17c478bd9Sstevel@tonic-gate /* 27c478bd9Sstevel@tonic-gate * CDDL HEADER START 37c478bd9Sstevel@tonic-gate * 47c478bd9Sstevel@tonic-gate * The contents of this file are subject to the terms of the 5a85a6733Sjosephb * Common Development and Distribution License (the "License"). 6a85a6733Sjosephb * You may not use this file except in compliance with the License. 77c478bd9Sstevel@tonic-gate * 87c478bd9Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 97c478bd9Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 107c478bd9Sstevel@tonic-gate * See the License for the specific language governing permissions 117c478bd9Sstevel@tonic-gate * and limitations under the License. 127c478bd9Sstevel@tonic-gate * 137c478bd9Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 147c478bd9Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 157c478bd9Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 167c478bd9Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 177c478bd9Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 187c478bd9Sstevel@tonic-gate * 197c478bd9Sstevel@tonic-gate * CDDL HEADER END 207c478bd9Sstevel@tonic-gate */ 217c478bd9Sstevel@tonic-gate /* 22ae115bc7Smrj * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate */ 257c478bd9Sstevel@tonic-gate 267c478bd9Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 277c478bd9Sstevel@tonic-gate 287c478bd9Sstevel@tonic-gate /* 297c478bd9Sstevel@tonic-gate * VM - Hardware Address Translation management for i386 and amd64 307c478bd9Sstevel@tonic-gate * 317c478bd9Sstevel@tonic-gate * Implementation of the interfaces described in <common/vm/hat.h> 327c478bd9Sstevel@tonic-gate * 337c478bd9Sstevel@tonic-gate * Nearly all the details of how the hardware is managed should not be 347c478bd9Sstevel@tonic-gate * visible outside this layer except for misc. machine specific functions 357c478bd9Sstevel@tonic-gate * that work in conjunction with this code. 367c478bd9Sstevel@tonic-gate * 377c478bd9Sstevel@tonic-gate * Routines used only inside of i86pc/vm start with hati_ for HAT Internal. 387c478bd9Sstevel@tonic-gate */ 397c478bd9Sstevel@tonic-gate 407c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 417c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 427c478bd9Sstevel@tonic-gate #include <sys/mman.h> 437c478bd9Sstevel@tonic-gate #include <sys/types.h> 447c478bd9Sstevel@tonic-gate #include <sys/systm.h> 457c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 467c478bd9Sstevel@tonic-gate #include <sys/thread.h> 477c478bd9Sstevel@tonic-gate #include <sys/proc.h> 487c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 497c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 507c478bd9Sstevel@tonic-gate #include <sys/disp.h> 517c478bd9Sstevel@tonic-gate #include <sys/shm.h> 527c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 537c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 547c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 557c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 567c478bd9Sstevel@tonic-gate #include <sys/promif.h> 577c478bd9Sstevel@tonic-gate #include <sys/var.h> 587c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 597c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 607c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 61ae115bc7Smrj #include <sys/controlregs.h> 62ae115bc7Smrj #include <sys/bootconf.h> 63ae115bc7Smrj #include <sys/bootsvcs.h> 64ae115bc7Smrj #include <sys/bootinfo.h> 6595c0a3c8Sjosephb #include <sys/archsystm.h> 667c478bd9Sstevel@tonic-gate 677c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 687c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 697c478bd9Sstevel@tonic-gate #include <vm/as.h> 707c478bd9Sstevel@tonic-gate #include <vm/seg.h> 717c478bd9Sstevel@tonic-gate #include <vm/page.h> 727c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 737c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 747c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 75*843e1988Sjohnlev #ifdef __xpv 76*843e1988Sjohnlev #include <sys/hypervisor.h> 77*843e1988Sjohnlev #endif 78ae115bc7Smrj #include <vm/kboot_mmu.h> 79250b7ff9Sjosephb #include <vm/seg_spt.h> 807c478bd9Sstevel@tonic-gate 817c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 827c478bd9Sstevel@tonic-gate 837c478bd9Sstevel@tonic-gate /* 847c478bd9Sstevel@tonic-gate * Basic parameters for hat operation. 857c478bd9Sstevel@tonic-gate */ 867c478bd9Sstevel@tonic-gate struct hat_mmu_info mmu; 877c478bd9Sstevel@tonic-gate 887c478bd9Sstevel@tonic-gate /* 897c478bd9Sstevel@tonic-gate * The page that is the kernel's top level pagetable. 907c478bd9Sstevel@tonic-gate * 91*843e1988Sjohnlev * For 32 bit PAE support on i86pc, the kernel hat will use the 1st 4 entries 927c478bd9Sstevel@tonic-gate * on this 4K page for its top level page table. The remaining groups of 937c478bd9Sstevel@tonic-gate * 4 entries are used for per processor copies of user VLP pagetables for 947c478bd9Sstevel@tonic-gate * running threads. See hat_switch() and reload_pae32() for details. 957c478bd9Sstevel@tonic-gate * 96*843e1988Sjohnlev * vlp_page[0..3] - level==2 PTEs for kernel HAT 97*843e1988Sjohnlev * vlp_page[4..7] - level==2 PTEs for user thread on cpu 0 98*843e1988Sjohnlev * vlp_page[8..11] - level==2 PTE for user thread on cpu 1 99*843e1988Sjohnlev * etc... 1007c478bd9Sstevel@tonic-gate */ 1017c478bd9Sstevel@tonic-gate static x86pte_t *vlp_page; 1027c478bd9Sstevel@tonic-gate 1037c478bd9Sstevel@tonic-gate /* 1047c478bd9Sstevel@tonic-gate * forward declaration of internal utility routines 1057c478bd9Sstevel@tonic-gate */ 1067c478bd9Sstevel@tonic-gate static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, 1077c478bd9Sstevel@tonic-gate x86pte_t new); 1087c478bd9Sstevel@tonic-gate 1097c478bd9Sstevel@tonic-gate /* 1107c478bd9Sstevel@tonic-gate * The kernel address space exists in all HATs. To implement this the 111*843e1988Sjohnlev * kernel reserves a fixed number of entries in the topmost level(s) of page 112*843e1988Sjohnlev * tables. The values are setup during startup and then copied to every user 113*843e1988Sjohnlev * hat created by hat_alloc(). This means that kernelbase must be: 1147c478bd9Sstevel@tonic-gate * 1157c478bd9Sstevel@tonic-gate * 4Meg aligned for 32 bit kernels 1167c478bd9Sstevel@tonic-gate * 512Gig aligned for x86_64 64 bit kernel 1177c478bd9Sstevel@tonic-gate * 118*843e1988Sjohnlev * The hat_kernel_range_ts describe what needs to be copied from kernel hat 119*843e1988Sjohnlev * to each user hat. 1207c478bd9Sstevel@tonic-gate */ 121*843e1988Sjohnlev typedef struct hat_kernel_range { 122*843e1988Sjohnlev level_t hkr_level; 123*843e1988Sjohnlev uintptr_t hkr_start_va; 124*843e1988Sjohnlev uintptr_t hkr_end_va; /* zero means to end of memory */ 125*843e1988Sjohnlev } hat_kernel_range_t; 126*843e1988Sjohnlev #define NUM_KERNEL_RANGE 2 127*843e1988Sjohnlev static hat_kernel_range_t kernel_ranges[NUM_KERNEL_RANGE]; 128*843e1988Sjohnlev static int num_kernel_ranges; 1297c478bd9Sstevel@tonic-gate 1307c478bd9Sstevel@tonic-gate uint_t use_boot_reserve = 1; /* cleared after early boot process */ 1317c478bd9Sstevel@tonic-gate uint_t can_steal_post_boot = 0; /* set late in boot to enable stealing */ 1327c478bd9Sstevel@tonic-gate 1337c478bd9Sstevel@tonic-gate /* 1347c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 1357c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 1367c478bd9Sstevel@tonic-gate */ 1377c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 1387c478bd9Sstevel@tonic-gate 1397c478bd9Sstevel@tonic-gate /* 1407c478bd9Sstevel@tonic-gate * management stuff for hat structures 1417c478bd9Sstevel@tonic-gate */ 1427c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 1437c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 1447c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 1457c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 1467c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 1477c478bd9Sstevel@tonic-gate 1487c478bd9Sstevel@tonic-gate /* 1497c478bd9Sstevel@tonic-gate * Simple statistics 1507c478bd9Sstevel@tonic-gate */ 1517c478bd9Sstevel@tonic-gate struct hatstats hatstat; 1527c478bd9Sstevel@tonic-gate 1537c478bd9Sstevel@tonic-gate /* 1547c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 1557c478bd9Sstevel@tonic-gate */ 1567c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 1577c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 1587c478bd9Sstevel@tonic-gate 1597c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 1607c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 1617c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 1627c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 1637c478bd9Sstevel@tonic-gate 1647c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 1657c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 1667c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 1677c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 1687c478bd9Sstevel@tonic-gate 1697c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 1707c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 1717c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 1727c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 1737c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 1747c478bd9Sstevel@tonic-gate 1757c478bd9Sstevel@tonic-gate /* 1767c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 1777c478bd9Sstevel@tonic-gate */ 1787c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 1797c478bd9Sstevel@tonic-gate static int 1807c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 1817c478bd9Sstevel@tonic-gate { 1827c478bd9Sstevel@tonic-gate hat_t *hat = buf; 1837c478bd9Sstevel@tonic-gate 1847c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 1857c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 1867c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 187250b7ff9Sjosephb hat->hat_ism_pgcnt = 0; 1887c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 1897c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 1907c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 1917c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 1927c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 1937c478bd9Sstevel@tonic-gate return (0); 1947c478bd9Sstevel@tonic-gate } 1957c478bd9Sstevel@tonic-gate 1967c478bd9Sstevel@tonic-gate /* 1977c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 1987c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 1997c478bd9Sstevel@tonic-gate */ 2007c478bd9Sstevel@tonic-gate hat_t * 2017c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 2027c478bd9Sstevel@tonic-gate { 203*843e1988Sjohnlev hat_t *hat; 204*843e1988Sjohnlev htable_t *ht; /* top level htable */ 205*843e1988Sjohnlev uint_t use_vlp; 206*843e1988Sjohnlev uint_t r; 207*843e1988Sjohnlev hat_kernel_range_t *rp; 208*843e1988Sjohnlev uintptr_t va; 209*843e1988Sjohnlev uintptr_t eva; 210*843e1988Sjohnlev uint_t start; 211*843e1988Sjohnlev uint_t cnt; 212*843e1988Sjohnlev htable_t *src; 2137c478bd9Sstevel@tonic-gate 2147c478bd9Sstevel@tonic-gate /* 2157c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 2167c478bd9Sstevel@tonic-gate * the htable_steal() code. 2177c478bd9Sstevel@tonic-gate */ 2187c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 2197c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 2207c478bd9Sstevel@tonic-gate 2217c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 2227c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 2237c478bd9Sstevel@tonic-gate hat->hat_as = as; 2247c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2257c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 2267c478bd9Sstevel@tonic-gate 227*843e1988Sjohnlev #if defined(__xpv) 2287c478bd9Sstevel@tonic-gate /* 229*843e1988Sjohnlev * No VLP stuff on the hypervisor due to the 64-bit split top level 230*843e1988Sjohnlev * page tables. On 32-bit it's not needed as the hypervisor takes 231*843e1988Sjohnlev * care of copying the top level PTEs to a below 4Gig page. 2327c478bd9Sstevel@tonic-gate */ 233*843e1988Sjohnlev use_vlp = 0; 234*843e1988Sjohnlev #else /* __xpv */ 235*843e1988Sjohnlev /* 32 bit processes uses a VLP style hat when running with PAE */ 2367c478bd9Sstevel@tonic-gate #if defined(__amd64) 2377c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 2387c478bd9Sstevel@tonic-gate #elif defined(__i386) 2397c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 2407c478bd9Sstevel@tonic-gate #endif 241*843e1988Sjohnlev #endif /* __xpv */ 2427c478bd9Sstevel@tonic-gate if (use_vlp) { 2437c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 2447c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 2457c478bd9Sstevel@tonic-gate } 2467c478bd9Sstevel@tonic-gate 2477c478bd9Sstevel@tonic-gate /* 2487c478bd9Sstevel@tonic-gate * Allocate the htable hash 2497c478bd9Sstevel@tonic-gate */ 2507c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 2517c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 2527c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 2537c478bd9Sstevel@tonic-gate } else { 2547c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 2557c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 2567c478bd9Sstevel@tonic-gate } 2577c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 2587c478bd9Sstevel@tonic-gate 2597c478bd9Sstevel@tonic-gate /* 2607c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 261*843e1988Sjohnlev * tables for the new hat. 2627c478bd9Sstevel@tonic-gate */ 2637c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2647c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 265*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2667c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 267*843e1988Sjohnlev hat->hat_htable = ht; 268ae115bc7Smrj 269*843e1988Sjohnlev #if defined(__amd64) 270*843e1988Sjohnlev if (hat->hat_flags & HAT_VLP) 271*843e1988Sjohnlev goto init_done; 2727c478bd9Sstevel@tonic-gate #endif 2737c478bd9Sstevel@tonic-gate 274*843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 275*843e1988Sjohnlev rp = &kernel_ranges[r]; 276*843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 277*843e1988Sjohnlev va += cnt * LEVEL_SIZE(rp->hkr_level)) { 278*843e1988Sjohnlev 279*843e1988Sjohnlev if (rp->hkr_level == TOP_LEVEL(hat)) 280*843e1988Sjohnlev ht = hat->hat_htable; 281*843e1988Sjohnlev else 282*843e1988Sjohnlev ht = htable_create(hat, va, rp->hkr_level, 283*843e1988Sjohnlev NULL); 284*843e1988Sjohnlev 285*843e1988Sjohnlev start = htable_va2entry(va, ht); 286*843e1988Sjohnlev cnt = HTABLE_NUM_PTES(ht) - start; 287*843e1988Sjohnlev eva = va + 288*843e1988Sjohnlev ((uintptr_t)cnt << LEVEL_SHIFT(rp->hkr_level)); 289*843e1988Sjohnlev if (rp->hkr_end_va != 0 && 290*843e1988Sjohnlev (eva > rp->hkr_end_va || eva == 0)) 291*843e1988Sjohnlev cnt = htable_va2entry(rp->hkr_end_va, ht) - 292*843e1988Sjohnlev start; 293*843e1988Sjohnlev 294*843e1988Sjohnlev #if defined(__i386) && !defined(__xpv) 295*843e1988Sjohnlev if (ht->ht_flags & HTABLE_VLP) { 296*843e1988Sjohnlev bcopy(&vlp_page[start], 297*843e1988Sjohnlev &hat->hat_vlp_ptes[start], 298*843e1988Sjohnlev cnt * sizeof (x86pte_t)); 299*843e1988Sjohnlev continue; 300*843e1988Sjohnlev } 301*843e1988Sjohnlev #endif 302*843e1988Sjohnlev src = htable_lookup(kas.a_hat, va, rp->hkr_level); 303*843e1988Sjohnlev ASSERT(src != NULL); 304*843e1988Sjohnlev x86pte_copy(src, ht, start, cnt); 305*843e1988Sjohnlev htable_release(src); 306*843e1988Sjohnlev } 307*843e1988Sjohnlev } 308*843e1988Sjohnlev 309*843e1988Sjohnlev init_done: 310*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 311*843e1988Sjohnlev 312*843e1988Sjohnlev #if defined(__xpv) 3137c478bd9Sstevel@tonic-gate /* 314*843e1988Sjohnlev * Pin top level page tables after initializing them 3157c478bd9Sstevel@tonic-gate */ 316*843e1988Sjohnlev xen_pin(hat->hat_htable->ht_pfn, mmu.max_level); 317*843e1988Sjohnlev #if defined(__amd64) 318*843e1988Sjohnlev xen_pin(hat->hat_user_ptable, mmu.max_level); 319*843e1988Sjohnlev #endif 3207c478bd9Sstevel@tonic-gate #endif 3217c478bd9Sstevel@tonic-gate 3227c478bd9Sstevel@tonic-gate /* 323a85a6733Sjosephb * Put it at the start of the global list of all hats (used by stealing) 324a85a6733Sjosephb * 325a85a6733Sjosephb * kas.a_hat is not in the list but is instead used to find the 326a85a6733Sjosephb * first and last items in the list. 327a85a6733Sjosephb * 328a85a6733Sjosephb * - kas.a_hat->hat_next points to the start of the user hats. 329a85a6733Sjosephb * The list ends where hat->hat_next == NULL 330a85a6733Sjosephb * 331a85a6733Sjosephb * - kas.a_hat->hat_prev points to the last of the user hats. 332a85a6733Sjosephb * The list begins where hat->hat_prev == NULL 3337c478bd9Sstevel@tonic-gate */ 3347c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 335a85a6733Sjosephb hat->hat_prev = NULL; 336a85a6733Sjosephb hat->hat_next = kas.a_hat->hat_next; 337a85a6733Sjosephb if (hat->hat_next) 338a85a6733Sjosephb hat->hat_next->hat_prev = hat; 339a85a6733Sjosephb else 340a85a6733Sjosephb kas.a_hat->hat_prev = hat; 3417c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 3427c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3437c478bd9Sstevel@tonic-gate 3447c478bd9Sstevel@tonic-gate return (hat); 3457c478bd9Sstevel@tonic-gate } 3467c478bd9Sstevel@tonic-gate 3477c478bd9Sstevel@tonic-gate /* 3487c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 3497c478bd9Sstevel@tonic-gate */ 3507c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3517c478bd9Sstevel@tonic-gate void 3527c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 3537c478bd9Sstevel@tonic-gate { 3547c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock)); 355a85a6733Sjosephb 356a85a6733Sjosephb /* 357a85a6733Sjosephb * If the hat is currently a stealing victim, wait for the stealing 358a85a6733Sjosephb * to finish. Once we mark it as HAT_FREEING, htable_steal() 359a85a6733Sjosephb * won't look at its pagetables anymore. 360a85a6733Sjosephb */ 3617c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 362a85a6733Sjosephb while (hat->hat_flags & HAT_VICTIM) 363a85a6733Sjosephb cv_wait(&hat_list_cv, &hat_list_lock); 3647c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 3657c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3667c478bd9Sstevel@tonic-gate } 3677c478bd9Sstevel@tonic-gate 3687c478bd9Sstevel@tonic-gate /* 3697c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 3707c478bd9Sstevel@tonic-gate */ 3717c478bd9Sstevel@tonic-gate void 3727c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 3737c478bd9Sstevel@tonic-gate { 3747c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 3757c478bd9Sstevel@tonic-gate 3767c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 3777c478bd9Sstevel@tonic-gate 3787c478bd9Sstevel@tonic-gate /* 3797c478bd9Sstevel@tonic-gate * must not be running on the given hat 3807c478bd9Sstevel@tonic-gate */ 3817c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 3827c478bd9Sstevel@tonic-gate 3837c478bd9Sstevel@tonic-gate /* 384a85a6733Sjosephb * Remove it from the list of HATs 3857c478bd9Sstevel@tonic-gate */ 3867c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 387a85a6733Sjosephb if (hat->hat_prev) 388a85a6733Sjosephb hat->hat_prev->hat_next = hat->hat_next; 389a85a6733Sjosephb else 3907c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 391a85a6733Sjosephb if (hat->hat_next) 392a85a6733Sjosephb hat->hat_next->hat_prev = hat->hat_prev; 393a85a6733Sjosephb else 394a85a6733Sjosephb kas.a_hat->hat_prev = hat->hat_prev; 3957c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 396a85a6733Sjosephb hat->hat_next = hat->hat_prev = NULL; 3977c478bd9Sstevel@tonic-gate 398*843e1988Sjohnlev #if defined(__xpv) 399*843e1988Sjohnlev /* 400*843e1988Sjohnlev * On the hypervisor, unpin top level page table(s) 401*843e1988Sjohnlev */ 402*843e1988Sjohnlev xen_unpin(hat->hat_htable->ht_pfn); 403*843e1988Sjohnlev #if defined(__amd64) 404*843e1988Sjohnlev xen_unpin(hat->hat_user_ptable); 405*843e1988Sjohnlev #endif 406*843e1988Sjohnlev #endif 407*843e1988Sjohnlev 4087c478bd9Sstevel@tonic-gate /* 4097c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 4107c478bd9Sstevel@tonic-gate */ 4117c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 4127c478bd9Sstevel@tonic-gate 4137c478bd9Sstevel@tonic-gate /* 4147c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 4157c478bd9Sstevel@tonic-gate */ 4167c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 4177c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 4187c478bd9Sstevel@tonic-gate else 4197c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 4207c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 4217c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 4227c478bd9Sstevel@tonic-gate 4237c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 4247c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 4257c478bd9Sstevel@tonic-gate } 4267c478bd9Sstevel@tonic-gate 4277c478bd9Sstevel@tonic-gate /* 4287c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 4297c478bd9Sstevel@tonic-gate * 4307c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 4317c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 4327c478bd9Sstevel@tonic-gate */ 4337c478bd9Sstevel@tonic-gate uintptr_t 4347c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 4357c478bd9Sstevel@tonic-gate { 4367c478bd9Sstevel@tonic-gate #if defined(__i386) 4377c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 4387c478bd9Sstevel@tonic-gate #endif 4397c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 4407c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 4417c478bd9Sstevel@tonic-gate return (va); 4427c478bd9Sstevel@tonic-gate } 4437c478bd9Sstevel@tonic-gate 4447c478bd9Sstevel@tonic-gate /* 4457c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 4467c478bd9Sstevel@tonic-gate */ 4477c478bd9Sstevel@tonic-gate void 4487c478bd9Sstevel@tonic-gate mmu_init(void) 4497c478bd9Sstevel@tonic-gate { 4507c478bd9Sstevel@tonic-gate uint_t max_htables; 4517c478bd9Sstevel@tonic-gate uint_t pa_bits; 4527c478bd9Sstevel@tonic-gate uint_t va_bits; 4537c478bd9Sstevel@tonic-gate int i; 4547c478bd9Sstevel@tonic-gate 4557c478bd9Sstevel@tonic-gate /* 456ae115bc7Smrj * If CPU enabled the page table global bit, use it for the kernel 457ae115bc7Smrj * This is bit 7 in CR4 (PGE - Page Global Enable). 4587c478bd9Sstevel@tonic-gate */ 459ae115bc7Smrj if ((x86_feature & X86_PGE) != 0 && (getcr4() & CR4_PGE) != 0) 4607c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 4617c478bd9Sstevel@tonic-gate 4627c478bd9Sstevel@tonic-gate /* 463ae115bc7Smrj * Detect NX and PAE usage. 