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> 75843e1988Sjohnlev #ifdef __xpv 76843e1988Sjohnlev #include <sys/hypervisor.h> 77843e1988Sjohnlev #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 * 91843e1988Sjohnlev * 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 * 96843e1988Sjohnlev * vlp_page[0..3] - level==2 PTEs for kernel HAT 97843e1988Sjohnlev * vlp_page[4..7] - level==2 PTEs for user thread on cpu 0 98843e1988Sjohnlev * vlp_page[8..11] - level==2 PTE for user thread on cpu 1 99843e1988Sjohnlev * 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 111843e1988Sjohnlev * kernel reserves a fixed number of entries in the topmost level(s) of page 112843e1988Sjohnlev * tables. The values are setup during startup and then copied to every user 113843e1988Sjohnlev * 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 * 118843e1988Sjohnlev * The hat_kernel_range_ts describe what needs to be copied from kernel hat 119843e1988Sjohnlev * to each user hat. 1207c478bd9Sstevel@tonic-gate */ 121843e1988Sjohnlev typedef struct hat_kernel_range { 122843e1988Sjohnlev level_t hkr_level; 123843e1988Sjohnlev uintptr_t hkr_start_va; 124843e1988Sjohnlev uintptr_t hkr_end_va; /* zero means to end of memory */ 125843e1988Sjohnlev } hat_kernel_range