4647c478bd9Sstevel@tonic-gate */ 465ae115bc7Smrj mmu.pae_hat = kbm_pae_support; 466ae115bc7Smrj if (kbm_nx_support) 4677c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 468ae115bc7Smrj else 4697c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 4707c478bd9Sstevel@tonic-gate 4717c478bd9Sstevel@tonic-gate /* 4727c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 4737c478bd9Sstevel@tonic-gate */ 4747c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 4757c478bd9Sstevel@tonic-gate 4767c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 4777c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 4787c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 4797c478bd9Sstevel@tonic-gate } else { 4807c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 4817c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 4827c478bd9Sstevel@tonic-gate } 4837c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 4847c478bd9Sstevel@tonic-gate /* 4857c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 4867c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 4877c478bd9Sstevel@tonic-gate */ 4887c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 4897c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 4907c478bd9Sstevel@tonic-gate #else 4917c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 4927c478bd9Sstevel@tonic-gate #endif 4937c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 4947c478bd9Sstevel@tonic-gate 4957c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 4967c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 4977c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 4987c478bd9Sstevel@tonic-gate 4997c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5007c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 5017c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 5027c478bd9Sstevel@tonic-gate } else { 5037c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 5047c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 5057c478bd9Sstevel@tonic-gate } 5067c478bd9Sstevel@tonic-gate 5077c478bd9Sstevel@tonic-gate if (mmu.pae_hat && (x86_feature & X86_PAE) == 0) 5087c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 5097c478bd9Sstevel@tonic-gate 5107c478bd9Sstevel@tonic-gate if ((x86_feature & X86_CX8) == 0) 5117c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 5127c478bd9Sstevel@tonic-gate 5137c478bd9Sstevel@tonic-gate /* 5147c478bd9Sstevel@tonic-gate * Initialize parameters based on the 64 or 32 bit kernels and 5157c478bd9Sstevel@tonic-gate * for the 32 bit kernel decide if we should use PAE. 5167c478bd9Sstevel@tonic-gate */ 517ae115bc7Smrj if (kbm_largepage_support) 5187c478bd9Sstevel@tonic-gate mmu.max_page_level = 1; 5197c478bd9Sstevel@tonic-gate else 5207c478bd9Sstevel@tonic-gate mmu.max_page_level = 0; 5217c478bd9Sstevel@tonic-gate mmu_page_sizes = mmu.max_page_level + 1; 5227c478bd9Sstevel@tonic-gate mmu_exported_page_sizes = mmu_page_sizes; 5237c478bd9Sstevel@tonic-gate 5247c478bd9Sstevel@tonic-gate #if defined(__amd64) 5257c478bd9Sstevel@tonic-gate 5267c478bd9Sstevel@tonic-gate mmu.num_level = 4; 5277c478bd9Sstevel@tonic-gate mmu.max_level = 3; 5287c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5297c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 5307c478bd9Sstevel@tonic-gate 5317c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5327c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5337c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5347c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 5357c478bd9Sstevel@tonic-gate 5367c478bd9Sstevel@tonic-gate #elif defined(__i386) 5377c478bd9Sstevel@tonic-gate 5387c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5397c478bd9Sstevel@tonic-gate mmu.num_level = 3; 5407c478bd9Sstevel@tonic-gate mmu.max_level = 2; 5417c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5427c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 5437c478bd9Sstevel@tonic-gate 5447c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5457c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5467c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5477c478bd9Sstevel@tonic-gate 5487c478bd9Sstevel@tonic-gate } else { 5497c478bd9Sstevel@tonic-gate mmu.num_level = 2; 5507c478bd9Sstevel@tonic-gate mmu.max_level = 1; 5517c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 5527c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 5537c478bd9Sstevel@tonic-gate 5547c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5557c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 5567c478bd9Sstevel@tonic-gate } 5577c478bd9Sstevel@tonic-gate 5587c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5597c478bd9Sstevel@tonic-gate 5607c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 5617c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 5627c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 5637c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 5647c478bd9Sstevel@tonic-gate } 5657c478bd9Sstevel@tonic-gate 566ae115bc7Smrj for (i = 0; i <= mmu.max_page_level; ++i) { 567ae115bc7Smrj mmu.pte_bits[i] = PT_VALID; 568*843e1988Sjohnlev #if defined(__xpv) && defined(__amd64) 569*843e1988Sjohnlev mmu.pte_bits[i] |= PT_USER; 570*843e1988Sjohnlev #endif 571ae115bc7Smrj if (i > 0) 572ae115bc7Smrj mmu.pte_bits[i] |= PT_PAGESIZE; 573ae115bc7Smrj } 5747c478bd9Sstevel@tonic-gate 5757c478bd9Sstevel@tonic-gate /* 5767c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 5777c478bd9Sstevel@tonic-gate */ 5787c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 5797c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 580ae115bc7Smrj 5817c478bd9Sstevel@tonic-gate #if defined(__i386) 5827c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 5837c478bd9Sstevel@tonic-gate #endif 5847c478bd9Sstevel@tonic-gate 5857c478bd9Sstevel@tonic-gate /* 5867c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 5877c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 5887c478bd9Sstevel@tonic-gate * 5897c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 5907c478bd9Sstevel@tonic-gate */ 5917c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 5927c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 5937c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 5947c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 5957c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 5967c478bd9Sstevel@tonic-gate 5977c478bd9Sstevel@tonic-gate #if defined(__amd64) 5987c478bd9Sstevel@tonic-gate /* 5997c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 6007c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 6017c478bd9Sstevel@tonic-gate * a 64 bit process. 6027c478bd9Sstevel@tonic-gate */ 6037c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 6047c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 6057c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 6067c478bd9Sstevel@tonic-gate #endif 6077c478bd9Sstevel@tonic-gate } 6087c478bd9Sstevel@tonic-gate 6097c478bd9Sstevel@tonic-gate 6107c478bd9Sstevel@tonic-gate /* 6117c478bd9Sstevel@tonic-gate * initialize hat data structures 6127c478bd9Sstevel@tonic-gate */ 6137c478bd9Sstevel@tonic-gate void 6147c478bd9Sstevel@tonic-gate hat_init() 6157c478bd9Sstevel@tonic-gate { 6167c478bd9Sstevel@tonic-gate #if defined(__i386) 6177c478bd9Sstevel@tonic-gate /* 6187c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 6197c478bd9Sstevel@tonic-gate */ 6207c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 6217c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 6227c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 6237c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 6247c478bd9Sstevel@tonic-gate } 6257c478bd9Sstevel@tonic-gate #endif 6267c478bd9Sstevel@tonic-gate 6277c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 6287c478bd9Sstevel@tonic-gate 6297c478bd9Sstevel@tonic-gate /* 6307c478bd9Sstevel@tonic-gate * initialize kmem caches 6317c478bd9Sstevel@tonic-gate */ 6327c478bd9Sstevel@tonic-gate htable_init(); 6337c478bd9Sstevel@tonic-gate hment_init(); 6347c478bd9Sstevel@tonic-gate 6357c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 6367c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 6377c478bd9Sstevel@tonic-gate NULL, 0, 0); 6387c478bd9Sstevel@tonic-gate 6397c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 6407c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6417c478bd9Sstevel@tonic-gate NULL, 0, 0); 6427c478bd9Sstevel@tonic-gate 6437c478bd9Sstevel@tonic-gate /* 6447c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 6457c478bd9Sstevel@tonic-gate */ 6467c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 6477c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 6487c478bd9Sstevel@tonic-gate } else { 6497c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 6507c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6517c478bd9Sstevel@tonic-gate NULL, 0, 0); 6527c478bd9Sstevel@tonic-gate } 6537c478bd9Sstevel@tonic-gate 6547c478bd9Sstevel@tonic-gate /* 6557c478bd9Sstevel@tonic-gate * Set up the kernel's hat 6567c478bd9Sstevel@tonic-gate */ 6577c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER); 6587c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 6597c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 6607c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 6617c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 6627c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(&kas, &kas.a_lock); 6637c478bd9Sstevel@tonic-gate 6647c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 6657c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 6667c478bd9Sstevel@tonic-gate 6677c478bd9Sstevel@tonic-gate /* 6687c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 669a85a6733Sjosephb * The kernel hat's prev pointer tracks the last hat on the list for 670a85a6733Sjosephb * htable_steal() to use. 6717c478bd9Sstevel@tonic-gate */ 6727c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 673a85a6733Sjosephb kas.a_hat->hat_prev = NULL; 6747c478bd9Sstevel@tonic-gate 6757c478bd9Sstevel@tonic-gate /* 6767c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 6777c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 6787c478bd9Sstevel@tonic-gate */ 6797c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 6807c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 6817c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 6827c478bd9Sstevel@tonic-gate 6837c478bd9Sstevel@tonic-gate /* 6847c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 6857c478bd9Sstevel@tonic-gate */ 6867c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 6877c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 6889d9461f9Strevtom 6899d9461f9Strevtom /* 6909d9461f9Strevtom * Pre-allocate hrm_hashtab before enabling the collection of 6919d9461f9Strevtom * refmod statistics. Allocating on the fly would mean us 6929d9461f9Strevtom * running the risk of suffering recursive mutex enters or 6939d9461f9Strevtom * deadlocks. 6949d9461f9Strevtom */ 6959d9461f9Strevtom hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *), 6969d9461f9Strevtom KM_SLEEP); 6977c478bd9Sstevel@tonic-gate } 6987c478bd9Sstevel@tonic-gate 6997c478bd9Sstevel@tonic-gate /* 7007c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 7017c478bd9Sstevel@tonic-gate * 7027c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 7037c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 7047c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 7057c478bd9Sstevel@tonic-gate */ 7067c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 7077c478bd9Sstevel@tonic-gate static void 7087c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 7097c478bd9Sstevel@tonic-gate { 710*843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 7117c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 7127c478bd9Sstevel@tonic-gate pfn_t pfn; 7137c478bd9Sstevel@tonic-gate 7147c478bd9Sstevel@tonic-gate /* 7157c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 7167c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 7177c478bd9Sstevel@tonic-gate */ 7187c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 7197c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7207c478bd9Sstevel@tonic-gate 7217c478bd9Sstevel@tonic-gate /* 7227c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 7237c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 7247c478bd9Sstevel@tonic-gate */ 7257c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7267c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 7277c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 7287c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 729*843e1988Sjohnlev bcopy(vlp_page, hci->hci_vlp_l3ptes, MMU_PAGESIZE); 7307c478bd9Sstevel@tonic-gate 7317c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 7327c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 7337c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 734*843e1988Sjohnlev #endif /* __amd64 && !__xpv */ 7357c478bd9Sstevel@tonic-gate } 7367c478bd9Sstevel@tonic-gate 737ae115bc7Smrj /*ARGSUSED*/ 738ae115bc7Smrj static void 739ae115bc7Smrj hat_vlp_teardown(cpu_t *cpu) 740ae115bc7Smrj { 741*843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 742ae115bc7Smrj struct hat_cpu_info *hci; 743ae115bc7Smrj 744ae115bc7Smrj if ((hci = cpu->cpu_hat_info) == NULL) 745ae115bc7Smrj return; 746ae115bc7Smrj if (hci->hci_vlp_l2ptes) 747ae115bc7Smrj kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE); 748ae115bc7Smrj if (hci->hci_vlp_l3ptes) 749ae115bc7Smrj kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE); 750*843e1988Sjohnlev #endif 751*843e1988Sjohnlev } 752*843e1988Sjohnlev 753*843e1988Sjohnlev #define NEXT_HKR(r, l, s, e) { \ 754*843e1988Sjohnlev kernel_ranges[r].hkr_level = l; \ 755*843e1988Sjohnlev kernel_ranges[r].hkr_start_va = s; \ 756*843e1988Sjohnlev kernel_ranges[r].hkr_end_va = e; \ 757*843e1988Sjohnlev ++r; \ 758ae115bc7Smrj } 759ae115bc7Smrj 7607c478bd9Sstevel@tonic-gate /* 7617c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 7627c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 7637c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 7647c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 765ae115bc7Smrj * 766ae115bc7Smrj * create the kmap mappings. 7677c478bd9Sstevel@tonic-gate */ 7687c478bd9Sstevel@tonic-gate void 7697c478bd9Sstevel@tonic-gate hat_init_finish(void) 7707c478bd9Sstevel@tonic-gate { 771ae115bc7Smrj size_t size; 772*843e1988Sjohnlev uint_t r = 0; 773*843e1988Sjohnlev uintptr_t va; 774*843e1988Sjohnlev hat_kernel_range_t *rp; 7757c478bd9Sstevel@tonic-gate 7767c478bd9Sstevel@tonic-gate 7777c478bd9Sstevel@tonic-gate /* 778*843e1988Sjohnlev * We are now effectively running on the kernel hat. 779*843e1988Sjohnlev * Clearing use_boot_reserve shuts off using the pre-allocated boot 780*843e1988Sjohnlev * reserve for all HAT allocations. From here on, the reserves are 781*843e1988Sjohnlev * only used when avoiding recursion in kmem_alloc(). 7827c478bd9Sstevel@tonic-gate */ 783*843e1988Sjohnlev use_boot_reserve = 0; 784*843e1988Sjohnlev htable_adjust_reserve(); 7857c478bd9Sstevel@tonic-gate 7867c478bd9Sstevel@tonic-gate /* 787*843e1988Sjohnlev * User HATs are initialized with copies of all kernel mappings in 788*843e1988Sjohnlev * higher level page tables. Ensure that those entries exist. 7897c478bd9Sstevel@tonic-gate */ 790*843e1988Sjohnlev #if defined(__amd64) 791*843e1988Sjohnlev 792*843e1988Sjohnlev NEXT_HKR(r, 3, kernelbase, 0); 793*843e1988Sjohnlev #if defined(__xpv) 794*843e1988Sjohnlev NEXT_HKR(r, 3, HYPERVISOR_VIRT_START, HYPERVISOR_VIRT_END); 795*843e1988Sjohnlev #endif 796*843e1988Sjohnlev 797*843e1988Sjohnlev #elif defined(__i386) 798*843e1988Sjohnlev 799*843e1988Sjohnlev #if !defined(__xpv) 800*843e1988Sjohnlev if (mmu.pae_hat) { 801*843e1988Sjohnlev va = kernelbase; 802*843e1988Sjohnlev if ((va & LEVEL_MASK(2)) != va) { 803*843e1988Sjohnlev va = P2ROUNDUP(va, LEVEL_SIZE(2)); 804*843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, va); 805*843e1988Sjohnlev } 806*843e1988Sjohnlev if (va != 0) 807*843e1988Sjohnlev NEXT_HKR(r, 2, va, 0); 808*843e1988Sjohnlev } else 809*843e1988Sjohnlev #endif /* __xpv */ 810*843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, 0); 811*843e1988Sjohnlev 812*843e1988Sjohnlev #endif /* __i386 */ 813*843e1988Sjohnlev 814*843e1988Sjohnlev num_kernel_ranges = r; 8157c478bd9Sstevel@tonic-gate 8167c478bd9Sstevel@tonic-gate /* 817*843e1988Sjohnlev * Create all the kernel pagetables that will have entries 818*843e1988Sjohnlev * shared to user HATs. 8197c478bd9Sstevel@tonic-gate */ 820*843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 821*843e1988Sjohnlev rp = &kernel_ranges[r]; 822*843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 823*843e1988Sjohnlev va += LEVEL_SIZE(rp->hkr_level)) { 824*843e1988Sjohnlev htable_t *ht; 825*843e1988Sjohnlev 826*843e1988Sjohnlev if (IN_HYPERVISOR_VA(va)) 827*843e1988Sjohnlev continue; 828*843e1988Sjohnlev 829*843e1988Sjohnlev /* can/must skip if a page mapping already exists */ 830*843e1988Sjohnlev if (rp->hkr_level <= mmu.max_page_level && 831*843e1988Sjohnlev (ht = htable_getpage(kas.a_hat, va, NULL)) != 832*843e1988Sjohnlev NULL) { 833*843e1988Sjohnlev htable_release(ht); 834*843e1988Sjohnlev continue; 835*843e1988Sjohnlev } 836*843e1988Sjohnlev 837*843e1988Sjohnlev (void) htable_create(kas.a_hat, va, rp->hkr_level - 1, 838*843e1988Sjohnlev NULL); 839*843e1988Sjohnlev } 840*843e1988Sjohnlev } 8417c478bd9Sstevel@tonic-gate 8427c478bd9Sstevel@tonic-gate /* 843*843e1988Sjohnlev * 32 bit PAE metal kernels use only 4 of the 512 entries in the 844*843e1988Sjohnlev * page holding the top level pagetable. We use the remainder for 845*843e1988Sjohnlev * the "per CPU" page tables for VLP processes. 846*843e1988Sjohnlev * Map the top level kernel pagetable into the kernel to make 847*843e1988Sjohnlev * it easy to use bcopy access these tables. 8487c478bd9Sstevel@tonic-gate */ 8497c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 8507c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 8517c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 8527c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 853*843e1988Sjohnlev #if !defined(__xpv) 854ae115bc7Smrj PROT_WRITE | 855*843e1988Sjohnlev #endif 856ae115bc7Smrj PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK, 8577c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 8587c478bd9Sstevel@tonic-gate } 8597c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 860ae115bc7Smrj 861ae115bc7Smrj /* 862ae115bc7Smrj * Create kmap (cached mappings of kernel PTEs) 863ae115bc7Smrj * for 32 bit we map from segmap_start .. ekernelheap 864ae115bc7Smrj * for 64 bit we map from segmap_start .. segmap_start + segmapsize; 865ae115bc7Smrj */ 866ae115bc7Smrj #if defined(__i386) 867ae115bc7Smrj size = (uintptr_t)ekernelheap - segmap_start; 868ae115bc7Smrj #elif defined(__amd64) 869ae115bc7Smrj size = segmapsize; 870ae115bc7Smrj #endif 871ae115bc7Smrj hat_kmap_init((uintptr_t)segmap_start, size); 8727c478bd9Sstevel@tonic-gate } 8737c478bd9Sstevel@tonic-gate 8747c478bd9Sstevel@tonic-gate /* 8757c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 8767c478bd9Sstevel@tonic-gate * are 32 bit, so for safety we must use cas64() to install these. 8777c478bd9Sstevel@tonic-gate */ 8787c478bd9Sstevel@tonic-gate #ifdef __i386 8797c478bd9Sstevel@tonic-gate static void 8807c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 8817c478bd9Sstevel@tonic-gate { 8827c478bd9Sstevel@tonic-gate x86pte_t *src; 8837c478bd9Sstevel@tonic-gate x86pte_t *dest; 8847c478bd9Sstevel@tonic-gate x86pte_t pte; 8857c478bd9Sstevel@tonic-gate int i; 8867c478bd9Sstevel@tonic-gate 8877c478bd9Sstevel@tonic-gate /* 8887c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 8897c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 8907c478bd9Sstevel@tonic-gate */ 8917c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 8927c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 8937c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 8947c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 8957c478bd9Sstevel@tonic-gate for (;;) { 8967c478bd9Sstevel@tonic-gate pte = dest[i]; 8977c478bd9Sstevel@tonic-gate if (pte == src[i]) 8987c478bd9Sstevel@tonic-gate break; 8997c478bd9Sstevel@tonic-gate if (cas64(dest + i, pte, src[i]) != src[i]) 9007c478bd9Sstevel@tonic-gate break; 9017c478bd9Sstevel@tonic-gate } 9027c478bd9Sstevel@tonic-gate } 9037c478bd9Sstevel@tonic-gate } 9047c478bd9Sstevel@tonic-gate #endif 9057c478bd9Sstevel@tonic-gate 9067c478bd9Sstevel@tonic-gate /* 9077c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 908*843e1988Sjohnlev * 909*843e1988Sjohnlev * On the 32-bit PAE hypervisor, %cr3 is a 64-bit value, on metal it 910*843e1988Sjohnlev * remains a 32-bit value. 9117c478bd9Sstevel@tonic-gate */ 9127c478bd9Sstevel@tonic-gate void 9137c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 9147c478bd9Sstevel@tonic-gate { 915*843e1988Sjohnlev uint64_t newcr3; 9167c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 9177c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 9187c478bd9Sstevel@tonic-gate 9197c478bd9Sstevel@tonic-gate /* 9207c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 9217c478bd9Sstevel@tonic-gate */ 9227c478bd9Sstevel@tonic-gate if (old != NULL) { 9237c478bd9Sstevel@tonic-gate if (old == hat) 9247c478bd9Sstevel@tonic-gate return; 9257c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 9267c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 9277c478bd9Sstevel@tonic-gate } 9287c478bd9Sstevel@tonic-gate 9297c478bd9Sstevel@tonic-gate /* 93095c0a3c8Sjosephb * Add this CPU to the active set for this HAT. 9317c478bd9Sstevel@tonic-gate */ 9327c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 9337c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 9347c478bd9Sstevel@tonic-gate } 9357c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 9367c478bd9Sstevel@tonic-gate 9377c478bd9Sstevel@tonic-gate /* 9387c478bd9Sstevel@tonic-gate * now go ahead and load cr3 9397c478bd9Sstevel@tonic-gate */ 9407c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 9417c478bd9Sstevel@tonic-gate #if defined(__amd64) 9427c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 9437c478bd9Sstevel@tonic-gate 9447c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 9457c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 9467c478bd9Sstevel@tonic-gate #elif defined(__i386) 9477c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 9487c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 9497c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 9507c478bd9Sstevel@tonic-gate #endif 9517c478bd9Sstevel@tonic-gate } else { 952*843e1988Sjohnlev newcr3 = MAKECR3((uint64_t)hat->hat_htable->ht_pfn); 9537c478bd9Sstevel@tonic-gate } 954*843e1988Sjohnlev #ifdef __xpv 955*843e1988Sjohnlev { 956*843e1988Sjohnlev struct mmuext_op t[2]; 957*843e1988Sjohnlev uint_t retcnt; 958*843e1988Sjohnlev uint_t opcnt = 1; 959*843e1988Sjohnlev 960*843e1988Sjohnlev t[0].cmd = MMUEXT_NEW_BASEPTR; 961*843e1988Sjohnlev t[0].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 962*843e1988Sjohnlev #if defined(__amd64) 963*843e1988Sjohnlev /* 964*843e1988Sjohnlev * There's an interesting problem here, as to what to 965*843e1988Sjohnlev * actually specify when switching to the kernel hat. 966*843e1988Sjohnlev * For now we'll reuse the kernel hat again. 967*843e1988Sjohnlev */ 968*843e1988Sjohnlev t[1].cmd = MMUEXT_NEW_USER_BASEPTR; 969*843e1988Sjohnlev if (hat == kas.a_hat) 970*843e1988Sjohnlev t[1].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 971*843e1988Sjohnlev else 972*843e1988Sjohnlev t[1].arg1.mfn = pfn_to_mfn(hat->hat_user_ptable); 973*843e1988Sjohnlev ++opcnt; 974*843e1988Sjohnlev #endif /* __amd64 */ 975*843e1988Sjohnlev if (HYPERVISOR_mmuext_op(t, opcnt, &retcnt, DOMID_SELF) < 0) 976*843e1988Sjohnlev panic("HYPERVISOR_mmu_update() failed"); 977*843e1988Sjohnlev ASSERT(retcnt == opcnt); 978*843e1988Sjohnlev 979*843e1988Sjohnlev } 980*843e1988Sjohnlev #else 9817c478bd9Sstevel@tonic-gate setcr3(newcr3); 982*843e1988Sjohnlev #endif 9837c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 9847c478bd9Sstevel@tonic-gate } 9857c478bd9Sstevel@tonic-gate 9867c478bd9Sstevel@tonic-gate /* 9877c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 9887c478bd9Sstevel@tonic-gate */ 9897c478bd9Sstevel@tonic-gate static x86pte_t 9907c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 9917c478bd9Sstevel@tonic-gate { 9927c478bd9Sstevel@tonic-gate x86pte_t pte; 9937c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 9947c478bd9Sstevel@tonic-gate 9957c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 9967c478bd9Sstevel@tonic-gate 9977c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 9987c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 9997c478bd9Sstevel@tonic-gate 10007c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 10017c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 10027c478bd9Sstevel@tonic-gate 10037c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 10047c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 10057c478bd9Sstevel@tonic-gate 10067c478bd9Sstevel@tonic-gate /* 1007ae115bc7Smrj * Set the software bits used track ref/mod sync's and hments. 1008ae115bc7Smrj * If not using REF/MOD, set them to avoid h/w rewriting PTEs. 10097c478bd9Sstevel@tonic-gate */ 10107c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 1011ae115bc7Smrj PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD); 1012ae115bc7Smrj else if (attr & HAT_NOSYNC) 1013ae115bc7Smrj PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD); 10147c478bd9Sstevel@tonic-gate 10157c478bd9Sstevel@tonic-gate /* 10167c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 10177c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 10187c478bd9Sstevel@tonic-gate * to them in the given order. 10197c478bd9Sstevel@tonic-gate * 10207c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 10217c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 10227c478bd9Sstevel@tonic-gate */ 10237c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 10247c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10257c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 10267c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 10277c478bd9Sstevel@tonic-gate /* nothing to set */; 10287c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 10297c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10307c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) 10317c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 10327c478bd9Sstevel@tonic-gate else 10337c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 10347c478bd9Sstevel@tonic-gate } else { 10357c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 10367c478bd9Sstevel@tonic-gate } 10377c478bd9Sstevel@tonic-gate 10387c478bd9Sstevel@tonic-gate return (pte); 10397c478bd9Sstevel@tonic-gate } 10407c478bd9Sstevel@tonic-gate 10417c478bd9Sstevel@tonic-gate /* 10427c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 10437c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 10447c478bd9Sstevel@tonic-gate */ 10457c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 10467c478bd9Sstevel@tonic-gate int 10477c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 10487c478bd9Sstevel@tonic-gate { 10497c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 10507c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 10517c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 10527c478bd9Sstevel@tonic-gate return (0); 10537c478bd9Sstevel@tonic-gate } 10547c478bd9Sstevel@tonic-gate 10557c478bd9Sstevel@tonic-gate /* 10567c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 10577c478bd9Sstevel@tonic-gate */ 10587c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 10597c478bd9Sstevel@tonic-gate void 10607c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 10617c478bd9Sstevel@tonic-gate { 10627c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 10637c478bd9Sstevel@tonic-gate } 10647c478bd9Sstevel@tonic-gate 10657c478bd9Sstevel@tonic-gate /* 10667c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 10677c478bd9Sstevel@tonic-gate * that is being swapped out. 10687c478bd9Sstevel@tonic-gate */ 10697c478bd9Sstevel@tonic-gate void 10707c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 10717c478bd9Sstevel@tonic-gate { 10727c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 10737c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 10747c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 10757c478bd9Sstevel@tonic-gate level_t l; 10767c478bd9Sstevel@tonic-gate 1077*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 10787c478bd9Sstevel@tonic-gate /* 10797c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 10807c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 10817c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 10827c478bd9Sstevel@tonic-gate * 10837c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 10847c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 10857c478bd9Sstevel@tonic-gate */ 10867c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 10877c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 10887c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 10897c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 10907c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 10917c478bd9Sstevel@tonic-gate 10927c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 10937c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 10947c478bd9Sstevel@tonic-gate if (ht == NULL) 10957c478bd9Sstevel@tonic-gate break; 10967c478bd9Sstevel@tonic-gate 10977c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 10987c478bd9Sstevel@tonic-gate 10997c478bd9Sstevel@tonic-gate /* 11007c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 11017c478bd9Sstevel@tonic-gate * This code knows that only level 0 page tables are shared 11027c478bd9Sstevel@tonic-gate */ 11037c478bd9Sstevel@tonic-gate l = ht->ht_level; 11047c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 11057c478bd9Sstevel@tonic-gate ASSERT(l == 0); 11067c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 11077c478bd9Sstevel@tonic-gate htable_release(ht); 11087c478bd9Sstevel@tonic-gate ht = NULL; 11097c478bd9Sstevel@tonic-gate continue; 11107c478bd9Sstevel@tonic-gate } 11117c478bd9Sstevel@tonic-gate 11127c478bd9Sstevel@tonic-gate /* 11137c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 11147c478bd9Sstevel@tonic-gate */ 11157c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 11167c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 11177c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 11187c478bd9Sstevel@tonic-gate 11197c478bd9Sstevel@tonic-gate /* 11207c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 11217c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 11227c478bd9Sstevel@tonic-gate */ 11237c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 11247c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 11257c478bd9Sstevel@tonic-gate else 11267c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 11277c478bd9Sstevel@tonic-gate } 11287c478bd9Sstevel@tonic-gate if (ht) 11297c478bd9Sstevel@tonic-gate htable_release(ht); 11307c478bd9Sstevel@tonic-gate 11317c478bd9Sstevel@tonic-gate /* 11327c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 11337c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 11347c478bd9Sstevel@tonic-gate */ 11357c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 1136*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 11377c478bd9Sstevel@tonic-gate } 11387c478bd9Sstevel@tonic-gate 11397c478bd9Sstevel@tonic-gate /* 11407c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 11417c478bd9Sstevel@tonic-gate */ 11427c478bd9Sstevel@tonic-gate size_t 11437c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 11447c478bd9Sstevel@tonic-gate { 11457c478bd9Sstevel@tonic-gate size_t total = 0; 11467c478bd9Sstevel@tonic-gate int l; 11477c478bd9Sstevel@tonic-gate 11487c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 11497c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 1150250b7ff9Sjosephb total += hat->hat_ism_pgcnt; 11517c478bd9Sstevel@tonic-gate 11527c478bd9Sstevel@tonic-gate return (total); 11537c478bd9Sstevel@tonic-gate } 11547c478bd9Sstevel@tonic-gate 11557c478bd9Sstevel@tonic-gate /* 11567c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 11577c478bd9Sstevel@tonic-gate */ 11587c478bd9Sstevel@tonic-gate int 11597c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 11607c478bd9Sstevel@tonic-gate { 11617c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, 1); 11627c478bd9Sstevel@tonic-gate return (1); 11637c478bd9Sstevel@tonic-gate } 11647c478bd9Sstevel@tonic-gate 11657c478bd9Sstevel@tonic-gate void 11667c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 11677c478bd9Sstevel@tonic-gate { 11687c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, -1); 11697c478bd9Sstevel@tonic-gate } 11707c478bd9Sstevel@tonic-gate 11717c478bd9Sstevel@tonic-gate /* 11727c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 11737c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 11747c478bd9Sstevel@tonic-gate */ 11757c478bd9Sstevel@tonic-gate static void 11767c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 11777c478bd9Sstevel@tonic-gate { 11787c478bd9Sstevel@tonic-gate uint_t rm = 0; 11797c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 11807c478bd9Sstevel@tonic-gate 1181ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 11827c478bd9Sstevel@tonic-gate return; 11837c478bd9Sstevel@tonic-gate 11847c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 11857c478bd9Sstevel@tonic-gate rm |= P_REF; 11867c478bd9Sstevel@tonic-gate 11877c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 11887c478bd9Sstevel@tonic-gate rm |= P_MOD; 11897c478bd9Sstevel@tonic-gate 11907c478bd9Sstevel@tonic-gate if (rm == 0) 11917c478bd9Sstevel@tonic-gate return; 11927c478bd9Sstevel@tonic-gate 11937c478bd9Sstevel@tonic-gate /* 11947c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 11957c478bd9Sstevel@tonic-gate */ 11967c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 11977c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 11987c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 11997c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 12007c478bd9Sstevel@tonic-gate /* 12017c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 12027c478bd9Sstevel@tonic-gate * pszc below this mapping size 12037c478bd9Sstevel@tonic-gate * since this large mapping existed after we 12047c478bd9Sstevel@tonic-gate * took mlist lock. 12057c478bd9Sstevel@tonic-gate */ 12067c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 12077c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 12087c478bd9Sstevel@tonic-gate ++pp; 12097c478bd9Sstevel@tonic-gate } 12107c478bd9Sstevel@tonic-gate } 12117c478bd9Sstevel@tonic-gate 12127c478bd9Sstevel@tonic-gate /* 12137c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1214*843e1988Sjohnlev * that are allowed to change on a HAT_LOAD_REMAP 12157c478bd9Sstevel@tonic-gate */ 12167c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 12177c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 1218*843e1988Sjohnlev PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE | PT_IGNORE | PT_REF | PT_MOD) 12197c478bd9Sstevel@tonic-gate 1220b193e412Skchow #define REMAPASSERT(EX) if (!(EX)) panic("hati_pte_map: " #EX) 12217c478bd9Sstevel@tonic-gate /* 12227c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 12237c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 12247c478bd9Sstevel@tonic-gate */ 1225ae115bc7Smrj static int 12267c478bd9Sstevel@tonic-gate hati_pte_map( 12277c478bd9Sstevel@tonic-gate htable_t *ht, 12287c478bd9Sstevel@tonic-gate uint_t entry, 12297c478bd9Sstevel@tonic-gate page_t *pp, 12307c478bd9Sstevel@tonic-gate x86pte_t pte, 12317c478bd9Sstevel@tonic-gate int flags, 12327c478bd9Sstevel@tonic-gate void *pte_ptr) 12337c478bd9Sstevel@tonic-gate { 12347c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 12357c478bd9Sstevel@tonic-gate x86pte_t old_pte; 12367c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 12377c478bd9Sstevel@tonic-gate hment_t *hm; 12387c478bd9Sstevel@tonic-gate uint_t is_consist; 1239ae115bc7Smrj int rv = 0; 12407c478bd9Sstevel@tonic-gate 12417c478bd9Sstevel@tonic-gate /* 12427c478bd9Sstevel@tonic-gate * Is this a consistant (ie. need mapping list lock) mapping? 12437c478bd9Sstevel@tonic-gate */ 12447c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 12457c478bd9Sstevel@tonic-gate 12467c478bd9Sstevel@tonic-gate /* 12477c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 12487c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 12497c478bd9Sstevel@tonic-gate */ 12507c478bd9Sstevel@tonic-gate if ((flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat) 12517c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 12527c478bd9Sstevel@tonic-gate 12537c478bd9Sstevel@tonic-gate /* 12547c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 12557c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 12567c478bd9Sstevel@tonic-gate */ 12577c478bd9Sstevel@tonic-gate if (is_consist) { 12587c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 12597c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 12607c478bd9Sstevel@tonic-gate } 12617c478bd9Sstevel@tonic-gate 12627c478bd9Sstevel@tonic-gate /* 12637c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 12647c478bd9Sstevel@tonic-gate */ 12657c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 12667c478bd9Sstevel@tonic-gate 12677c478bd9Sstevel@tonic-gate /* 1268ae115bc7Smrj * did we get a large page / page table collision? 12697c478bd9Sstevel@tonic-gate */ 1270ae115bc7Smrj if (old_pte == LPAGE_ERROR) { 1271ae115bc7Smrj rv = -1; 1272ae115bc7Smrj goto done; 12737c478bd9Sstevel@tonic-gate } 12747c478bd9Sstevel@tonic-gate 1275ae115bc7Smrj /* 1276ae115bc7Smrj * If the mapping didn't change there is nothing more to do. 1277ae115bc7Smrj */ 1278ae115bc7Smrj if (PTE_EQUIV(pte, old_pte)) 1279ae115bc7Smrj goto done; 1280ae115bc7Smrj 12817c478bd9Sstevel@tonic-gate /* 12827c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 12837c478bd9Sstevel@tonic-gate */ 12847c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 12857c478bd9Sstevel@tonic-gate if (is_consist) { 12867c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 12877c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 12887c478bd9Sstevel@tonic-gate } else { 12897c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 12907c478bd9Sstevel@tonic-gate } 12917c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 12927c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1293ae115bc7Smrj return (rv); 12947c478bd9Sstevel@tonic-gate } 12957c478bd9Sstevel@tonic-gate 12967c478bd9Sstevel@tonic-gate /* 12977c478bd9Sstevel@tonic-gate * Remap's are more complicated: 12987c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 12997c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 13007c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 13017c478bd9Sstevel@tonic-gate */ 13027c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 13037c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 13047c478bd9Sstevel@tonic-gate 13057c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1306b193e412Skchow REMAPASSERT(flags & HAT_LOAD_REMAP); 1307b193e412Skchow REMAPASSERT(flags & HAT_LOAD_NOCONSIST); 1308ae115bc7Smrj REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 1309b193e412Skchow REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 13107c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1311b193e412Skchow REMAPASSERT(!is_consist); 13127c478bd9Sstevel@tonic-gate } 13137c478bd9Sstevel@tonic-gate 13147c478bd9Sstevel@tonic-gate /* 1315*843e1988Sjohnlev * We only let remaps change the certain bits in the PTE. 13167c478bd9Sstevel@tonic-gate */ 1317*843e1988Sjohnlev if (PTE_GET(old_pte, ~PT_REMAP_BITS) != PTE_GET(pte, ~PT_REMAP_BITS)) 1318*843e1988Sjohnlev panic("remap bits changed: old_pte="FMT_PTE", pte="FMT_PTE"\n", 1319*843e1988Sjohnlev old_pte, pte); 13207c478bd9Sstevel@tonic-gate 13217c478bd9Sstevel@tonic-gate /* 13227c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 13237c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 13247c478bd9Sstevel@tonic-gate */ 1325ae115bc7Smrj done: 13267c478bd9Sstevel@tonic-gate if (is_consist) { 13277c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 13287c478bd9Sstevel@tonic-gate if (hm != NULL) 13297c478bd9Sstevel@tonic-gate hment_free(hm); 13307c478bd9Sstevel@tonic-gate } 1331ae115bc7Smrj return (rv); 13327c478bd9Sstevel@tonic-gate } 13337c478bd9Sstevel@tonic-gate 13347c478bd9Sstevel@tonic-gate /* 1335ae115bc7Smrj * Internal routine to load a single page table entry. This only fails if 1336ae115bc7Smrj * we attempt to overwrite a page table link with a large page. 13377c478bd9Sstevel@tonic-gate */ 1338ae115bc7Smrj static int 13397c478bd9Sstevel@tonic-gate hati_load_common( 13407c478bd9Sstevel@tonic-gate hat_t *hat, 13417c478bd9Sstevel@tonic-gate uintptr_t va, 13427c478bd9Sstevel@tonic-gate page_t *pp, 13437c478bd9Sstevel@tonic-gate uint_t attr, 13447c478bd9Sstevel@tonic-gate uint_t flags, 13457c478bd9Sstevel@tonic-gate level_t level, 13467c478bd9Sstevel@tonic-gate pfn_t pfn) 13477c478bd9Sstevel@tonic-gate { 13487c478bd9Sstevel@tonic-gate htable_t *ht; 13497c478bd9Sstevel@tonic-gate uint_t entry; 13507c478bd9Sstevel@tonic-gate x86pte_t pte; 1351ae115bc7Smrj int rv = 0; 13527c478bd9Sstevel@tonic-gate 1353aac11643Sjosephb /* 1354aac11643Sjosephb * The number 16 is arbitrary and here to catch a recursion problem 1355aac11643Sjosephb * early before we blow out the kernel stack. 1356aac11643Sjosephb */ 1357aac11643Sjosephb ++curthread->t_hatdepth; 1358aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1359aac11643Sjosephb 13607c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 13617c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 13627c478bd9Sstevel@tonic-gate 13637c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 13647c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 13657c478bd9Sstevel@tonic-gate 13667c478bd9Sstevel@tonic-gate /* 13677c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 13687c478bd9Sstevel@tonic-gate */ 13697c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 13707c478bd9Sstevel@tonic-gate 13717c478bd9Sstevel@tonic-gate /* 1372aac11643Sjosephb * We must have HAT_LOAD_NOCONSIST if page_t is NULL. 13737c478bd9Sstevel@tonic-gate */ 1374aac11643Sjosephb if (pp == NULL) 13757c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 13767c478bd9Sstevel@tonic-gate 13777c478bd9Sstevel@tonic-gate if (ht == NULL) { 13787c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 13797c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 13807c478bd9Sstevel@tonic-gate } 13817c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 13827c478bd9Sstevel@tonic-gate 13837c478bd9Sstevel@tonic-gate /* 13847c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 13857c478bd9Sstevel@tonic-gate */ 13867c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 13877c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 13887c478bd9Sstevel@tonic-gate panic("hati_load_common: bad htable %p, va %p", ht, (void *)va); 13897c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 13907c478bd9Sstevel@tonic-gate 13917c478bd9Sstevel@tonic-gate /* 13927c478bd9Sstevel@tonic-gate * construct the new PTE 13937c478bd9Sstevel@tonic-gate */ 13947c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 13957c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 13967c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 13977c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 13987c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 13997c478bd9Sstevel@tonic-gate 14007c478bd9Sstevel@tonic-gate /* 14017c478bd9Sstevel@tonic-gate * establish the mapping 14027c478bd9Sstevel@tonic-gate */ 1403ae115bc7Smrj rv = hati_pte_map(ht, entry, pp, pte, flags, NULL); 14047c478bd9Sstevel@tonic-gate 14057c478bd9Sstevel@tonic-gate /* 14067c478bd9Sstevel@tonic-gate * release the htable and any reserves 14077c478bd9Sstevel@tonic-gate */ 14087c478bd9Sstevel@tonic-gate htable_release(ht); 1409aac11643Sjosephb --curthread->t_hatdepth; 1410ae115bc7Smrj return (rv); 14117c478bd9Sstevel@tonic-gate } 14127c478bd9Sstevel@tonic-gate 14137c478bd9Sstevel@tonic-gate /* 14147c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 14157c478bd9Sstevel@tonic-gate */ 14167c478bd9Sstevel@tonic-gate static void 14177c478bd9Sstevel@tonic-gate hat_kmap_load( 14187c478bd9Sstevel@tonic-gate caddr_t addr, 14197c478bd9Sstevel@tonic-gate page_t *pp, 14207c478bd9Sstevel@tonic-gate uint_t attr, 14217c478bd9Sstevel@tonic-gate uint_t flags) 14227c478bd9Sstevel@tonic-gate { 14237c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 14247c478bd9Sstevel@tonic-gate x86pte_t pte; 14257c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 14267c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 14277c478bd9Sstevel@tonic-gate htable_t *ht; 14287c478bd9Sstevel@tonic-gate uint_t entry; 14297c478bd9Sstevel@tonic-gate void *pte_ptr; 14307c478bd9Sstevel@tonic-gate 14317c478bd9Sstevel@tonic-gate /* 14327c478bd9Sstevel@tonic-gate * construct the requested PTE 14337c478bd9Sstevel@tonic-gate */ 14347c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 14357c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 14367c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 14377c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 14387c478bd9Sstevel@tonic-gate 14397c478bd9Sstevel@tonic-gate /* 14407c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 14417c478bd9Sstevel@tonic-gate */ 14427c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 14437c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 14447c478bd9Sstevel@tonic-gate else 14457c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 14467c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 14477c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 14487c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1449aac11643Sjosephb ++curthread->t_hatdepth; 1450aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1451ae115bc7Smrj (void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1452aac11643Sjosephb --curthread->t_hatdepth; 14537c478bd9Sstevel@tonic-gate } 14547c478bd9Sstevel@tonic-gate 14557c478bd9Sstevel@tonic-gate /* 14567c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 14577c478bd9Sstevel@tonic-gate * 14587c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 14597c478bd9Sstevel@tonic-gate * 14607c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 14617c478bd9Sstevel@tonic-gate * 14627c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 14637c478bd9Sstevel@tonic-gate * and hat_devload(). 14647c478bd9Sstevel@tonic-gate * 14657c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1466ae115bc7Smrj * sets PT_NOCONSIST 14677c478bd9Sstevel@tonic-gate * 14687c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 14697c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 14707c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 14717c478bd9Sstevel@tonic-gate * 14727c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 14737c478bd9Sstevel@tonic-gate * 14747c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 14757c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 14767c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 14777c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 14787c478bd9Sstevel@tonic-gate * recursion. 14797c478bd9Sstevel@tonic-gate * 14807c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 14817c478bd9Sstevel@tonic-gate * 1482ae115bc7Smrj * HAT_NOSYNC set PT_NOSYNC - this mapping's ref/mod bits 14837c478bd9Sstevel@tonic-gate * are never cleared. 14847c478bd9Sstevel@tonic-gate * 14857c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 14867c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 14877c478bd9Sstevel@tonic-gate * page_t being mapped. 14887c478bd9Sstevel@tonic-gate */ 14897c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 14907c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 14917c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 14927c478bd9Sstevel@tonic-gate 14937c478bd9Sstevel@tonic-gate void 14947c478bd9Sstevel@tonic-gate hat_memload( 14957c478bd9Sstevel@tonic-gate hat_t *hat, 14967c478bd9Sstevel@tonic-gate caddr_t addr, 14977c478bd9Sstevel@tonic-gate page_t *pp, 14987c478bd9Sstevel@tonic-gate uint_t attr, 14997c478bd9Sstevel@tonic-gate uint_t flags) 15007c478bd9Sstevel@tonic-gate { 15017c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15027c478bd9Sstevel@tonic-gate level_t level = 0; 15037c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 15047c478bd9Sstevel@tonic-gate 1505*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 15067c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1507ae115bc7Smrj ASSERT(hat == kas.a_hat || va < _userlimit); 15087c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 15097c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 15107c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 15117c478bd9Sstevel@tonic-gate 15127c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 15137c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 15147c478bd9Sstevel@tonic-gate 15157c478bd9Sstevel@tonic-gate /* 15167c478bd9Sstevel@tonic-gate * kernel address special case for performance. 15177c478bd9Sstevel@tonic-gate */ 15187c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 15197c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 15207c478bd9Sstevel@tonic-gate hat_kmap_load(addr, pp, attr, flags); 1521*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 15227c478bd9Sstevel@tonic-gate return; 15237c478bd9Sstevel@tonic-gate } 15247c478bd9Sstevel@tonic-gate 15257c478bd9Sstevel@tonic-gate /* 15267c478bd9Sstevel@tonic-gate * This is used for memory with normal caching enabled, so 15277c478bd9Sstevel@tonic-gate * always set HAT_STORECACHING_OK. 15287c478bd9Sstevel@tonic-gate */ 15297c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1530ae115bc7Smrj if (hati_load_common(hat, va, pp, attr, flags, level, pfn) != 0) 1531ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1532*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 15337c478bd9Sstevel@tonic-gate } 15347c478bd9Sstevel@tonic-gate 153505d3dc4bSpaulsan /* ARGSUSED */ 153605d3dc4bSpaulsan void 153705d3dc4bSpaulsan hat_memload_region(struct hat *hat, caddr_t addr, struct page *pp, 153805d3dc4bSpaulsan uint_t attr, uint_t flags, hat_region_cookie_t rcookie) 153905d3dc4bSpaulsan { 154005d3dc4bSpaulsan hat_memload(hat, addr, pp, attr, flags); 154105d3dc4bSpaulsan } 154205d3dc4bSpaulsan 15437c478bd9Sstevel@tonic-gate /* 15447c478bd9Sstevel@tonic-gate * Load the given array of page structs using large pages when possible 15457c478bd9Sstevel@tonic-gate */ 15467c478bd9Sstevel@tonic-gate void 15477c478bd9Sstevel@tonic-gate hat_memload_array( 15487c478bd9Sstevel@tonic-gate hat_t *hat, 15497c478bd9Sstevel@tonic-gate caddr_t addr, 15507c478bd9Sstevel@tonic-gate size_t len, 15517c478bd9Sstevel@tonic-gate page_t **pages, 15527c478bd9Sstevel@tonic-gate uint_t attr, 15537c478bd9Sstevel@tonic-gate uint_t flags) 15547c478bd9Sstevel@tonic-gate { 15557c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15567c478bd9Sstevel@tonic-gate uintptr_t eaddr = va + len; 15577c478bd9Sstevel@tonic-gate level_t level; 15587c478bd9Sstevel@tonic-gate size_t pgsize; 15597c478bd9Sstevel@tonic-gate pgcnt_t pgindx = 0; 15607c478bd9Sstevel@tonic-gate pfn_t pfn; 15617c478bd9Sstevel@tonic-gate pgcnt_t i; 15627c478bd9Sstevel@tonic-gate 1563*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 15647c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1565ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 15667c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 15677c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 15687c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 15697c478bd9Sstevel@tonic-gate 15707c478bd9Sstevel@tonic-gate /* 15717c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 15727c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 15737c478bd9Sstevel@tonic-gate */ 15747c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 15757c478bd9Sstevel@tonic-gate 15767c478bd9Sstevel@tonic-gate /* 15777c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 15787c478bd9Sstevel@tonic-gate */ 15797c478bd9Sstevel@tonic-gate while (va < eaddr) { 15807c478bd9Sstevel@tonic-gate /* 15817c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 15827c478bd9Sstevel@tonic-gate */ 15837c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 15847c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 15857c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 15867c478bd9Sstevel@tonic-gate if (level == 0) 15877c478bd9Sstevel@tonic-gate break; 1588ae115bc7Smrj 15897c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 15907c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1591ae115bc7Smrj !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize)) 15927c478bd9Sstevel@tonic-gate continue; 15937c478bd9Sstevel@tonic-gate 15947c478bd9Sstevel@tonic-gate /* 15957c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 15967c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 15977c478bd9Sstevel@tonic-gate * of the large page. 15987c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 15997c478bd9Sstevel@tonic-gate * all pages are locked. 16007c478bd9Sstevel@tonic-gate */ 16017c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 16027c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 16037c478bd9Sstevel@tonic-gate if (pfn + i != 16047c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 16057c478bd9Sstevel@tonic-gate break; 16067c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 16077c478bd9Sstevel@tonic-gate level); 16087c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 16097c478bd9Sstevel@tonic-gate pages[pgindx + i]); 16107c478bd9Sstevel@tonic-gate } 16117c478bd9Sstevel@tonic-gate if (i == mmu_btop(pgsize)) 16127c478bd9Sstevel@tonic-gate break; 16137c478bd9Sstevel@tonic-gate } 16147c478bd9Sstevel@tonic-gate } 16157c478bd9Sstevel@tonic-gate 16167c478bd9Sstevel@tonic-gate /* 1617ae115bc7Smrj * Load this page mapping. If the load fails, try a smaller 1618ae115bc7Smrj * pagesize. 16197c478bd9Sstevel@tonic-gate */ 16207c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1621ae115bc7Smrj while (hati_load_common(hat, va, pages[pgindx], attr, 1622250b7ff9Sjosephb flags, level, pfn) != 0) { 1623ae115bc7Smrj if (level == 0) 1624ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1625ae115bc7Smrj --level; 1626ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1627ae115bc7Smrj } 16287c478bd9Sstevel@tonic-gate 16297c478bd9Sstevel@tonic-gate /* 16307c478bd9Sstevel@tonic-gate * move to next page 16317c478bd9Sstevel@tonic-gate */ 16327c478bd9Sstevel@tonic-gate va += pgsize; 16337c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 16347c478bd9Sstevel@tonic-gate } 1635*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 16367c478bd9Sstevel@tonic-gate } 16377c478bd9Sstevel@tonic-gate 163805d3dc4bSpaulsan /* ARGSUSED */ 163905d3dc4bSpaulsan void 164005d3dc4bSpaulsan hat_memload_array_region(struct hat *hat, caddr_t addr, size_t len, 164105d3dc4bSpaulsan struct page **pps, uint_t attr, uint_t flags, 164205d3dc4bSpaulsan hat_region_cookie_t rcookie) 164305d3dc4bSpaulsan { 164405d3dc4bSpaulsan hat_memload_array(hat, addr, len, pps, attr, flags); 164505d3dc4bSpaulsan } 164605d3dc4bSpaulsan 16477c478bd9Sstevel@tonic-gate /* 16487c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 16497c478bd9Sstevel@tonic-gate * load/lock the given page frame number 16507c478bd9Sstevel@tonic-gate * 16517c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 16527c478bd9Sstevel@tonic-gate * 16537c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 16547c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 16557c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 16567c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 16577c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 16587c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 16597c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 16607c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 16617c478bd9Sstevel@tonic-gate * This also implies re-ordering. 16627c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 16637c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 16647c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 16657c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 16667c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 16677c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 16687c478bd9Sstevel@tonic-gate * 16697c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 16707c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 16717c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 16727c478bd9Sstevel@tonic-gate */ 16737c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 16747c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 16757c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 16767c478bd9Sstevel@tonic-gate 16777c478bd9Sstevel@tonic-gate void 16787c478bd9Sstevel@tonic-gate hat_devload( 16797c478bd9Sstevel@tonic-gate hat_t *hat, 16807c478bd9Sstevel@tonic-gate caddr_t addr, 16817c478bd9Sstevel@tonic-gate size_t len, 16827c478bd9Sstevel@tonic-gate pfn_t pfn, 16837c478bd9Sstevel@tonic-gate uint_t attr, 16847c478bd9Sstevel@tonic-gate int flags) 16857c478bd9Sstevel@tonic-gate { 16867c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 16877c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 16887c478bd9Sstevel@tonic-gate level_t level; 16897c478bd9Sstevel@tonic-gate size_t pgsize; 16907c478bd9Sstevel@tonic-gate page_t *pp; 16917c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 16927c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 16937c478bd9Sstevel@tonic-gate 1694*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 16957c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1696ae115bc7Smrj ASSERT(hat == kas.a_hat || eva <= _userlimit); 16977c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 16987c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16997c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 17007c478bd9Sstevel@tonic-gate 17017c478bd9Sstevel@tonic-gate /* 17027c478bd9Sstevel@tonic-gate * handle all pages 17037c478bd9Sstevel@tonic-gate */ 17047c478bd9Sstevel@tonic-gate while (va < eva) { 17057c478bd9Sstevel@tonic-gate 17067c478bd9Sstevel@tonic-gate /* 17077c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 17087c478bd9Sstevel@tonic-gate */ 17097c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 17107c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 17117c478bd9Sstevel@tonic-gate if (level == 0) 17127c478bd9Sstevel@tonic-gate break; 17137c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 17147c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 17157c478bd9Sstevel@tonic-gate IS_P2ALIGNED(pfn, mmu_btop(pgsize))) 17167c478bd9Sstevel@tonic-gate break; 17177c478bd9Sstevel@tonic-gate } 17187c478bd9Sstevel@tonic-gate 17197c478bd9Sstevel@tonic-gate /* 1720ae115bc7Smrj * If this is just memory then allow caching (this happens 17217c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1722ae115bc7Smrj * to override that. If we don't have a page_t then make sure 17237c478bd9Sstevel@tonic-gate * NOCONSIST is set. 17247c478bd9Sstevel@tonic-gate */ 17257c478bd9Sstevel@tonic-gate a = attr; 17267c478bd9Sstevel@tonic-gate f = flags; 1727*843e1988Sjohnlev if (!pf_is_memory(pfn)) 1728*843e1988Sjohnlev f |= HAT_LOAD_NOCONSIST; 1729*843e1988Sjohnlev else if (!(a & HAT_PLAT_NOCACHE)) 1730*843e1988Sjohnlev a |= HAT_STORECACHING_OK; 17317c478bd9Sstevel@tonic-gate 1732*843e1988Sjohnlev if (f & HAT_LOAD_NOCONSIST) 17337c478bd9Sstevel@tonic-gate pp = NULL; 1734*843e1988Sjohnlev else 1735*843e1988Sjohnlev pp = page_numtopp_nolock(pfn); 17367c478bd9Sstevel@tonic-gate 17377c478bd9Sstevel@tonic-gate /* 17387c478bd9Sstevel@tonic-gate * load this page mapping 17397c478bd9Sstevel@tonic-gate */ 17407c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1741ae115bc7Smrj while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) { 1742ae115bc7Smrj if (level == 0) 1743ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1744ae115bc7Smrj --level; 1745ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1746ae115bc7Smrj } 17477c478bd9Sstevel@tonic-gate 17487c478bd9Sstevel@tonic-gate /* 17497c478bd9Sstevel@tonic-gate * move to next page 17507c478bd9Sstevel@tonic-gate */ 17517c478bd9Sstevel@tonic-gate va += pgsize; 17527c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 17537c478bd9Sstevel@tonic-gate } 1754*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 17557c478bd9Sstevel@tonic-gate } 17567c478bd9Sstevel@tonic-gate 17577c478bd9Sstevel@tonic-gate /* 17587c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 17597c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 17607c478bd9Sstevel@tonic-gate * 17617c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 17627c478bd9Sstevel@tonic-gate */ 17637c478bd9Sstevel@tonic-gate void 17647c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 17657c478bd9Sstevel@tonic-gate { 17667c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 17677c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 17687c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 17697c478bd9Sstevel@tonic-gate 17707c478bd9Sstevel@tonic-gate /* 17717c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 17727c478bd9Sstevel@tonic-gate */ 1773ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 17747c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 17757c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 17767c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 17777c478bd9Sstevel@tonic-gate return; 17787c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 17797c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 17807c478bd9Sstevel@tonic-gate 1781*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 17827c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 17837c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 17847c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 17857c478bd9Sstevel@tonic-gate if (ht == NULL) 17867c478bd9Sstevel@tonic-gate break; 17877c478bd9Sstevel@tonic-gate 17887c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 17897c478bd9Sstevel@tonic-gate 17907c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 17917c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 17927c478bd9Sstevel@tonic-gate "htable=%p, vaddr=%p\n", ht, (caddr_t)vaddr); 17937c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 17947c478bd9Sstevel@tonic-gate 17957c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 17967c478bd9Sstevel@tonic-gate } 17977c478bd9Sstevel@tonic-gate if (ht) 17987c478bd9Sstevel@tonic-gate htable_release(ht); 1799*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 18007c478bd9Sstevel@tonic-gate } 18017c478bd9Sstevel@tonic-gate 180205d3dc4bSpaulsan /* ARGSUSED */ 180305d3dc4bSpaulsan void 18047dacfc44Spaulsan hat_unlock_region(struct hat *hat, caddr_t addr, size_t len, 180505d3dc4bSpaulsan hat_region_cookie_t rcookie) 180605d3dc4bSpaulsan { 180705d3dc4bSpaulsan panic("No shared region support on x86"); 180805d3dc4bSpaulsan } 180905d3dc4bSpaulsan 1810*843e1988Sjohnlev #if !defined(__xpv) 18117c478bd9Sstevel@tonic-gate /* 18127c478bd9Sstevel@tonic-gate * Cross call service routine to demap a virtual page on 18137c478bd9Sstevel@tonic-gate * the current CPU or flush all mappings in TLB. 18147c478bd9Sstevel@tonic-gate */ 18157c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 18167c478bd9Sstevel@tonic-gate static int 18177c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 18187c478bd9Sstevel@tonic-gate { 18197c478bd9Sstevel@tonic-gate hat_t *hat = (hat_t *)a1; 18207c478bd9Sstevel@tonic-gate caddr_t addr = (caddr_t)a2; 18217c478bd9Sstevel@tonic-gate 18227c478bd9Sstevel@tonic-gate /* 18237c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 18247c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 18257c478bd9Sstevel@tonic-gate */ 18267c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 18277c478bd9Sstevel@tonic-gate return (0); 18287c478bd9Sstevel@tonic-gate 18297c478bd9Sstevel@tonic-gate /* 18307c478bd9Sstevel@tonic-gate * For a normal address, we just flush one page mapping 18317c478bd9Sstevel@tonic-gate */ 18327c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1833ae115bc7Smrj mmu_tlbflush_entry(addr); 18347c478bd9Sstevel@tonic-gate return (0); 18357c478bd9Sstevel@tonic-gate } 18367c478bd9Sstevel@tonic-gate 18377c478bd9Sstevel@tonic-gate /* 18387c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 18397c478bd9Sstevel@tonic-gate * 18407c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 18417c478bd9Sstevel@tonic-gate * the pte values from the struct hat 18427c478bd9Sstevel@tonic-gate */ 18437c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 18447c478bd9Sstevel@tonic-gate #if defined(__amd64) 18457c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 18467c478bd9Sstevel@tonic-gate 18477c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 18487c478bd9Sstevel@tonic-gate #elif defined(__i386) 18497c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 18507c478bd9Sstevel@tonic-gate #endif 18517c478bd9Sstevel@tonic-gate } 18527c478bd9Sstevel@tonic-gate reload_cr3(); 18537c478bd9Sstevel@tonic-gate return (0); 18547c478bd9Sstevel@tonic-gate } 18557c478bd9Sstevel@tonic-gate 185695c0a3c8Sjosephb /* 185795c0a3c8Sjosephb * Flush all TLB entries, including global (ie. kernel) ones. 185895c0a3c8Sjosephb */ 185995c0a3c8Sjosephb static void 186095c0a3c8Sjosephb flush_all_tlb_entries(void) 186195c0a3c8Sjosephb { 186295c0a3c8Sjosephb ulong_t cr4 = getcr4(); 186395c0a3c8Sjosephb 186495c0a3c8Sjosephb if (cr4 & CR4_PGE) { 186595c0a3c8Sjosephb setcr4(cr4 & ~(ulong_t)CR4_PGE); 186695c0a3c8Sjosephb setcr4(cr4); 186795c0a3c8Sjosephb 186895c0a3c8Sjosephb /* 186995c0a3c8Sjosephb * 32 bit PAE also needs to always reload_cr3() 187095c0a3c8Sjosephb */ 187195c0a3c8Sjosephb if (mmu.max_level == 2) 187295c0a3c8Sjosephb reload_cr3(); 187395c0a3c8Sjosephb } else { 187495c0a3c8Sjosephb reload_cr3(); 187595c0a3c8Sjosephb } 187695c0a3c8Sjosephb } 187795c0a3c8Sjosephb 187895c0a3c8Sjosephb #define TLB_CPU_HALTED (01ul) 187995c0a3c8Sjosephb #define TLB_INVAL_ALL (02ul) 188095c0a3c8Sjosephb #define CAS_TLB_INFO(cpu, old, new) \ 188195c0a3c8Sjosephb caslong((ulong_t *)&(cpu)->cpu_m.mcpu_tlb_info, (old), (new)) 188295c0a3c8Sjosephb 188395c0a3c8Sjosephb /* 188495c0a3c8Sjosephb * Record that a CPU is going idle 188595c0a3c8Sjosephb */ 188695c0a3c8Sjosephb void 188795c0a3c8Sjosephb tlb_going_idle(void) 188895c0a3c8Sjosephb { 188995c0a3c8Sjosephb atomic_or_long((ulong_t *)&CPU->cpu_m.mcpu_tlb_info, TLB_CPU_HALTED); 189095c0a3c8Sjosephb } 189195c0a3c8Sjosephb 189295c0a3c8Sjosephb /* 189395c0a3c8Sjosephb * Service a delayed TLB flush if coming out of being idle. 189495c0a3c8Sjosephb */ 189595c0a3c8Sjosephb void 189695c0a3c8Sjosephb tlb_service(void) 189795c0a3c8Sjosephb { 189895c0a3c8Sjosephb ulong_t flags = getflags(); 189995c0a3c8Sjosephb ulong_t tlb_info; 190095c0a3c8Sjosephb ulong_t found; 190195c0a3c8Sjosephb 190295c0a3c8Sjosephb /* 190395c0a3c8Sjosephb * Be sure interrupts are off while doing this so that 190495c0a3c8Sjosephb * higher level interrupts correctly wait for flushes to finish. 190595c0a3c8Sjosephb */ 190695c0a3c8Sjosephb if (flags & PS_IE) 190795c0a3c8Sjosephb flags = intr_clear(); 190895c0a3c8Sjosephb 190995c0a3c8Sjosephb /* 191095c0a3c8Sjosephb * We only have to do something if coming out of being idle. 191195c0a3c8Sjosephb */ 191295c0a3c8Sjosephb tlb_info = CPU->cpu_m.mcpu_tlb_info; 191395c0a3c8Sjosephb if (tlb_info & TLB_CPU_HALTED) { 191495c0a3c8Sjosephb ASSERT(CPU->cpu_current_hat == kas.a_hat); 191595c0a3c8Sjosephb 191695c0a3c8Sjosephb /* 191795c0a3c8Sjosephb * Atomic clear and fetch of old state. 191895c0a3c8Sjosephb */ 191995c0a3c8Sjosephb while ((found = CAS_TLB_INFO(CPU, tlb_info, 0)) != tlb_info) { 192095c0a3c8Sjosephb ASSERT(found & TLB_CPU_HALTED); 192195c0a3c8Sjosephb tlb_info = found; 192295c0a3c8Sjosephb SMT_PAUSE(); 192395c0a3c8Sjosephb } 192495c0a3c8Sjosephb if (tlb_info & TLB_INVAL_ALL) 192595c0a3c8Sjosephb flush_all_tlb_entries(); 192695c0a3c8Sjosephb } 192795c0a3c8Sjosephb 192895c0a3c8Sjosephb /* 192995c0a3c8Sjosephb * Restore interrupt enable control bit. 193095c0a3c8Sjosephb */ 193195c0a3c8Sjosephb if (flags & PS_IE) 193295c0a3c8Sjosephb sti(); 193395c0a3c8Sjosephb } 1934*843e1988Sjohnlev #endif /* !__xpv */ 193595c0a3c8Sjosephb 19367c478bd9Sstevel@tonic-gate /* 19377c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 19387c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 19397c478bd9Sstevel@tonic-gate */ 19407c478bd9Sstevel@tonic-gate void 1941ae115bc7Smrj hat_tlb_inval(hat_t *hat, uintptr_t va) 19427c478bd9Sstevel@tonic-gate { 19437c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 19447c478bd9Sstevel@tonic-gate cpuset_t justme; 1945ae115bc7Smrj cpuset_t cpus_to_shootdown; 1946*843e1988Sjohnlev #ifndef __xpv 1947*843e1988Sjohnlev cpuset_t check_cpus; 194895c0a3c8Sjosephb cpu_t *cpup; 194995c0a3c8Sjosephb int c; 1950*843e1988Sjohnlev #endif 19517c478bd9Sstevel@tonic-gate 19527c478bd9Sstevel@tonic-gate /* 19537c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 19547c478bd9Sstevel@tonic-gate * demap need not do anything. 19557c478bd9Sstevel@tonic-gate */ 19567c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 19577c478bd9Sstevel@tonic-gate return; 19587c478bd9Sstevel@tonic-gate 19597c478bd9Sstevel@tonic-gate /* 19607c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 19617c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 19627c478bd9Sstevel@tonic-gate * these were shared at. 19637c478bd9Sstevel@tonic-gate */ 19647c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 19657c478bd9Sstevel@tonic-gate hat = kas.a_hat; 19667c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 19677c478bd9Sstevel@tonic-gate } 19687c478bd9Sstevel@tonic-gate 19697c478bd9Sstevel@tonic-gate /* 19707c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 19717c478bd9Sstevel@tonic-gate */ 19727c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 1973*843e1988Sjohnlev #ifdef __xpv 1974*843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 1975*843e1988Sjohnlev xen_flush_tlb(); 1976*843e1988Sjohnlev else 1977*843e1988Sjohnlev xen_flush_va((caddr_t)va); 1978*843e1988Sjohnlev #else 19797c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 1980*843e1988Sjohnlev #endif 19817c478bd9Sstevel@tonic-gate return; 19827c478bd9Sstevel@tonic-gate } 19837c478bd9Sstevel@tonic-gate 19847c478bd9Sstevel@tonic-gate 19857c478bd9Sstevel@tonic-gate /* 1986ae115bc7Smrj * Determine CPUs to shootdown. Kernel changes always do all CPUs. 1987ae115bc7Smrj * Otherwise it's just CPUs currently executing in this hat. 19887c478bd9Sstevel@tonic-gate */ 19897c478bd9Sstevel@tonic-gate kpreempt_disable(); 19907c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 1991ae115bc7Smrj if (hat == kas.a_hat) 1992ae115bc7Smrj cpus_to_shootdown = khat_cpuset; 19937c478bd9Sstevel@tonic-gate else 1994ae115bc7Smrj cpus_to_shootdown = hat->hat_cpus; 1995ae115bc7Smrj 1996*843e1988Sjohnlev #ifndef __xpv 199795c0a3c8Sjosephb /* 199895c0a3c8Sjosephb * If any CPUs in the set are idle, just request a delayed flush 199995c0a3c8Sjosephb * and avoid waking them up. 200095c0a3c8Sjosephb */ 200195c0a3c8Sjosephb check_cpus = cpus_to_shootdown; 200295c0a3c8Sjosephb for (c = 0; c < NCPU && !CPUSET_ISNULL(check_cpus); ++c) { 200395c0a3c8Sjosephb ulong_t tlb_info; 200495c0a3c8Sjosephb 200595c0a3c8Sjosephb if (!CPU_IN_SET(check_cpus, c)) 200695c0a3c8Sjosephb continue; 200795c0a3c8Sjosephb CPUSET_DEL(check_cpus, c); 200895c0a3c8Sjosephb cpup = cpu[c]; 200995c0a3c8Sjosephb if (cpup == NULL) 201095c0a3c8Sjosephb continue; 201195c0a3c8Sjosephb 201295c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 201395c0a3c8Sjosephb while (tlb_info == TLB_CPU_HALTED) { 201495c0a3c8Sjosephb (void) CAS_TLB_INFO(cpup, TLB_CPU_HALTED, 2015250b7ff9Sjosephb TLB_CPU_HALTED | TLB_INVAL_ALL); 201695c0a3c8Sjosephb SMT_PAUSE(); 201795c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 201895c0a3c8Sjosephb } 201995c0a3c8Sjosephb if (tlb_info == (TLB_CPU_HALTED | TLB_INVAL_ALL)) { 202095c0a3c8Sjosephb HATSTAT_INC(hs_tlb_inval_delayed); 202195c0a3c8Sjosephb CPUSET_DEL(cpus_to_shootdown, c); 202295c0a3c8Sjosephb } 202395c0a3c8Sjosephb } 2024*843e1988Sjohnlev #endif 202595c0a3c8Sjosephb 2026ae115bc7Smrj if (CPUSET_ISNULL(cpus_to_shootdown) || 2027ae115bc7Smrj CPUSET_ISEQUAL(cpus_to_shootdown, justme)) { 2028ae115bc7Smrj 2029*843e1988Sjohnlev #ifdef __xpv 2030*843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2031*843e1988Sjohnlev xen_flush_tlb(); 2032*843e1988Sjohnlev else 2033*843e1988Sjohnlev xen_flush_va((caddr_t)va); 2034*843e1988Sjohnlev #else 2035ae115bc7Smrj (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 2036*843e1988Sjohnlev #endif 2037ae115bc7Smrj 2038ae115bc7Smrj } else { 2039ae115bc7Smrj 2040ae115bc7Smrj CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id); 2041*843e1988Sjohnlev #ifdef __xpv 2042*843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2043*843e1988Sjohnlev xen_gflush_tlb(cpus_to_shootdown); 2044*843e1988Sjohnlev else 2045*843e1988Sjohnlev xen_gflush_va((caddr_t)va, cpus_to_shootdown); 2046*843e1988Sjohnlev #else 2047ae115bc7Smrj xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, X_CALL_HIPRI, 2048ae115bc7Smrj cpus_to_shootdown, hati_demap_func); 2049*843e1988Sjohnlev #endif 2050ae115bc7Smrj 2051ae115bc7Smrj } 20527c478bd9Sstevel@tonic-gate kpreempt_enable(); 20537c478bd9Sstevel@tonic-gate } 20547c478bd9Sstevel@tonic-gate 20557c478bd9Sstevel@tonic-gate /* 20567c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 20577c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 20587c478bd9Sstevel@tonic-gate * handle releasing of the htables. 20597c478bd9Sstevel@tonic-gate */ 20607c478bd9Sstevel@tonic-gate void 20617c478bd9Sstevel@tonic-gate hat_pte_unmap( 20627c478bd9Sstevel@tonic-gate htable_t *ht, 20637c478bd9Sstevel@tonic-gate uint_t entry, 20647c478bd9Sstevel@tonic-gate uint_t flags, 20657c478bd9Sstevel@tonic-gate x86pte_t old_pte, 20667c478bd9Sstevel@tonic-gate void *pte_ptr) 20677c478bd9Sstevel@tonic-gate { 20687c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 20697c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 20707c478bd9Sstevel@tonic-gate page_t *pp = NULL; 20717c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 20727c478bd9Sstevel@tonic-gate pfn_t pfn; 20737c478bd9Sstevel@tonic-gate 20747c478bd9Sstevel@tonic-gate /* 20757c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 20767c478bd9Sstevel@tonic-gate */ 20777c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 20787c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 20797c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 20807c478bd9Sstevel@tonic-gate } 20817c478bd9Sstevel@tonic-gate 20827c478bd9Sstevel@tonic-gate /* 20837c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 20847c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 20857c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 20867c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 20877c478bd9Sstevel@tonic-gate */ 20887c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 20897c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 20907c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 2091ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) { 20927c478bd9Sstevel@tonic-gate pp = NULL; 20937c478bd9Sstevel@tonic-gate } else { 2094*843e1988Sjohnlev #ifdef __xpv 2095*843e1988Sjohnlev if (pfn == PFN_INVALID) 2096*843e1988Sjohnlev panic("Invalid PFN, but not PT_NOCONSIST"); 2097*843e1988Sjohnlev #endif 20987c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 2099aa2ed9e5Sjosephb if (pp == NULL) { 2100aa2ed9e5Sjosephb panic("no page_t, not NOCONSIST: old_pte=" 2101aa2ed9e5Sjosephb FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx", 2102aa2ed9e5Sjosephb old_pte, (uintptr_t)ht, entry, 2103aa2ed9e5Sjosephb (uintptr_t)pte_ptr); 2104aa2ed9e5Sjosephb } 21057c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 21067c478bd9Sstevel@tonic-gate } 2107aa2ed9e5Sjosephb 2108aa2ed9e5Sjosephb /* 2109aa2ed9e5Sjosephb * If freeing the address space, check that the PTE 2110aa2ed9e5Sjosephb * hasn't changed, as the mappings are no longer in use by 2111aa2ed9e5Sjosephb * any thread, invalidation is unnecessary. 2112aa2ed9e5Sjosephb * If not freeing, do a full invalidate. 2113*843e1988Sjohnlev * 2114*843e1988Sjohnlev * On the hypervisor we must always remove mappings, as a 2115*843e1988Sjohnlev * writable mapping left behind could cause a page table 2116*843e1988Sjohnlev * allocation to fail. 2117aa2ed9e5Sjosephb */ 2118*843e1988Sjohnlev #if !defined(__xpv) 2119aa2ed9e5Sjosephb if (hat->hat_flags & HAT_FREEING) 2120aa2ed9e5Sjosephb old_pte = x86pte_get(ht, entry); 2121aa2ed9e5Sjosephb else 2122*843e1988Sjohnlev #endif 2123ae115bc7Smrj old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr); 21247c478bd9Sstevel@tonic-gate 21257c478bd9Sstevel@tonic-gate /* 21267c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 21277c478bd9Sstevel@tonic-gate */ 21287c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 21297c478bd9Sstevel@tonic-gate break; 21307c478bd9Sstevel@tonic-gate 21317c478bd9Sstevel@tonic-gate /* 21327c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 21337c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 21347c478bd9Sstevel@tonic-gate */ 21357c478bd9Sstevel@tonic-gate if (pp != NULL) { 21367c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 21377c478bd9Sstevel@tonic-gate pp = NULL; 21387c478bd9Sstevel@tonic-gate } else { 2139ae115bc7Smrj ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 21407c478bd9Sstevel@tonic-gate } 21417c478bd9Sstevel@tonic-gate } 21427c478bd9Sstevel@tonic-gate 21437c478bd9Sstevel@tonic-gate /* 21447c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 21457c478bd9Sstevel@tonic-gate */ 21467c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 21477c478bd9Sstevel@tonic-gate if (pp != NULL) 21487c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 21497c478bd9Sstevel@tonic-gate return; 21507c478bd9Sstevel@tonic-gate } 21517c478bd9Sstevel@tonic-gate 21527c478bd9Sstevel@tonic-gate /* 21537c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 21547c478bd9Sstevel@tonic-gate */ 21557c478bd9Sstevel@tonic-gate if (pp != NULL) { 21567c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 21577c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 21587c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 21597c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 21607c478bd9Sstevel@tonic-gate if (hm != NULL) 21617c478bd9Sstevel@tonic-gate hment_free(hm); 21627c478bd9Sstevel@tonic-gate } 21637c478bd9Sstevel@tonic-gate 21647c478bd9Sstevel@tonic-gate /* 21657c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 21667c478bd9Sstevel@tonic-gate */ 21677c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 21687c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 21697c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 21707c478bd9Sstevel@tonic-gate } 21717c478bd9Sstevel@tonic-gate 21727c478bd9Sstevel@tonic-gate /* 21737c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 21747c478bd9Sstevel@tonic-gate */ 21757c478bd9Sstevel@tonic-gate static void 21767c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 21777c478bd9Sstevel@tonic-gate { 21787c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 21797c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2180ae115bc7Smrj pgcnt_t pg_index; 21817c478bd9Sstevel@tonic-gate htable_t *ht; 21827c478bd9Sstevel@tonic-gate uint_t entry; 2183ae115bc7Smrj x86pte_t *pte_ptr; 21847c478bd9Sstevel@tonic-gate x86pte_t old_pte; 21857c478bd9Sstevel@tonic-gate 21867c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 21877c478bd9Sstevel@tonic-gate /* 21887c478bd9Sstevel@tonic-gate * Get the PTE 21897c478bd9Sstevel@tonic-gate */ 2190ae115bc7Smrj pg_index = mmu_btop(va - mmu.kmap_addr); 2191ae115bc7Smrj pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index); 2192ae115bc7Smrj old_pte = GET_PTE(pte_ptr); 21937c478bd9Sstevel@tonic-gate 21947c478bd9Sstevel@tonic-gate /* 21957c478bd9Sstevel@tonic-gate * get the htable / entry 21967c478bd9Sstevel@tonic-gate */ 21977c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 21987c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 21997c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 22007c478bd9Sstevel@tonic-gate 22017c478bd9Sstevel@tonic-gate /* 22027c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 22037c478bd9Sstevel@tonic-gate */ 22047c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr); 22057c478bd9Sstevel@tonic-gate } 22067c478bd9Sstevel@tonic-gate } 22077c478bd9Sstevel@tonic-gate 22087c478bd9Sstevel@tonic-gate 22097c478bd9Sstevel@tonic-gate /* 22107c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 22117c478bd9Sstevel@tonic-gate */ 22127c478bd9Sstevel@tonic-gate void 22137c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 22147c478bd9Sstevel@tonic-gate { 22157c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2216ae115bc7Smrj 2217*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2218ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 22197c478bd9Sstevel@tonic-gate 22207c478bd9Sstevel@tonic-gate /* 22217c478bd9Sstevel@tonic-gate * special case for performance. 22227c478bd9Sstevel@tonic-gate */ 22237c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 22247c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 22257c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2226ae115bc7Smrj } else { 2227ae115bc7Smrj hat_unload_callback(hat, addr, len, flags, NULL); 22287c478bd9Sstevel@tonic-gate } 2229*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 22307c478bd9Sstevel@tonic-gate } 22317c478bd9Sstevel@tonic-gate 22327c478bd9Sstevel@tonic-gate /* 22337c478bd9Sstevel@tonic-gate * Do the callbacks for ranges being unloaded. 22347c478bd9Sstevel@tonic-gate */ 22357c478bd9Sstevel@tonic-gate typedef struct range_info { 22367c478bd9Sstevel@tonic-gate uintptr_t rng_va; 22377c478bd9Sstevel@tonic-gate ulong_t rng_cnt; 22387c478bd9Sstevel@tonic-gate level_t rng_level; 22397c478bd9Sstevel@tonic-gate } range_info_t; 22407c478bd9Sstevel@tonic-gate 22417c478bd9Sstevel@tonic-gate static void 22427c478bd9Sstevel@tonic-gate handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range) 22437c478bd9Sstevel@tonic-gate { 22447c478bd9Sstevel@tonic-gate /* 22457c478bd9Sstevel@tonic-gate * do callbacks to upper level VM system 22467c478bd9Sstevel@tonic-gate */ 22477c478bd9Sstevel@tonic-gate while (cb != NULL && cnt > 0) { 22487c478bd9Sstevel@tonic-gate --cnt; 22497c478bd9Sstevel@tonic-gate cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 22507c478bd9Sstevel@tonic-gate cb->hcb_end_addr = cb->hcb_start_addr; 22517c478bd9Sstevel@tonic-gate cb->hcb_end_addr += 22527c478bd9Sstevel@tonic-gate range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level); 22537c478bd9Sstevel@tonic-gate cb->hcb_function(cb); 22547c478bd9Sstevel@tonic-gate } 22557c478bd9Sstevel@tonic-gate } 22567c478bd9Sstevel@tonic-gate 22577c478bd9Sstevel@tonic-gate /* 22587c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 22597c478bd9Sstevel@tonic-gate * 22607c478bd9Sstevel@tonic-gate * Flags: 22617c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 22627c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 22637c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 22647c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 22657c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 22667c478bd9Sstevel@tonic-gate */ 22677c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 22687c478bd9Sstevel@tonic-gate void 22697c478bd9Sstevel@tonic-gate hat_unload_callback( 22707c478bd9Sstevel@tonic-gate hat_t *hat, 22717c478bd9Sstevel@tonic-gate caddr_t addr, 22727c478bd9Sstevel@tonic-gate size_t len, 22737c478bd9Sstevel@tonic-gate uint_t flags, 22747c478bd9Sstevel@tonic-gate hat_callback_t *cb) 22757c478bd9Sstevel@tonic-gate { 22767c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 22777c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 22787c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 22797c478bd9Sstevel@tonic-gate uint_t entry; 2280aa2ed9e5Sjosephb uintptr_t contig_va = (uintptr_t)-1L; 22817c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 22827c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 22837c478bd9Sstevel@tonic-gate x86pte_t old_pte; 22847c478bd9Sstevel@tonic-gate 2285*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2286ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 22877c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 22887c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 22897c478bd9Sstevel@tonic-gate 2290ae115bc7Smrj /* 2291ae115bc7Smrj * Special case a single page being unloaded for speed. This happens 2292ae115bc7Smrj * quite frequently, COW faults after a fork() for example. 2293ae115bc7Smrj */ 2294ae115bc7Smrj if (cb == NULL && len == MMU_PAGESIZE) { 2295ae115bc7Smrj ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0); 2296ae115bc7Smrj if (ht != NULL) { 2297ae115bc7Smrj if (PTE_ISVALID(old_pte)) 2298ae115bc7Smrj hat_pte_unmap(ht, entry, flags, old_pte, NULL); 2299ae115bc7Smrj htable_release(ht); 2300ae115bc7Smrj } 2301*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2302ae115bc7Smrj return; 2303ae115bc7Smrj } 2304ae115bc7Smrj 23057c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 23067c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 23077c478bd9Sstevel@tonic-gate if (ht == NULL) 23087c478bd9Sstevel@tonic-gate break; 23097c478bd9Sstevel@tonic-gate 23107c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 23117c478bd9Sstevel@tonic-gate 23127c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 23137c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 23147c478bd9Sstevel@tonic-gate 23157c478bd9Sstevel@tonic-gate /* 23167c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 23177c478bd9Sstevel@tonic-gate */ 2318aa2ed9e5Sjosephb if (vaddr != contig_va || 23197c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 23207c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 23217c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 23227c478bd9Sstevel@tonic-gate r_cnt = 0; 23237c478bd9Sstevel@tonic-gate } 23247c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 23257c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 23267c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 23277c478bd9Sstevel@tonic-gate ++r_cnt; 23287c478bd9Sstevel@tonic-gate } 23297c478bd9Sstevel@tonic-gate 23307c478bd9Sstevel@tonic-gate /* 23317c478bd9Sstevel@tonic-gate * Unload one mapping from the page tables. 23327c478bd9Sstevel@tonic-gate */ 23337c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 23347c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, NULL); 23357c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 23367c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2337aa2ed9e5Sjosephb contig_va = vaddr; 23387c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 23397c478bd9Sstevel@tonic-gate } 23407c478bd9Sstevel@tonic-gate if (ht) 23417c478bd9Sstevel@tonic-gate htable_release(ht); 23427c478bd9Sstevel@tonic-gate 23437c478bd9Sstevel@tonic-gate /* 23447c478bd9Sstevel@tonic-gate * handle last range for callbacks 23457c478bd9Sstevel@tonic-gate */ 23467c478bd9Sstevel@tonic-gate if (r_cnt > 0) 23477c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 2348*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 23497c478bd9Sstevel@tonic-gate } 23507c478bd9Sstevel@tonic-gate 23517c478bd9Sstevel@tonic-gate /* 23527c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 23537c478bd9Sstevel@tonic-gate * 23547c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 23557c478bd9Sstevel@tonic-gate * driver. 23567c478bd9Sstevel@tonic-gate */ 23577c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 23587c478bd9Sstevel@tonic-gate void 23597c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 23607c478bd9Sstevel@tonic-gate { 23617c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 23627c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 23637c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23647c478bd9Sstevel@tonic-gate uint_t entry; 23657c478bd9Sstevel@tonic-gate x86pte_t pte; 23667c478bd9Sstevel@tonic-gate x86pte_t save_pte; 23677c478bd9Sstevel@tonic-gate x86pte_t new; 23687c478bd9Sstevel@tonic-gate page_t *pp; 23697c478bd9Sstevel@tonic-gate 23707c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 23717c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 23727c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2373ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 23747c478bd9Sstevel@tonic-gate 2375*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 23767c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 23777c478bd9Sstevel@tonic-gate try_again: 23787c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 23797c478bd9Sstevel@tonic-gate if (ht == NULL) 23807c478bd9Sstevel@tonic-gate break; 23817c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 23827c478bd9Sstevel@tonic-gate 2383ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 23847c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 23857c478bd9Sstevel@tonic-gate continue; 23867c478bd9Sstevel@tonic-gate 23877c478bd9Sstevel@tonic-gate /* 23887c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 23897c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 23907c478bd9Sstevel@tonic-gate */ 23917c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 23927c478bd9Sstevel@tonic-gate if (pp == NULL) 23937c478bd9Sstevel@tonic-gate break; 23947c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 23957c478bd9Sstevel@tonic-gate save_pte = pte; 23967c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 23977c478bd9Sstevel@tonic-gate if (pte != save_pte) { 23987c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 23997c478bd9Sstevel@tonic-gate goto try_again; 24007c478bd9Sstevel@tonic-gate } 2401ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 24027c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 24037c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24047c478bd9Sstevel@tonic-gate continue; 24057c478bd9Sstevel@tonic-gate } 24067c478bd9Sstevel@tonic-gate 24077c478bd9Sstevel@tonic-gate /* 24087c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 24097c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 24107c478bd9Sstevel@tonic-gate */ 24117c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 24127c478bd9Sstevel@tonic-gate new = pte; 24137c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 24147c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 24157c478bd9Sstevel@tonic-gate if (pte != 0) { 24167c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24177c478bd9Sstevel@tonic-gate goto try_again; 24187c478bd9Sstevel@tonic-gate } 24197c478bd9Sstevel@tonic-gate } else { 24207c478bd9Sstevel@tonic-gate /* 24217c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 24227c478bd9Sstevel@tonic-gate */ 24237c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 24247c478bd9Sstevel@tonic-gate } 24257c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24267c478bd9Sstevel@tonic-gate } 24277c478bd9Sstevel@tonic-gate if (ht) 24287c478bd9Sstevel@tonic-gate htable_release(ht); 2429*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 24307c478bd9Sstevel@tonic-gate } 24317c478bd9Sstevel@tonic-gate 24327c478bd9Sstevel@tonic-gate /* 24337c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 24347c478bd9Sstevel@tonic-gate */ 24357c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24367c478bd9Sstevel@tonic-gate void 24377c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 24387c478bd9Sstevel@tonic-gate { 24397c478bd9Sstevel@tonic-gate /* does nothing */ 24407c478bd9Sstevel@tonic-gate } 24417c478bd9Sstevel@tonic-gate 24427c478bd9Sstevel@tonic-gate /* 24437c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 24447c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 24457c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 24467c478bd9Sstevel@tonic-gate * *attr is not valid. 24477c478bd9Sstevel@tonic-gate */ 24487c478bd9Sstevel@tonic-gate uint_t 24497c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 24507c478bd9Sstevel@tonic-gate { 24517c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 24527c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 24537c478bd9Sstevel@tonic-gate x86pte_t pte; 24547c478bd9Sstevel@tonic-gate 2455ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 24567c478bd9Sstevel@tonic-gate 24577c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 24587c478bd9Sstevel@tonic-gate return ((uint_t)-1); 24597c478bd9Sstevel@tonic-gate 2460ae115bc7Smrj ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level); 24617c478bd9Sstevel@tonic-gate if (ht == NULL) 24627c478bd9Sstevel@tonic-gate return ((uint_t)-1); 24637c478bd9Sstevel@tonic-gate 24647c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 24657c478bd9Sstevel@tonic-gate htable_release(ht); 24667c478bd9Sstevel@tonic-gate return ((uint_t)-1); 24677c478bd9Sstevel@tonic-gate } 24687c478bd9Sstevel@tonic-gate 24697c478bd9Sstevel@tonic-gate *attr = PROT_READ; 24707c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 24717c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 24727c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 24737c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 24747c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 24757c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2476ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 24777c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 24787c478bd9Sstevel@tonic-gate htable_release(ht); 24797c478bd9Sstevel@tonic-gate return (0); 24807c478bd9Sstevel@tonic-gate } 24817c478bd9Sstevel@tonic-gate 24827c478bd9Sstevel@tonic-gate /* 24837c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 24847c478bd9Sstevel@tonic-gate */ 24857c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 24867c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 24877c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 24887c478bd9Sstevel@tonic-gate 24897c478bd9Sstevel@tonic-gate static void 24907c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 24917c478bd9Sstevel@tonic-gate { 24927c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 24937c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 24947c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 24957c478bd9Sstevel@tonic-gate uint_t entry; 24967c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 24977c478bd9Sstevel@tonic-gate page_t *pp; 24987c478bd9Sstevel@tonic-gate 2499*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 25007c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 25017c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 25027c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 25037c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 25047c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 25057c478bd9Sstevel@tonic-gate try_again: 25067c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 25077c478bd9Sstevel@tonic-gate if (ht == NULL) 25087c478bd9Sstevel@tonic-gate break; 2509ae115bc7Smrj if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST) 25107c478bd9Sstevel@tonic-gate continue; 25117c478bd9Sstevel@tonic-gate 25127c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 25137c478bd9Sstevel@tonic-gate if (pp == NULL) 25147c478bd9Sstevel@tonic-gate continue; 25157c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 25167c478bd9Sstevel@tonic-gate 25177c478bd9Sstevel@tonic-gate newpte = oldpte; 25187c478bd9Sstevel@tonic-gate /* 25197c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 25207c478bd9Sstevel@tonic-gate * figure out the new attributes. 25217c478bd9Sstevel@tonic-gate */ 25227c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 25237c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 25247c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 25257c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 25267c478bd9Sstevel@tonic-gate 2527ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2528ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC) 25297c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 25307c478bd9Sstevel@tonic-gate 25317c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 25327c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 25337c478bd9Sstevel@tonic-gate } 25347c478bd9Sstevel@tonic-gate 25357c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 25367c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 25377c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 25387c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 25397c478bd9Sstevel@tonic-gate 2540ae115bc7Smrj if (!(attr & HAT_NOSYNC) && 2541ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2542ae115bc7Smrj newpte &= ~PT_SOFTWARE; 25437c478bd9Sstevel@tonic-gate 25447c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 25457c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 25467c478bd9Sstevel@tonic-gate } 25477c478bd9Sstevel@tonic-gate 25487c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 25497c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 25507c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 25517c478bd9Sstevel@tonic-gate 2552ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2553ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2554ae115bc7Smrj newpte &= ~PT_SOFTWARE; 25557c478bd9Sstevel@tonic-gate 25567c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 25577c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 25587c478bd9Sstevel@tonic-gate } 25597c478bd9Sstevel@tonic-gate 2560ae115bc7Smrj /* 2561ae115bc7Smrj * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set. 2562ae115bc7Smrj * x86pte_set() depends on this. 2563ae115bc7Smrj */ 2564ae115bc7Smrj if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC) 2565ae115bc7Smrj newpte |= PT_REF | PT_MOD; 2566ae115bc7Smrj 25677c478bd9Sstevel@tonic-gate /* 25687c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 25697c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 25707c478bd9Sstevel@tonic-gate */ 25717c478bd9Sstevel@tonic-gate 25727c478bd9Sstevel@tonic-gate /* 25737c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 25747c478bd9Sstevel@tonic-gate */ 25757c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 25767c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 25777c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 25787c478bd9Sstevel@tonic-gate if (oldpte != 0) { 25797c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25807c478bd9Sstevel@tonic-gate goto try_again; 25817c478bd9Sstevel@tonic-gate } 25827c478bd9Sstevel@tonic-gate } 25837c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 25847c478bd9Sstevel@tonic-gate } 25857c478bd9Sstevel@tonic-gate if (ht) 25867c478bd9Sstevel@tonic-gate htable_release(ht); 2587*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 25887c478bd9Sstevel@tonic-gate } 25897c478bd9Sstevel@tonic-gate 25907c478bd9Sstevel@tonic-gate /* 25917c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 25927c478bd9Sstevel@tonic-gate */ 25937c478bd9Sstevel@tonic-gate void 25947c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 25957c478bd9Sstevel@tonic-gate { 2596ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 25977c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 25987c478bd9Sstevel@tonic-gate } 25997c478bd9Sstevel@tonic-gate 26007c478bd9Sstevel@tonic-gate void 26017c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26027c478bd9Sstevel@tonic-gate { 2603ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26047c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 26057c478bd9Sstevel@tonic-gate } 26067c478bd9Sstevel@tonic-gate 26077c478bd9Sstevel@tonic-gate void 26087c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26097c478bd9Sstevel@tonic-gate { 2610ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26117c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 26127c478bd9Sstevel@tonic-gate } 26137c478bd9Sstevel@tonic-gate 26147c478bd9Sstevel@tonic-gate void 26157c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 26167c478bd9Sstevel@tonic-gate { 2617ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26187c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 26197c478bd9Sstevel@tonic-gate } 26207c478bd9Sstevel@tonic-gate 26217c478bd9Sstevel@tonic-gate /* 26227c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 26237c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 26247c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 26257c478bd9Sstevel@tonic-gate */ 26267c478bd9Sstevel@tonic-gate ssize_t 26277c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 26287c478bd9Sstevel@tonic-gate { 26297c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 26307c478bd9Sstevel@tonic-gate htable_t *ht; 26317c478bd9Sstevel@tonic-gate size_t pagesize; 26327c478bd9Sstevel@tonic-gate 2633ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 26347c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 26357c478bd9Sstevel@tonic-gate return (-1); 26367c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 26377c478bd9Sstevel@tonic-gate if (ht == NULL) 26387c478bd9Sstevel@tonic-gate return (-1); 26397c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 26407c478bd9Sstevel@tonic-gate htable_release(ht); 26417c478bd9Sstevel@tonic-gate return (pagesize); 26427c478bd9Sstevel@tonic-gate } 26437c478bd9Sstevel@tonic-gate 26447c478bd9Sstevel@tonic-gate 26457c478bd9Sstevel@tonic-gate 26467c478bd9Sstevel@tonic-gate /* 26477c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 26487c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 26497c478bd9Sstevel@tonic-gate */ 26507c478bd9Sstevel@tonic-gate pfn_t 26517c478bd9Sstevel@tonic-gate hat_getpfnum(hat_t *hat, caddr_t addr) 26527c478bd9Sstevel@tonic-gate { 26537c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 26547c478bd9Sstevel@tonic-gate htable_t *ht; 26557c478bd9Sstevel@tonic-gate uint_t entry; 26567c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 26577c478bd9Sstevel@tonic-gate 2658ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 26597c478bd9Sstevel@tonic-gate if (khat_running == 0) 2660ae115bc7Smrj return (PFN_INVALID); 26617c478bd9Sstevel@tonic-gate 26627c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 26637c478bd9Sstevel@tonic-gate return (PFN_INVALID); 26647c478bd9Sstevel@tonic-gate 2665*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 26667c478bd9Sstevel@tonic-gate /* 26677c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 26687c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 26697c478bd9Sstevel@tonic-gate * this up. 26707c478bd9Sstevel@tonic-gate */ 26717c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 26727c478bd9Sstevel@tonic-gate x86pte_t pte; 2673ae115bc7Smrj pgcnt_t pg_index; 26747c478bd9Sstevel@tonic-gate 2675ae115bc7Smrj pg_index = mmu_btop(vaddr - mmu.kmap_addr); 2676ae115bc7Smrj pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index)); 2677*843e1988Sjohnlev if (PTE_ISVALID(pte)) 2678*843e1988Sjohnlev /*LINTED [use of constant 0 causes a lint warning] */ 2679*843e1988Sjohnlev pfn = PTE2PFN(pte, 0); 2680*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2681*843e1988Sjohnlev return (pfn); 26827c478bd9Sstevel@tonic-gate } 26837c478bd9Sstevel@tonic-gate 26847c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 2685*843e1988Sjohnlev if (ht == NULL) { 2686*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 26877c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2688*843e1988Sjohnlev } 26897c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 26907c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 26917c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 26927c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 26937c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 26947c478bd9Sstevel@tonic-gate htable_release(ht); 2695*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 26967c478bd9Sstevel@tonic-gate return (pfn); 26977c478bd9Sstevel@tonic-gate } 26987c478bd9Sstevel@tonic-gate 26997c478bd9Sstevel@tonic-gate /* 27007c478bd9Sstevel@tonic-gate * hat_getkpfnum() is an obsolete DDI routine, and its use is discouraged. 27017c478bd9Sstevel@tonic-gate * Use hat_getpfnum(kas.a_hat, ...) instead. 27027c478bd9Sstevel@tonic-gate * 27037c478bd9Sstevel@tonic-gate * We'd like to return PFN_INVALID if the mappings have underlying page_t's 27047c478bd9Sstevel@tonic-gate * but can't right now due to the fact that some software has grown to use 27057c478bd9Sstevel@tonic-gate * this interface incorrectly. So for now when the interface is misused, 27067c478bd9Sstevel@tonic-gate * return a warning to the user that in the future it won't work in the 27077c478bd9Sstevel@tonic-gate * way they're abusing it, and carry on. 27087c478bd9Sstevel@tonic-gate * 27097c478bd9Sstevel@tonic-gate * Note that hat_getkpfnum() is never supported on amd64. 27107c478bd9Sstevel@tonic-gate */ 27117c478bd9Sstevel@tonic-gate #if !defined(__amd64) 27127c478bd9Sstevel@tonic-gate pfn_t 27137c478bd9Sstevel@tonic-gate hat_getkpfnum(caddr_t addr) 27147c478bd9Sstevel@tonic-gate { 27157c478bd9Sstevel@tonic-gate pfn_t pfn; 27167c478bd9Sstevel@tonic-gate int badcaller = 0; 27177c478bd9Sstevel@tonic-gate 27187c478bd9Sstevel@tonic-gate if (khat_running == 0) 27197c478bd9Sstevel@tonic-gate panic("hat_getkpfnum(): called too early\n"); 27207c478bd9Sstevel@tonic-gate if ((uintptr_t)addr < kernelbase) 27217c478bd9Sstevel@tonic-gate return (PFN_INVALID); 27227c478bd9Sstevel@tonic-gate 2723*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 27247c478bd9Sstevel@tonic-gate if (segkpm && IS_KPM_ADDR(addr)) { 27257c478bd9Sstevel@tonic-gate badcaller = 1; 27267c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(addr); 27277c478bd9Sstevel@tonic-gate } else { 27287c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, addr); 27297c478bd9Sstevel@tonic-gate badcaller = pf_is_memory(pfn); 27307c478bd9Sstevel@tonic-gate } 27317c478bd9Sstevel@tonic-gate 27327c478bd9Sstevel@tonic-gate if (badcaller) 27337c478bd9Sstevel@tonic-gate hat_getkpfnum_badcall(caller()); 2734*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27357c478bd9Sstevel@tonic-gate return (pfn); 27367c478bd9Sstevel@tonic-gate } 27377c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 27387c478bd9Sstevel@tonic-gate 27397c478bd9Sstevel@tonic-gate /* 27407c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 27417c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 27427c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 27437c478bd9Sstevel@tonic-gate */ 27447c478bd9Sstevel@tonic-gate int 27457c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 27467c478bd9Sstevel@tonic-gate { 27477c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 27487c478bd9Sstevel@tonic-gate uint_t entry; 27497c478bd9Sstevel@tonic-gate htable_t *ht; 27507c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 27517c478bd9Sstevel@tonic-gate 2752ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27537c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 27547c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 27557c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 27567c478bd9Sstevel@tonic-gate return (0); 27577c478bd9Sstevel@tonic-gate 27587c478bd9Sstevel@tonic-gate /* 27597c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 27607c478bd9Sstevel@tonic-gate * for performance. 27617c478bd9Sstevel@tonic-gate */ 27627c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 27637c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 27647c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 27657c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 27667c478bd9Sstevel@tonic-gate else 27677c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 27687c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 27697c478bd9Sstevel@tonic-gate } 27707c478bd9Sstevel@tonic-gate 27717c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 27727c478bd9Sstevel@tonic-gate htable_release(ht); 2773*843e1988Sjohnlev return (ht != NULL); 27747c478bd9Sstevel@tonic-gate } 27757c478bd9Sstevel@tonic-gate 27767c478bd9Sstevel@tonic-gate /* 2777250b7ff9Sjosephb * Find out if the segment for hat_share()/hat_unshare() is DISM or locked ISM. 2778250b7ff9Sjosephb */ 2779250b7ff9Sjosephb static int 2780250b7ff9Sjosephb is_it_dism(hat_t *hat, caddr_t va) 2781250b7ff9Sjosephb { 2782250b7ff9Sjosephb struct seg *seg; 2783250b7ff9Sjosephb struct shm_data *shmd; 2784250b7ff9Sjosephb struct spt_data *sptd; 2785250b7ff9Sjosephb 2786250b7ff9Sjosephb seg = as_findseg(hat->hat_as, va, 0); 2787250b7ff9Sjosephb ASSERT(seg != NULL); 2788250b7ff9Sjosephb ASSERT(seg->s_base <= va); 2789250b7ff9Sjosephb shmd = (struct shm_data *)seg->s_data; 2790250b7ff9Sjosephb ASSERT(shmd != NULL); 2791250b7ff9Sjosephb sptd = (struct spt_data *)shmd->shm_sptseg->s_data; 2792250b7ff9Sjosephb ASSERT(sptd != NULL); 2793250b7ff9Sjosephb if (sptd->spt_flags & SHM_PAGEABLE) 2794250b7ff9Sjosephb return (1); 2795250b7ff9Sjosephb return (0); 2796250b7ff9Sjosephb } 2797250b7ff9Sjosephb 2798250b7ff9Sjosephb /* 2799250b7ff9Sjosephb * Simple implementation of ISM. hat_share() is similar to hat_memload_array(), 28007c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 2801250b7ff9Sjosephb * and protections to use for this hat. If we find a full properly aligned 2802250b7ff9Sjosephb * and sized pagetable, we will attempt to share the pagetable itself. 28037c478bd9Sstevel@tonic-gate */ 28047c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28057c478bd9Sstevel@tonic-gate int 28067c478bd9Sstevel@tonic-gate hat_share( 28077c478bd9Sstevel@tonic-gate hat_t *hat, 28087c478bd9Sstevel@tonic-gate caddr_t addr, 28097c478bd9Sstevel@tonic-gate hat_t *ism_hat, 28107c478bd9Sstevel@tonic-gate caddr_t src_addr, 28117c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 28127c478bd9Sstevel@tonic-gate uint_t ismszc) 28137c478bd9Sstevel@tonic-gate { 28147c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 28157c478bd9Sstevel@tonic-gate uintptr_t vaddr; 28167c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 28177c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 28187c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 28197c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 28207c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 28217c478bd9Sstevel@tonic-gate htable_t *ht; 28227c478bd9Sstevel@tonic-gate x86pte_t pte; 28237c478bd9Sstevel@tonic-gate page_t *pp; 28247c478bd9Sstevel@tonic-gate pfn_t pfn; 28257c478bd9Sstevel@tonic-gate level_t l; 28267c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 28277c478bd9Sstevel@tonic-gate uint_t prot; 2828250b7ff9Sjosephb int is_dism; 2829250b7ff9Sjosephb int flags; 28307c478bd9Sstevel@tonic-gate 28317c478bd9Sstevel@tonic-gate /* 28327c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 28337c478bd9Sstevel@tonic-gate */ 28347c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2835ae115bc7Smrj ASSERT(eaddr <= _userlimit); 28367c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 28377c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 28387c478bd9Sstevel@tonic-gate return (0); 28397c478bd9Sstevel@tonic-gate } 2840*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 28417c478bd9Sstevel@tonic-gate 28427c478bd9Sstevel@tonic-gate /* 28437c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 28447c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 28457c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 28467c478bd9Sstevel@tonic-gate */ 28477c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 28487c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 28497c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 2850250b7ff9Sjosephb is_dism = is_it_dism(hat, addr); 28517c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 28527c478bd9Sstevel@tonic-gate /* 28537c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 28547c478bd9Sstevel@tonic-gate */ 28557c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 28567c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 28577c478bd9Sstevel@tonic-gate break; 28587c478bd9Sstevel@tonic-gate 28597c478bd9Sstevel@tonic-gate /* 2860250b7ff9Sjosephb * First check to see if we already share the page table. 28617c478bd9Sstevel@tonic-gate */ 2862250b7ff9Sjosephb l = ism_ht->ht_level; 28637c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 2864250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 2865250b7ff9Sjosephb if (ht != NULL) { 2866250b7ff9Sjosephb if (ht->ht_flags & HTABLE_SHARED_PFN) 2867250b7ff9Sjosephb goto shared; 2868250b7ff9Sjosephb htable_release(ht); 2869250b7ff9Sjosephb goto not_shared; 28707c478bd9Sstevel@tonic-gate } 28717c478bd9Sstevel@tonic-gate 28727c478bd9Sstevel@tonic-gate /* 2873250b7ff9Sjosephb * Can't ever share top table. 28747c478bd9Sstevel@tonic-gate */ 2875250b7ff9Sjosephb if (l == mmu.max_level) 2876250b7ff9Sjosephb goto not_shared; 28777c478bd9Sstevel@tonic-gate 2878250b7ff9Sjosephb /* 2879250b7ff9Sjosephb * Avoid level mismatches later due to DISM faults. 2880250b7ff9Sjosephb */ 2881250b7ff9Sjosephb if (is_dism && l > 0) 2882250b7ff9Sjosephb goto not_shared; 28837c478bd9Sstevel@tonic-gate 2884250b7ff9Sjosephb /* 2885250b7ff9Sjosephb * addresses and lengths must align 2886250b7ff9Sjosephb * table must be fully populated 2887250b7ff9Sjosephb * no lower level page tables 2888250b7ff9Sjosephb */ 2889250b7ff9Sjosephb if (ism_addr != ism_ht->ht_vaddr || 2890250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l + 1)) != 0) 2891250b7ff9Sjosephb goto not_shared; 28927c478bd9Sstevel@tonic-gate 2893250b7ff9Sjosephb /* 2894250b7ff9Sjosephb * The range of address space must cover a full table. 2895250b7ff9Sjosephb */ 2896250b7ff9Sjosephb if (e_ism_addr - ism_addr < LEVEL_SIZE(1 + 1)) 2897250b7ff9Sjosephb goto not_shared; 28987c478bd9Sstevel@tonic-gate 2899250b7ff9Sjosephb /* 2900250b7ff9Sjosephb * All entries in the ISM page table must be leaf PTEs. 2901250b7ff9Sjosephb */ 2902250b7ff9Sjosephb if (l > 0) { 2903250b7ff9Sjosephb int e; 29047c478bd9Sstevel@tonic-gate 2905250b7ff9Sjosephb /* 2906250b7ff9Sjosephb * We know the 0th is from htable_walk() above. 2907250b7ff9Sjosephb */ 2908250b7ff9Sjosephb for (e = 1; e < HTABLE_NUM_PTES(ism_ht); ++e) { 2909250b7ff9Sjosephb x86pte_t pte; 2910250b7ff9Sjosephb pte = x86pte_get(ism_ht, e); 2911250b7ff9Sjosephb if (!PTE_ISPAGE(pte, l)) 2912250b7ff9Sjosephb goto not_shared; 29137c478bd9Sstevel@tonic-gate } 29147c478bd9Sstevel@tonic-gate } 29157c478bd9Sstevel@tonic-gate 2916250b7ff9Sjosephb /* 2917250b7ff9Sjosephb * share the page table 2918250b7ff9Sjosephb */ 2919250b7ff9Sjosephb ht = htable_create(hat, vaddr, l, ism_ht); 2920250b7ff9Sjosephb shared: 2921250b7ff9Sjosephb ASSERT(ht->ht_flags & HTABLE_SHARED_PFN); 2922250b7ff9Sjosephb ASSERT(ht->ht_shares == ism_ht); 2923250b7ff9Sjosephb hat->hat_ism_pgcnt += 2924250b7ff9Sjosephb (ism_ht->ht_valid_cnt - ht->ht_valid_cnt) << 2925250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 2926250b7ff9Sjosephb ht->ht_valid_cnt = ism_ht->ht_valid_cnt; 2927250b7ff9Sjosephb htable_release(ht); 2928250b7ff9Sjosephb ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(l + 1); 2929250b7ff9Sjosephb htable_release(ism_ht); 2930250b7ff9Sjosephb ism_ht = NULL; 2931250b7ff9Sjosephb continue; 2932250b7ff9Sjosephb 2933250b7ff9Sjosephb not_shared: 29347c478bd9Sstevel@tonic-gate /* 29357c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 29367c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 2937250b7ff9Sjosephb * Figure out what level size mappings to use; 2938250b7ff9Sjosephb */ 2939250b7ff9Sjosephb for (l = ism_ht->ht_level; l > 0; --l) { 2940250b7ff9Sjosephb if (LEVEL_SIZE(l) <= eaddr - vaddr && 2941250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l)) == 0) 2942250b7ff9Sjosephb break; 2943250b7ff9Sjosephb } 2944250b7ff9Sjosephb 2945250b7ff9Sjosephb /* 29467c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 2947250b7ff9Sjosephb * be careful to truncate it if needed. 29487c478bd9Sstevel@tonic-gate */ 29497c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 29507c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 29517c478bd9Sstevel@tonic-gate } else { 29527c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 29537c478bd9Sstevel@tonic-gate l = 0; 29547c478bd9Sstevel@tonic-gate } 29557c478bd9Sstevel@tonic-gate 29567c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 29577c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 29587c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 29597c478bd9Sstevel@tonic-gate /* 29607c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 29617c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 29627c478bd9Sstevel@tonic-gate */ 29637c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 29647c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 29657c478bd9Sstevel@tonic-gate 29667c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 29677c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 29687c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 29697c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 29707c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 29717c478bd9Sstevel@tonic-gate 2972250b7ff9Sjosephb flags = HAT_LOAD; 2973250b7ff9Sjosephb if (!is_dism) 2974250b7ff9Sjosephb flags |= HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST; 2975250b7ff9Sjosephb while (hati_load_common(hat, vaddr, pp, prot, flags, 2976ae115bc7Smrj l, pfn) != 0) { 2977ae115bc7Smrj if (l == 0) 2978ae115bc7Smrj panic("hati_load_common() failure"); 2979ae115bc7Smrj --l; 2980ae115bc7Smrj } 29817c478bd9Sstevel@tonic-gate 29827c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 29837c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 29847c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 29857c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 29867c478bd9Sstevel@tonic-gate } 29877c478bd9Sstevel@tonic-gate } 29887c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 29897c478bd9Sstevel@tonic-gate htable_release(ism_ht); 2990*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 29917c478bd9Sstevel@tonic-gate return (0); 29927c478bd9Sstevel@tonic-gate } 29937c478bd9Sstevel@tonic-gate 29947c478bd9Sstevel@tonic-gate 29957c478bd9Sstevel@tonic-gate /* 29967c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 29977c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 29987c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 29997c478bd9Sstevel@tonic-gate */ 30007c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30017c478bd9Sstevel@tonic-gate void 30027c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 30037c478bd9Sstevel@tonic-gate { 30047173d045Sjosephb uint64_t vaddr = (uintptr_t)addr; 30057c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 30067c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 30077c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 3008250b7ff9Sjosephb int flags = HAT_UNLOAD_UNMAP; 3009250b7ff9Sjosephb level_t l; 30107c478bd9Sstevel@tonic-gate 30117c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 3012ae115bc7Smrj ASSERT(eaddr <= _userlimit); 30137c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 30147c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 3015*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 30167c478bd9Sstevel@tonic-gate 30177c478bd9Sstevel@tonic-gate /* 30187c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 30197c478bd9Sstevel@tonic-gate * 3020ae115bc7Smrj * Note that it's ok to delay the TLB shootdown till the entire range is 30217c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 3022ae115bc7Smrj * pagetable page, its hat_tlb_inval() will do a global TLB invalidate. 30237c478bd9Sstevel@tonic-gate */ 3024250b7ff9Sjosephb l = mmu.max_page_level; 3025250b7ff9Sjosephb if (l == mmu.max_level) 3026250b7ff9Sjosephb --l; 3027250b7ff9Sjosephb for (; l >= 0; --l) { 3028250b7ff9Sjosephb for (vaddr = (uintptr_t)addr; vaddr < eaddr; 3029250b7ff9Sjosephb vaddr = (vaddr & LEVEL_MASK(l + 1)) + LEVEL_SIZE(l + 1)) { 3030250b7ff9Sjosephb ASSERT(!IN_VA_HOLE(vaddr)); 3031250b7ff9Sjosephb /* 3032250b7ff9Sjosephb * find a pagetable that maps the current address 3033250b7ff9Sjosephb */ 3034250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 3035250b7ff9Sjosephb if (ht == NULL) 3036250b7ff9Sjosephb continue; 30377c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 30387c478bd9Sstevel@tonic-gate /* 3039250b7ff9Sjosephb * clear page count, set valid_cnt to 0, 3040250b7ff9Sjosephb * let htable_release() finish the job 30417c478bd9Sstevel@tonic-gate */ 3042250b7ff9Sjosephb hat->hat_ism_pgcnt -= ht->ht_valid_cnt << 3043250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 30447c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 30457c478bd9Sstevel@tonic-gate need_demaps = 1; 30467c478bd9Sstevel@tonic-gate } 30477c478bd9Sstevel@tonic-gate htable_release(ht); 30487c478bd9Sstevel@tonic-gate } 30497c478bd9Sstevel@tonic-gate } 30507c478bd9Sstevel@tonic-gate 30517c478bd9Sstevel@tonic-gate /* 30527c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 30537c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 30547c478bd9Sstevel@tonic-gate */ 30557c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 3056ae115bc7Smrj hat_tlb_inval(hat, DEMAP_ALL_ADDR); 30577c478bd9Sstevel@tonic-gate 30587c478bd9Sstevel@tonic-gate /* 30597c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 30607c478bd9Sstevel@tonic-gate * couldn't share pagetables. 30617c478bd9Sstevel@tonic-gate */ 3062250b7ff9Sjosephb if (!is_it_dism(hat, addr)) 3063250b7ff9Sjosephb flags |= HAT_UNLOAD_UNLOCK; 3064250b7ff9Sjosephb hat_unload(hat, addr, len, flags); 3065*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 30667c478bd9Sstevel@tonic-gate } 30677c478bd9Sstevel@tonic-gate 30687c478bd9Sstevel@tonic-gate 30697c478bd9Sstevel@tonic-gate /* 30707c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 30717c478bd9Sstevel@tonic-gate */ 30727c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30737c478bd9Sstevel@tonic-gate void 30747c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 30757c478bd9Sstevel@tonic-gate { 30767c478bd9Sstevel@tonic-gate } 30777c478bd9Sstevel@tonic-gate 30787c478bd9Sstevel@tonic-gate 30797c478bd9Sstevel@tonic-gate /* 30807c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 30817c478bd9Sstevel@tonic-gate * Mostly stolem from hat_pagesync() 30827c478bd9Sstevel@tonic-gate */ 30837c478bd9Sstevel@tonic-gate static void 30847c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 30857c478bd9Sstevel@tonic-gate { 30867c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 30877c478bd9Sstevel@tonic-gate htable_t *ht; 30887c478bd9Sstevel@tonic-gate uint_t entry; 30897c478bd9Sstevel@tonic-gate x86pte_t old; 30907c478bd9Sstevel@tonic-gate x86pte_t new; 30917c478bd9Sstevel@tonic-gate uint_t pszc = 0; 30927c478bd9Sstevel@tonic-gate 3093*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 30947c478bd9Sstevel@tonic-gate next_size: 30957c478bd9Sstevel@tonic-gate /* 30967c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 30977c478bd9Sstevel@tonic-gate */ 30987c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 30997c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 31007c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 31017c478bd9Sstevel@tonic-gate continue; 31027c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 31037c478bd9Sstevel@tonic-gate 31047c478bd9Sstevel@tonic-gate for (;;) { 31057c478bd9Sstevel@tonic-gate /* 31067c478bd9Sstevel@tonic-gate * Is this mapping of interest? 31077c478bd9Sstevel@tonic-gate */ 31087c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 31097c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 31107c478bd9Sstevel@tonic-gate break; 31117c478bd9Sstevel@tonic-gate 31127c478bd9Sstevel@tonic-gate /* 31137c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 31147c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 31157c478bd9Sstevel@tonic-gate */ 31167c478bd9Sstevel@tonic-gate new = old; 31177c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 31187c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 31197c478bd9Sstevel@tonic-gate if (old != 0) 31207c478bd9Sstevel@tonic-gate continue; 31217c478bd9Sstevel@tonic-gate 31227c478bd9Sstevel@tonic-gate break; 31237c478bd9Sstevel@tonic-gate } 31247c478bd9Sstevel@tonic-gate } 31257c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 31267c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 31277c478bd9Sstevel@tonic-gate page_t *tpp; 31287c478bd9Sstevel@tonic-gate pszc++; 31297c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 31307c478bd9Sstevel@tonic-gate if (pp != tpp) { 31317c478bd9Sstevel@tonic-gate pp = tpp; 31327c478bd9Sstevel@tonic-gate goto next_size; 31337c478bd9Sstevel@tonic-gate } 31347c478bd9Sstevel@tonic-gate } 3135*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 31367c478bd9Sstevel@tonic-gate } 31377c478bd9Sstevel@tonic-gate 31387c478bd9Sstevel@tonic-gate /* 31397c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 31407c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 31417c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 31427c478bd9Sstevel@tonic-gate */ 31437c478bd9Sstevel@tonic-gate void 31447c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 31457c478bd9Sstevel@tonic-gate { 31467c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 31477c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 31487c478bd9Sstevel@tonic-gate page_t **listp; 3149d9615970Sqiao int noshuffle; 3150d9615970Sqiao 3151d9615970Sqiao noshuffle = flag & P_NSH; 3152d9615970Sqiao flag &= ~P_NSH; 31537c478bd9Sstevel@tonic-gate 31547c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 31557c478bd9Sstevel@tonic-gate return; 31567c478bd9Sstevel@tonic-gate 3157d9615970Sqiao if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp) && 3158d9615970Sqiao !noshuffle) { 31597c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 31607c478bd9Sstevel@tonic-gate mutex_enter(vphm); 31617c478bd9Sstevel@tonic-gate } 31627c478bd9Sstevel@tonic-gate 31637c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 31647c478bd9Sstevel@tonic-gate 31657c478bd9Sstevel@tonic-gate if (vphm != NULL) { 31667c478bd9Sstevel@tonic-gate 31677c478bd9Sstevel@tonic-gate /* 31687c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 31697c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 31707c478bd9Sstevel@tonic-gate * pp is the only page on the list. 31717c478bd9Sstevel@tonic-gate */ 31727c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 31737c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 31747c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 31757c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 31767c478bd9Sstevel@tonic-gate else 31777c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 31787c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 31797c478bd9Sstevel@tonic-gate } 31807c478bd9Sstevel@tonic-gate mutex_exit(vphm); 31817c478bd9Sstevel@tonic-gate } 31827c478bd9Sstevel@tonic-gate } 31837c478bd9Sstevel@tonic-gate 31847c478bd9Sstevel@tonic-gate void 31857c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 31867c478bd9Sstevel@tonic-gate { 31877c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 31887c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 31897c478bd9Sstevel@tonic-gate 31907c478bd9Sstevel@tonic-gate /* 3191a71e32b6Sstans * Caller is expected to hold page's io lock for VMODSORT to work 3192a71e32b6Sstans * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod 3193a71e32b6Sstans * bit is cleared. 3194a71e32b6Sstans * We don't have assert to avoid tripping some existing third party 3195a71e32b6Sstans * code. The dirty page is moved back to top of the v_page list 3196a71e32b6Sstans * after IO is done in pvn_write_done(). 31977c478bd9Sstevel@tonic-gate */ 31987c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 31997c478bd9Sstevel@tonic-gate 3200a71e32b6Sstans if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 32017c478bd9Sstevel@tonic-gate 32027c478bd9Sstevel@tonic-gate /* 32037c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 32047c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 32057c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 32067c478bd9Sstevel@tonic-gate * to this page. 32077c478bd9Sstevel@tonic-gate */ 32087c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 32097c478bd9Sstevel@tonic-gate } 32107c478bd9Sstevel@tonic-gate } 32117c478bd9Sstevel@tonic-gate 32127c478bd9Sstevel@tonic-gate /* 32137c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 32147c478bd9Sstevel@tonic-gate * and non zero if enabled. If flag specifes multiple attributs 32157c478bd9Sstevel@tonic-gate * then returns 0 if ALL atriibutes are disabled. This is an advisory 32167c478bd9Sstevel@tonic-gate * call. 32177c478bd9Sstevel@tonic-gate */ 32187c478bd9Sstevel@tonic-gate uint_t 32197c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 32207c478bd9Sstevel@tonic-gate { 32217c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 32227c478bd9Sstevel@tonic-gate } 32237c478bd9Sstevel@tonic-gate 32247c478bd9Sstevel@tonic-gate 32257c478bd9Sstevel@tonic-gate /* 32267c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 32277c478bd9Sstevel@tonic-gate */ 32287c478bd9Sstevel@tonic-gate hment_t * 32297c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 32307c478bd9Sstevel@tonic-gate { 32317c478bd9Sstevel@tonic-gate x86pte_t old_pte; 32327c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 32337c478bd9Sstevel@tonic-gate hment_t *hm; 32347c478bd9Sstevel@tonic-gate 32357c478bd9Sstevel@tonic-gate /* 32367c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 32377c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 32387c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 32397c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 32407c478bd9Sstevel@tonic-gate */ 32417c478bd9Sstevel@tonic-gate htable_acquire(ht); 32427c478bd9Sstevel@tonic-gate 32437c478bd9Sstevel@tonic-gate /* 32447c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 32457c478bd9Sstevel@tonic-gate */ 3246ae115bc7Smrj old_pte = x86pte_inval(ht, entry, 0, NULL); 3247aa2ed9e5Sjosephb if (PTE2PFN(old_pte, ht->ht_level) != pfn) { 3248ae115bc7Smrj panic("x86pte_inval() failure found PTE = " FMT_PTE 3249aa2ed9e5Sjosephb " pfn being unmapped is %lx ht=0x%lx entry=0x%x", 3250aa2ed9e5Sjosephb old_pte, pfn, (uintptr_t)ht, entry); 3251aa2ed9e5Sjosephb } 32527c478bd9Sstevel@tonic-gate 32537c478bd9Sstevel@tonic-gate /* 32547c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 32557c478bd9Sstevel@tonic-gate */ 32567c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 32577c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 32587c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 32597c478bd9Sstevel@tonic-gate 32607c478bd9Sstevel@tonic-gate /* 32617c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 32627c478bd9Sstevel@tonic-gate */ 3263ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC) 32647c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 32657c478bd9Sstevel@tonic-gate 32667c478bd9Sstevel@tonic-gate /* 32677c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 32687c478bd9Sstevel@tonic-gate */ 32697c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 32707c478bd9Sstevel@tonic-gate 32717c478bd9Sstevel@tonic-gate /* 32727c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 32737c478bd9Sstevel@tonic-gate * hment and htable. 32747c478bd9Sstevel@tonic-gate */ 32757c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 32767c478bd9Sstevel@tonic-gate htable_release(ht); 32777c478bd9Sstevel@tonic-gate return (hm); 32787c478bd9Sstevel@tonic-gate } 32797c478bd9Sstevel@tonic-gate 3280a5652762Spraks extern int vpm_enable; 32817c478bd9Sstevel@tonic-gate /* 32827c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 32837c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 32847c478bd9Sstevel@tonic-gate * 32857c478bd9Sstevel@tonic-gate * The forceflags are unused. 32867c478bd9Sstevel@tonic-gate */ 32877c478bd9Sstevel@tonic-gate 32887c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 32897c478bd9Sstevel@tonic-gate static int 32907c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 32917c478bd9Sstevel@tonic-gate { 32927c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 32937c478bd9Sstevel@tonic-gate hment_t *hm; 32947c478bd9Sstevel@tonic-gate hment_t *prev; 32957c478bd9Sstevel@tonic-gate htable_t *ht; 32967c478bd9Sstevel@tonic-gate uint_t entry; 32977c478bd9Sstevel@tonic-gate level_t level; 32987c478bd9Sstevel@tonic-gate 3299*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 3300a5652762Spraks #if defined(__amd64) 3301a5652762Spraks /* 3302a5652762Spraks * clear the vpm ref. 3303a5652762Spraks */ 3304a5652762Spraks if (vpm_enable) { 3305a5652762Spraks pp->p_vpmref = 0; 3306a5652762Spraks } 3307a5652762Spraks #endif 33087c478bd9Sstevel@tonic-gate /* 33097c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 33107c478bd9Sstevel@tonic-gate */ 33117c478bd9Sstevel@tonic-gate next_size: 33127c478bd9Sstevel@tonic-gate for (;;) { 33137c478bd9Sstevel@tonic-gate 33147c478bd9Sstevel@tonic-gate /* 33157c478bd9Sstevel@tonic-gate * Get a mapping list entry 33167c478bd9Sstevel@tonic-gate */ 33177c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 33187c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 33197c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 33207c478bd9Sstevel@tonic-gate if (hm == NULL) { 33217c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 33227c478bd9Sstevel@tonic-gate 33237c478bd9Sstevel@tonic-gate /* 33247c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 33257c478bd9Sstevel@tonic-gate */ 3326ae115bc7Smrj if (cur_pp->p_szc <= pg_szcd) { 3327*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 33287c478bd9Sstevel@tonic-gate return (0); 3329ae115bc7Smrj } 33307c478bd9Sstevel@tonic-gate 33317c478bd9Sstevel@tonic-gate /* 33327c478bd9Sstevel@tonic-gate * Else check the next larger page size. 33337c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 33347c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 33357c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 33367c478bd9Sstevel@tonic-gate * and return. 33377c478bd9Sstevel@tonic-gate */ 33387c478bd9Sstevel@tonic-gate ++pg_szcd; 33397c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 33407c478bd9Sstevel@tonic-gate goto next_size; 33417c478bd9Sstevel@tonic-gate } 33427c478bd9Sstevel@tonic-gate 33437c478bd9Sstevel@tonic-gate /* 33447c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 33457c478bd9Sstevel@tonic-gate */ 33467c478bd9Sstevel@tonic-gate level = ht->ht_level; 33477c478bd9Sstevel@tonic-gate if (level == pg_szcd) 33487c478bd9Sstevel@tonic-gate break; 33497c478bd9Sstevel@tonic-gate } 33507c478bd9Sstevel@tonic-gate 33517c478bd9Sstevel@tonic-gate /* 33527c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 33537c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 33547c478bd9Sstevel@tonic-gate */ 33557c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 33567c478bd9Sstevel@tonic-gate if (hm != NULL) 33577c478bd9Sstevel@tonic-gate hment_free(hm); 33587c478bd9Sstevel@tonic-gate } 33597c478bd9Sstevel@tonic-gate } 33607c478bd9Sstevel@tonic-gate 33617c478bd9Sstevel@tonic-gate int 33627c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 33637c478bd9Sstevel@tonic-gate { 33647c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 33657c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 33667c478bd9Sstevel@tonic-gate } 33677c478bd9Sstevel@tonic-gate 33687c478bd9Sstevel@tonic-gate /* 33697c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 33707c478bd9Sstevel@tonic-gate * page level that included pp. 33717c478bd9Sstevel@tonic-gate * 33727c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 33737c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 33747c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 33757c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 33767c478bd9Sstevel@tonic-gate * locked. 33777c478bd9Sstevel@tonic-gate * 33787c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 33797c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 33807c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 33817c478bd9Sstevel@tonic-gate * 33827c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 33837c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 33847c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 33857c478bd9Sstevel@tonic-gate */ 33867c478bd9Sstevel@tonic-gate void 33877c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 33887c478bd9Sstevel@tonic-gate { 33897c478bd9Sstevel@tonic-gate uint_t pszc; 33907c478bd9Sstevel@tonic-gate uint_t rszc; 33917c478bd9Sstevel@tonic-gate uint_t szc; 33927c478bd9Sstevel@tonic-gate page_t *rootpp; 33937c478bd9Sstevel@tonic-gate page_t *firstpp; 33947c478bd9Sstevel@tonic-gate page_t *lastpp; 33957c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 33967c478bd9Sstevel@tonic-gate 33977c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 33987c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 33997c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 34007c478bd9Sstevel@tonic-gate 34017c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 34027c478bd9Sstevel@tonic-gate return; 34037c478bd9Sstevel@tonic-gate 34047c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 34057c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 34067c478bd9Sstevel@tonic-gate 34077c478bd9Sstevel@tonic-gate /* 34087c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 34097c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 34107c478bd9Sstevel@tonic-gate * 34117c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 34127c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 34137c478bd9Sstevel@tonic-gate * 34147c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 34157c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 34167c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 34177c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 34187c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 34197c478bd9Sstevel@tonic-gate * page). 34207c478bd9Sstevel@tonic-gate */ 34217c478bd9Sstevel@tonic-gate again: 34227c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 34237c478bd9Sstevel@tonic-gate if (pszc == 0) 34247c478bd9Sstevel@tonic-gate return; 34257c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 34267c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 34277c478bd9Sstevel@tonic-gate /* 34287c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 34297c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 34307c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 34317c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 34327c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 34337c478bd9Sstevel@tonic-gate */ 34347c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 34357c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 34367c478bd9Sstevel@tonic-gate if (rszc > pszc) { 34377c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 34387c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 34397c478bd9Sstevel@tonic-gate 34407c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 34417c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 34427c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 34437c478bd9Sstevel@tonic-gate } 34447c478bd9Sstevel@tonic-gate goto again; 34457c478bd9Sstevel@tonic-gate } 34467c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 34477c478bd9Sstevel@tonic-gate 34487c478bd9Sstevel@tonic-gate /* 34497c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 34507c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 34517c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 34527c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 34537c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 34547c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 34557c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 34567c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 34577c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 34587c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 34597c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 34607c478bd9Sstevel@tonic-gate * 34617c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 34627c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 34637c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 34647c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 34657c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 34667c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 34677c478bd9Sstevel@tonic-gate * 34687c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 34697c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 34707c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 34717c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 34727c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 34737c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 34747c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 34757c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 34767c478bd9Sstevel@tonic-gate */ 34777c478bd9Sstevel@tonic-gate szc = pszc; 34787c478bd9Sstevel@tonic-gate while (szc > 1) { 34797c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 34807c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 34817c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 34827c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 34837c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 34847c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 34857c478bd9Sstevel@tonic-gate szc--; 34867c478bd9Sstevel@tonic-gate continue; 34877c478bd9Sstevel@tonic-gate } 34887c478bd9Sstevel@tonic-gate firstpp += pgcnt; 34897c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 34907c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 34917c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 34927c478bd9Sstevel@tonic-gate lastpp--; 34937c478bd9Sstevel@tonic-gate } 34947c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 34957c478bd9Sstevel@tonic-gate szc--; 34967c478bd9Sstevel@tonic-gate } 34977c478bd9Sstevel@tonic-gate 34987c478bd9Sstevel@tonic-gate /* 34997c478bd9Sstevel@tonic-gate * Second loop description: 35007c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 35017c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 35027c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 35037c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 35047c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 35057c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 35067c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 35077c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 35087c478bd9Sstevel@tonic-gate * skip to the next iteration. 35097c478bd9Sstevel@tonic-gate * 35107c478bd9Sstevel@tonic-gate */ 35117c478bd9Sstevel@tonic-gate szc = 0; 35127c478bd9Sstevel@tonic-gate while (szc < pszc) { 35137c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 35147c478bd9Sstevel@tonic-gate if (szc == 0) { 35157c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 35167c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 35177c478bd9Sstevel@tonic-gate } else { 35187c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 35197c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 35207c478bd9Sstevel@tonic-gate szc++; 35217c478bd9Sstevel@tonic-gate continue; 35227c478bd9Sstevel@tonic-gate } 35237c478bd9Sstevel@tonic-gate lastpp--; 35247c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 35257c478bd9Sstevel@tonic-gate } 35267c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 35277c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 35287c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 35297c478bd9Sstevel@tonic-gate lastpp--; 35307c478bd9Sstevel@tonic-gate } 35317c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 35327c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 35337c478bd9Sstevel@tonic-gate break; 35347c478bd9Sstevel@tonic-gate szc++; 35357c478bd9Sstevel@tonic-gate } 35367c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35377c478bd9Sstevel@tonic-gate } 35387c478bd9Sstevel@tonic-gate 35397c478bd9Sstevel@tonic-gate /* 35407c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 35417c478bd9Sstevel@tonic-gate * returns attributes of page 35427c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 35437c478bd9Sstevel@tonic-gate * 35447c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 35457c478bd9Sstevel@tonic-gate * 35467c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 35477c478bd9Sstevel@tonic-gate * 35487c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 35497c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 35507c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 35517c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 35527c478bd9Sstevel@tonic-gate */ 35537c478bd9Sstevel@tonic-gate uint_t 35547c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 35557c478bd9Sstevel@tonic-gate { 35567c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 35577c478bd9Sstevel@tonic-gate htable_t *ht; 35587c478bd9Sstevel@tonic-gate uint_t entry; 35597c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 35607c478bd9Sstevel@tonic-gate x86pte_t new; 35617c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 35627c478bd9Sstevel@tonic-gate extern ulong_t po_share; 35637c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 35647c478bd9Sstevel@tonic-gate uint_t pszc = 0; 35657c478bd9Sstevel@tonic-gate 35667c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 35677c478bd9Sstevel@tonic-gate 35687c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 35697c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 35707c478bd9Sstevel@tonic-gate 35717c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 35727c478bd9Sstevel@tonic-gate 35737c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 35747c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 35757c478bd9Sstevel@tonic-gate 35767c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 35777c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 35787c478bd9Sstevel@tonic-gate 35797c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 35807c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 35817c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 35827c478bd9Sstevel@tonic-gate PP_SETREF(pp); 35837c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 35847c478bd9Sstevel@tonic-gate } 35857c478bd9Sstevel@tonic-gate } 35867c478bd9Sstevel@tonic-gate 3587*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 35887c478bd9Sstevel@tonic-gate next_size: 35897c478bd9Sstevel@tonic-gate /* 35907c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 35917c478bd9Sstevel@tonic-gate */ 35927c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 35937c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 35947c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 35957c478bd9Sstevel@tonic-gate continue; 35967c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 35977c478bd9Sstevel@tonic-gate try_again: 35987c478bd9Sstevel@tonic-gate 35997c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 36007c478bd9Sstevel@tonic-gate 36017c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 36027c478bd9Sstevel@tonic-gate continue; 36037c478bd9Sstevel@tonic-gate 36047c478bd9Sstevel@tonic-gate save_old = old; 36057c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 36067c478bd9Sstevel@tonic-gate 36077c478bd9Sstevel@tonic-gate /* 36087c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 36097c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 36107c478bd9Sstevel@tonic-gate */ 36117c478bd9Sstevel@tonic-gate new = old; 36127c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 36137c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 36147c478bd9Sstevel@tonic-gate if (old != 0) 36157c478bd9Sstevel@tonic-gate goto try_again; 36167c478bd9Sstevel@tonic-gate 36177c478bd9Sstevel@tonic-gate old = save_old; 36187c478bd9Sstevel@tonic-gate } 36197c478bd9Sstevel@tonic-gate 36207c478bd9Sstevel@tonic-gate /* 36217c478bd9Sstevel@tonic-gate * Sync the PTE 36227c478bd9Sstevel@tonic-gate */ 3623ae115bc7Smrj if (!(flags & HAT_SYNC_ZERORM) && 3624ae115bc7Smrj PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC) 36257c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 36267c478bd9Sstevel@tonic-gate 36277c478bd9Sstevel@tonic-gate /* 36287c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 36297c478bd9Sstevel@tonic-gate */ 36307c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 36317c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 36327c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3633ae115bc7Smrj goto done; 36347c478bd9Sstevel@tonic-gate } 36357c478bd9Sstevel@tonic-gate } 36367c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 36377c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 36387c478bd9Sstevel@tonic-gate page_t *tpp; 36397c478bd9Sstevel@tonic-gate pszc++; 36407c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 36417c478bd9Sstevel@tonic-gate if (pp != tpp) { 36427c478bd9Sstevel@tonic-gate pp = tpp; 36437c478bd9Sstevel@tonic-gate goto next_size; 36447c478bd9Sstevel@tonic-gate } 36457c478bd9Sstevel@tonic-gate } 3646ae115bc7Smrj done: 3647*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 36487c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 36497c478bd9Sstevel@tonic-gate } 36507c478bd9Sstevel@tonic-gate 36517c478bd9Sstevel@tonic-gate /* 36527c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 36537c478bd9Sstevel@tonic-gate * there are no mappings to the page. 36547c478bd9Sstevel@tonic-gate */ 36557c478bd9Sstevel@tonic-gate ulong_t 36567c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 36577c478bd9Sstevel@tonic-gate { 36587c478bd9Sstevel@tonic-gate uint_t cnt; 36597c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3660a5652762Spraks #if defined(__amd64) 3661a5652762Spraks if (vpm_enable && pp->p_vpmref) { 3662a5652762Spraks cnt += 1; 3663a5652762Spraks } 3664a5652762Spraks #endif 36657c478bd9Sstevel@tonic-gate return (cnt); 36667c478bd9Sstevel@tonic-gate } 36677c478bd9Sstevel@tonic-gate 366805d3dc4bSpaulsan /* 366905d3dc4bSpaulsan * Return 1 the number of mappings exceeds sh_thresh. Return 0 367005d3dc4bSpaulsan * otherwise. 367105d3dc4bSpaulsan */ 367205d3dc4bSpaulsan int 367305d3dc4bSpaulsan hat_page_checkshare(page_t *pp, ulong_t sh_thresh) 367405d3dc4bSpaulsan { 367505d3dc4bSpaulsan return (hat_page_getshare(pp) > sh_thresh); 367605d3dc4bSpaulsan } 367705d3dc4bSpaulsan 36787c478bd9Sstevel@tonic-gate /* 36797c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 36807c478bd9Sstevel@tonic-gate */ 36817c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 36827c478bd9Sstevel@tonic-gate faultcode_t 36837c478bd9Sstevel@tonic-gate hat_softlock( 36847c478bd9Sstevel@tonic-gate hat_t *hat, 36857c478bd9Sstevel@tonic-gate caddr_t addr, 36867c478bd9Sstevel@tonic-gate size_t *len, 36877c478bd9Sstevel@tonic-gate struct page **page_array, 36887c478bd9Sstevel@tonic-gate uint_t flags) 36897c478bd9Sstevel@tonic-gate { 36907c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 36917c478bd9Sstevel@tonic-gate } 36927c478bd9Sstevel@tonic-gate 36937c478bd9Sstevel@tonic-gate 36947c478bd9Sstevel@tonic-gate 36957c478bd9Sstevel@tonic-gate /* 36967c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 36977c478bd9Sstevel@tonic-gate */ 36987c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 36997c478bd9Sstevel@tonic-gate int 37007c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 37017c478bd9Sstevel@tonic-gate { 37027c478bd9Sstevel@tonic-gate switch (feature) { 37037c478bd9Sstevel@tonic-gate 37047c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 37057c478bd9Sstevel@tonic-gate return (1); 37067c478bd9Sstevel@tonic-gate 37077c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 37087c478bd9Sstevel@tonic-gate return (0); 37097c478bd9Sstevel@tonic-gate 37107c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 37117c478bd9Sstevel@tonic-gate return (1); 37127c478bd9Sstevel@tonic-gate 371305d3dc4bSpaulsan case HAT_SHARED_REGIONS: 371405d3dc4bSpaulsan return (0); 371505d3dc4bSpaulsan 37167c478bd9Sstevel@tonic-gate default: 37177c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 37187c478bd9Sstevel@tonic-gate } 37197c478bd9Sstevel@tonic-gate return (0); 37207c478bd9Sstevel@tonic-gate } 37217c478bd9Sstevel@tonic-gate 37227c478bd9Sstevel@tonic-gate /* 37237c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 37247c478bd9Sstevel@tonic-gate */ 37257c478bd9Sstevel@tonic-gate void 37267c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 37277c478bd9Sstevel@tonic-gate { 37287c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 3729*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 37307c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 3731*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 37327c478bd9Sstevel@tonic-gate } 37337c478bd9Sstevel@tonic-gate 37347c478bd9Sstevel@tonic-gate /* 37357c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 37367c478bd9Sstevel@tonic-gate */ 37377c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37387c478bd9Sstevel@tonic-gate void 37397c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 37407c478bd9Sstevel@tonic-gate { 3741*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 37427c478bd9Sstevel@tonic-gate kpreempt_disable(); 37437c478bd9Sstevel@tonic-gate 37447c478bd9Sstevel@tonic-gate hat_switch(hat); 37457c478bd9Sstevel@tonic-gate 37467c478bd9Sstevel@tonic-gate kpreempt_enable(); 3747*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 37487c478bd9Sstevel@tonic-gate } 37497c478bd9Sstevel@tonic-gate 37507c478bd9Sstevel@tonic-gate /* 37517c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 37527c478bd9Sstevel@tonic-gate * 37537c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 37547c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 37557c478bd9Sstevel@tonic-gate * 37567c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 37577c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 37587c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 37597c478bd9Sstevel@tonic-gate */ 3760ae115bc7Smrj hat_mempte_t 3761ae115bc7Smrj hat_mempte_setup(caddr_t addr) 37627c478bd9Sstevel@tonic-gate { 37637c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 37647c478bd9Sstevel@tonic-gate htable_t *ht; 37657c478bd9Sstevel@tonic-gate uint_t entry; 37667c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3767ae115bc7Smrj hat_mempte_t p; 37687c478bd9Sstevel@tonic-gate 37697c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 37707c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3771aac11643Sjosephb ++curthread->t_hatdepth; 37727c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 37737c478bd9Sstevel@tonic-gate if (ht == NULL) { 37747c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 37757c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 37767c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 37777c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 37787c478bd9Sstevel@tonic-gate } 37797c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 37807c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 37817c478bd9Sstevel@tonic-gate "ht=%p, entry=%d, pte=" FMT_PTE, ht, entry, oldpte); 37827c478bd9Sstevel@tonic-gate 37837c478bd9Sstevel@tonic-gate /* 37847c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 37857c478bd9Sstevel@tonic-gate */ 37867c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 37877c478bd9Sstevel@tonic-gate 37887c478bd9Sstevel@tonic-gate /* 3789ae115bc7Smrj * return the PTE physical address to the caller. 37907c478bd9Sstevel@tonic-gate */ 37917c478bd9Sstevel@tonic-gate htable_release(ht); 3792ae115bc7Smrj p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry); 3793aac11643Sjosephb --curthread->t_hatdepth; 3794ae115bc7Smrj return (p); 37957c478bd9Sstevel@tonic-gate } 37967c478bd9Sstevel@tonic-gate 37977c478bd9Sstevel@tonic-gate /* 37987c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 37997c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 38007c478bd9Sstevel@tonic-gate */ 3801ae115bc7Smrj /*ARGSUSED1*/ 38027c478bd9Sstevel@tonic-gate void 3803ae115bc7Smrj hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa) 38047c478bd9Sstevel@tonic-gate { 38057c478bd9Sstevel@tonic-gate htable_t *ht; 38067c478bd9Sstevel@tonic-gate 38077c478bd9Sstevel@tonic-gate /* 3808ae115bc7Smrj * invalidate any left over mapping and decrement the htable valid count 38097c478bd9Sstevel@tonic-gate */ 3810*843e1988Sjohnlev #ifdef __xpv 3811*843e1988Sjohnlev if (HYPERVISOR_update_va_mapping((uintptr_t)addr, 0, 3812*843e1988Sjohnlev UVMF_INVLPG | UVMF_LOCAL)) 3813*843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3814*843e1988Sjohnlev #else 3815ae115bc7Smrj { 3816ae115bc7Smrj x86pte_t *pteptr; 3817ae115bc7Smrj 3818ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3819ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3820ae115bc7Smrj if (mmu.pae_hat) 3821ae115bc7Smrj *pteptr = 0; 3822ae115bc7Smrj else 3823ae115bc7Smrj *(x86pte32_t *)pteptr = 0; 3824ae115bc7Smrj mmu_tlbflush_entry(addr); 3825ae115bc7Smrj x86pte_mapout(); 3826ae115bc7Smrj } 3827*843e1988Sjohnlev #endif 3828ae115bc7Smrj 38297c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 38307c478bd9Sstevel@tonic-gate if (ht == NULL) 38317c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 38327c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 38337c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 38347c478bd9Sstevel@tonic-gate htable_release(ht); 38357c478bd9Sstevel@tonic-gate } 38367c478bd9Sstevel@tonic-gate 38377c478bd9Sstevel@tonic-gate /* 38387c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 38397c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 38407c478bd9Sstevel@tonic-gate */ 38417c478bd9Sstevel@tonic-gate void 38427c478bd9Sstevel@tonic-gate hat_mempte_remap( 3843ae115bc7Smrj pfn_t pfn, 3844ae115bc7Smrj caddr_t addr, 3845ae115bc7Smrj hat_mempte_t pte_pa, 3846ae115bc7Smrj uint_t attr, 3847ae115bc7Smrj uint_t flags) 38487c478bd9Sstevel@tonic-gate { 38497c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 38507c478bd9Sstevel@tonic-gate x86pte_t pte; 38517c478bd9Sstevel@tonic-gate 38527c478bd9Sstevel@tonic-gate /* 38537c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 38547c478bd9Sstevel@tonic-gate * on this CPU. 38557c478bd9Sstevel@tonic-gate */ 38567c478bd9Sstevel@tonic-gate #ifdef DEBUG 38577c478bd9Sstevel@tonic-gate htable_t *ht; 38587c478bd9Sstevel@tonic-gate uint_t entry; 38597c478bd9Sstevel@tonic-gate 38607c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 38617c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 38627c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 38637c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 38647c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 38657c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3866ae115bc7Smrj ASSERT(ht->ht_pfn == mmu_btop(pte_pa)); 38677c478bd9Sstevel@tonic-gate htable_release(ht); 38687c478bd9Sstevel@tonic-gate #endif 3869*843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 38707c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3871*843e1988Sjohnlev #ifdef __xpv 3872*843e1988Sjohnlev if (HYPERVISOR_update_va_mapping(va, pte, UVMF_INVLPG | UVMF_LOCAL)) 3873*843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3874*843e1988Sjohnlev #else 3875ae115bc7Smrj { 3876ae115bc7Smrj x86pte_t *pteptr; 3877ae115bc7Smrj 3878ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3879ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3880ae115bc7Smrj if (mmu.pae_hat) 3881ae115bc7Smrj *(x86pte_t *)pteptr = pte; 3882ae115bc7Smrj else 3883ae115bc7Smrj *(x86pte32_t *)pteptr = (x86pte32_t)pte; 3884ae115bc7Smrj mmu_tlbflush_entry(addr); 3885ae115bc7Smrj x86pte_mapout(); 3886ae115bc7Smrj } 3887*843e1988Sjohnlev #endif 3888*843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 38897c478bd9Sstevel@tonic-gate } 38907c478bd9Sstevel@tonic-gate 38917c478bd9Sstevel@tonic-gate 38927c478bd9Sstevel@tonic-gate 38937c478bd9Sstevel@tonic-gate /* 38947c478bd9Sstevel@tonic-gate * Hat locking functions 38957c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 38967c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 38977c478bd9Sstevel@tonic-gate */ 38987c478bd9Sstevel@tonic-gate void 38997c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 39007c478bd9Sstevel@tonic-gate { 39017c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 39027c478bd9Sstevel@tonic-gate } 39037c478bd9Sstevel@tonic-gate 39047c478bd9Sstevel@tonic-gate void 39057c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 39067c478bd9Sstevel@tonic-gate { 39077c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 39087c478bd9Sstevel@tonic-gate } 39097c478bd9Sstevel@tonic-gate 39107c478bd9Sstevel@tonic-gate /* 3911ae115bc7Smrj * HAT part of cpu initialization. 39127c478bd9Sstevel@tonic-gate */ 39137c478bd9Sstevel@tonic-gate void 3914ae115bc7Smrj hat_cpu_online(struct cpu *cpup) 39157c478bd9Sstevel@tonic-gate { 3916ae115bc7Smrj if (cpup != CPU) { 3917ae115bc7Smrj x86pte_cpu_init(cpup); 3918ae115bc7Smrj hat_vlp_setup(cpup); 39197c478bd9Sstevel@tonic-gate } 3920ae115bc7Smrj CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 39217c478bd9Sstevel@tonic-gate } 39227c478bd9Sstevel@tonic-gate 39237c478bd9Sstevel@tonic-gate /* 3924ae115bc7Smrj * HAT part of cpu deletion. 3925ae115bc7Smrj * (currently, we only call this after the cpu is safely passivated.) 39267c478bd9Sstevel@tonic-gate */ 39277c478bd9Sstevel@tonic-gate void 3928ae115bc7Smrj hat_cpu_offline(struct cpu *cpup) 39297c478bd9Sstevel@tonic-gate { 3930ae115bc7Smrj ASSERT(cpup != CPU); 3931ae115bc7Smrj 3932ae115bc7Smrj CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id); 3933ae115bc7Smrj x86pte_cpu_fini(cpup); 3934ae115bc7Smrj hat_vlp_teardown(cpup); 39357c478bd9Sstevel@tonic-gate } 39367c478bd9Sstevel@tonic-gate 39377c478bd9Sstevel@tonic-gate /* 39387c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 39397c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 39407c478bd9Sstevel@tonic-gate */ 39417c478bd9Sstevel@tonic-gate void 39427c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 39437c478bd9Sstevel@tonic-gate { 39447c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 39457c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 39467c478bd9Sstevel@tonic-gate level_t level; 39477c478bd9Sstevel@tonic-gate uint_t entry; 39487c478bd9Sstevel@tonic-gate x86pte_t pte; 39497c478bd9Sstevel@tonic-gate 39507c478bd9Sstevel@tonic-gate /* 39517c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 3952ae115bc7Smrj * all virtual mappings under _userlimit. 39537c478bd9Sstevel@tonic-gate */ 39547c478bd9Sstevel@tonic-gate while (vaddr < high) { 39557c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 39567c478bd9Sstevel@tonic-gate if (ht == NULL) 39577c478bd9Sstevel@tonic-gate break; 39587c478bd9Sstevel@tonic-gate 39597c478bd9Sstevel@tonic-gate level = ht->ht_level; 39607c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 39617c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 39627c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 39637c478bd9Sstevel@tonic-gate 39647c478bd9Sstevel@tonic-gate /* 39657c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 39667c478bd9Sstevel@tonic-gate */ 3967ae115bc7Smrj (void) x86pte_inval(ht, entry, 0, NULL); 39687c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 39697c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 39707c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 39717c478bd9Sstevel@tonic-gate 39727c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 39737c478bd9Sstevel@tonic-gate } 39747c478bd9Sstevel@tonic-gate if (ht) 39757c478bd9Sstevel@tonic-gate htable_release(ht); 39767c478bd9Sstevel@tonic-gate } 39777c478bd9Sstevel@tonic-gate 39787c478bd9Sstevel@tonic-gate /* 39797c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 39807c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 39817c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 39827c478bd9Sstevel@tonic-gate * 39837c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 39847c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 39857c478bd9Sstevel@tonic-gate */ 39867c478bd9Sstevel@tonic-gate static x86pte_t 39877c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 39887c478bd9Sstevel@tonic-gate { 39897c478bd9Sstevel@tonic-gate page_t *pp; 39907c478bd9Sstevel@tonic-gate uint_t rm = 0; 39917c478bd9Sstevel@tonic-gate x86pte_t replaced; 39927c478bd9Sstevel@tonic-gate 3993ae115bc7Smrj if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC && 39947c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 39957c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 3996250b7ff9Sjosephb !PTE_GET(new, PT_MOD | PT_REF))) { 39977c478bd9Sstevel@tonic-gate 3998ae115bc7Smrj ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level))); 39997c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 40007c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 40017c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 40027c478bd9Sstevel@tonic-gate rm |= P_MOD; 40037c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 40047c478bd9Sstevel@tonic-gate rm |= P_REF; 40057c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 40067c478bd9Sstevel@tonic-gate } 40077c478bd9Sstevel@tonic-gate 40087c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 40097c478bd9Sstevel@tonic-gate if (replaced != expected) 40107c478bd9Sstevel@tonic-gate return (replaced); 40117c478bd9Sstevel@tonic-gate 40127c478bd9Sstevel@tonic-gate if (rm) { 40137c478bd9Sstevel@tonic-gate /* 40147c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 40157c478bd9Sstevel@tonic-gate */ 40167c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 40177c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 40187c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 40197c478bd9Sstevel@tonic-gate /* 40207c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 40217c478bd9Sstevel@tonic-gate * pszc below this mapping size 40227c478bd9Sstevel@tonic-gate * since large mapping existed after we 40237c478bd9Sstevel@tonic-gate * took mlist lock. 40247c478bd9Sstevel@tonic-gate */ 40257c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 40267c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 40277c478bd9Sstevel@tonic-gate ++pp; 40287c478bd9Sstevel@tonic-gate } 40297c478bd9Sstevel@tonic-gate } 40307c478bd9Sstevel@tonic-gate 40317c478bd9Sstevel@tonic-gate return (0); 40327c478bd9Sstevel@tonic-gate } 40337c478bd9Sstevel@tonic-gate 403405d3dc4bSpaulsan /* ARGSUSED */ 403505d3dc4bSpaulsan void 40367dacfc44Spaulsan hat_join_srd(struct hat *hat, vnode_t *evp) 403705d3dc4bSpaulsan { 403805d3dc4bSpaulsan } 403905d3dc4bSpaulsan 404005d3dc4bSpaulsan /* ARGSUSED */ 404105d3dc4bSpaulsan hat_region_cookie_t 40427dacfc44Spaulsan hat_join_region(struct hat *hat, 404305d3dc4bSpaulsan caddr_t r_saddr, 404405d3dc4bSpaulsan size_t r_size, 404505d3dc4bSpaulsan void *r_obj, 404605d3dc4bSpaulsan u_offset_t r_objoff, 404705d3dc4bSpaulsan uchar_t r_perm, 404805d3dc4bSpaulsan uchar_t r_pgszc, 404905d3dc4bSpaulsan hat_rgn_cb_func_t r_cb_function, 405005d3dc4bSpaulsan uint_t flags) 405105d3dc4bSpaulsan { 405205d3dc4bSpaulsan panic("No shared region support on x86"); 405305d3dc4bSpaulsan return (HAT_INVALID_REGION_COOKIE); 405405d3dc4bSpaulsan } 405505d3dc4bSpaulsan 405605d3dc4bSpaulsan /* ARGSUSED */ 405705d3dc4bSpaulsan void 40587dacfc44Spaulsan hat_leave_region(struct hat *hat, hat_region_cookie_t rcookie, uint_t flags) 405905d3dc4bSpaulsan { 406005d3dc4bSpaulsan panic("No shared region support on x86"); 406105d3dc4bSpaulsan } 406205d3dc4bSpaulsan 406305d3dc4bSpaulsan /* ARGSUSED */ 406405d3dc4bSpaulsan void 40657dacfc44Spaulsan hat_dup_region(struct hat *hat, hat_region_cookie_t rcookie) 406605d3dc4bSpaulsan { 406705d3dc4bSpaulsan panic("No shared region support on x86"); 406805d3dc4bSpaulsan } 406905d3dc4bSpaulsan 407005d3dc4bSpaulsan 40717c478bd9Sstevel@tonic-gate /* 40727c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 40737c478bd9Sstevel@tonic-gate * 40747c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 40757c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 40767c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 40777c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 40787c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 40797c478bd9Sstevel@tonic-gate * caches. 40807c478bd9Sstevel@tonic-gate * 40817c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 40827c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 40837c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 40847c478bd9Sstevel@tonic-gate * upper levels. 40857c478bd9Sstevel@tonic-gate */ 40867c478bd9Sstevel@tonic-gate 40877c478bd9Sstevel@tonic-gate /* 40887c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 40897c478bd9Sstevel@tonic-gate */ 40907c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 40917c478bd9Sstevel@tonic-gate caddr_t 40927c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 40937c478bd9Sstevel@tonic-gate { 40947c478bd9Sstevel@tonic-gate caddr_t vaddr; 40957c478bd9Sstevel@tonic-gate 40967c478bd9Sstevel@tonic-gate #ifdef DEBUG 40977c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 40987c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 40997c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41007c478bd9Sstevel@tonic-gate } 41017c478bd9Sstevel@tonic-gate 41027c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 41037c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 41047c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41057c478bd9Sstevel@tonic-gate } 41067c478bd9Sstevel@tonic-gate #endif 41077c478bd9Sstevel@tonic-gate 41087c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 41097c478bd9Sstevel@tonic-gate 41107c478bd9Sstevel@tonic-gate return (vaddr); 41117c478bd9Sstevel@tonic-gate } 41127c478bd9Sstevel@tonic-gate 41137c478bd9Sstevel@tonic-gate /* 41147c478bd9Sstevel@tonic-gate * Mapout a locked page. 41157c478bd9Sstevel@tonic-gate */ 41167c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 41177c478bd9Sstevel@tonic-gate void 41187c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 41197c478bd9Sstevel@tonic-gate { 41207c478bd9Sstevel@tonic-gate #ifdef DEBUG 41217c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 41227c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 41237c478bd9Sstevel@tonic-gate return; 41247c478bd9Sstevel@tonic-gate } 41257c478bd9Sstevel@tonic-gate 41267c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 41277c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 41287c478bd9Sstevel@tonic-gate return; 41297c478bd9Sstevel@tonic-gate } 41307c478bd9Sstevel@tonic-gate 41317c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 41327c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 41337c478bd9Sstevel@tonic-gate return; 41347c478bd9Sstevel@tonic-gate } 41357c478bd9Sstevel@tonic-gate #endif 41367c478bd9Sstevel@tonic-gate } 41377c478bd9Sstevel@tonic-gate 41387c478bd9Sstevel@tonic-gate /* 41397c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 41407c478bd9Sstevel@tonic-gate */ 41417c478bd9Sstevel@tonic-gate caddr_t 41427c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 41437c478bd9Sstevel@tonic-gate { 4144ae115bc7Smrj uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 41457c478bd9Sstevel@tonic-gate 41467c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 41477c478bd9Sstevel@tonic-gate } 41487c478bd9Sstevel@tonic-gate 41497c478bd9Sstevel@tonic-gate /* 41507c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 41517c478bd9Sstevel@tonic-gate */ 41527c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 41537c478bd9Sstevel@tonic-gate caddr_t 41547c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 41557c478bd9Sstevel@tonic-gate { 41567c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 41577c478bd9Sstevel@tonic-gate } 41587c478bd9Sstevel@tonic-gate 41597c478bd9Sstevel@tonic-gate /* 41607c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 41617c478bd9Sstevel@tonic-gate */ 41627c478bd9Sstevel@tonic-gate pfn_t 41637c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 41647c478bd9Sstevel@tonic-gate { 41657c478bd9Sstevel@tonic-gate pfn_t pfn; 41667c478bd9Sstevel@tonic-gate 41677c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 41687c478bd9Sstevel@tonic-gate 41697c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 41707c478bd9Sstevel@tonic-gate 41717c478bd9Sstevel@tonic-gate return (pfn); 41727c478bd9Sstevel@tonic-gate } 41737c478bd9Sstevel@tonic-gate 41747c478bd9Sstevel@tonic-gate 41757c478bd9Sstevel@tonic-gate /* 41767c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 41777c478bd9Sstevel@tonic-gate */ 41787c478bd9Sstevel@tonic-gate page_t * 41797c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 41807c478bd9Sstevel@tonic-gate { 41817c478bd9Sstevel@tonic-gate pfn_t pfn; 41827c478bd9Sstevel@tonic-gate 41837c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 41847c478bd9Sstevel@tonic-gate 41857c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 41867c478bd9Sstevel@tonic-gate 41877c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 41887c478bd9Sstevel@tonic-gate } 41897c478bd9Sstevel@tonic-gate 41907c478bd9Sstevel@tonic-gate /* 41917c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 41927c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 41937c478bd9Sstevel@tonic-gate */ 41947c478bd9Sstevel@tonic-gate int 41957c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 41967c478bd9Sstevel@tonic-gate { 41977c478bd9Sstevel@tonic-gate panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", hat, vaddr); 41987c478bd9Sstevel@tonic-gate 41997c478bd9Sstevel@tonic-gate return (0); 42007c478bd9Sstevel@tonic-gate } 42017c478bd9Sstevel@tonic-gate 42027c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42037c478bd9Sstevel@tonic-gate void 42047c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 42057c478bd9Sstevel@tonic-gate {} 42067c478bd9Sstevel@tonic-gate 42077c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42087c478bd9Sstevel@tonic-gate void 42097c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 42107c478bd9Sstevel@tonic-gate {} 4211*843e1988Sjohnlev 4212*843e1988Sjohnlev #ifdef __xpv 4213*843e1988Sjohnlev /* 4214*843e1988Sjohnlev * There are specific Hypervisor calls to establish and remove mappings 4215*843e1988Sjohnlev * to grant table references and the privcmd driver. We have to ensure 4216*843e1988Sjohnlev * that a page table actually exists. 4217*843e1988Sjohnlev */ 4218*843e1988Sjohnlev void 4219*843e1988Sjohnlev hat_prepare_mapping(hat_t *hat, caddr_t addr) 4220*843e1988Sjohnlev { 4221*843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4222*843e1988Sjohnlev (void) htable_create(hat, (uintptr_t)addr, 0, NULL); 4223*843e1988Sjohnlev } 4224*843e1988Sjohnlev 4225*843e1988Sjohnlev void 4226*843e1988Sjohnlev hat_release_mapping(hat_t *hat, caddr_t addr) 4227*843e1988Sjohnlev { 4228*843e1988Sjohnlev htable_t *ht; 4229*843e1988Sjohnlev 4230*843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4231*843e1988Sjohnlev ht = htable_lookup(hat, (uintptr_t)addr, 0); 4232*843e1988Sjohnlev ASSERT(ht != NULL); 4233*843e1988Sjohnlev ASSERT(ht->ht_busy >= 2); 4234*843e1988Sjohnlev htable_release(ht); 4235*843e1988Sjohnlev htable_release(ht); 4236*843e1988Sjohnlev } 4237*843e1988Sjohnlev #endif 4238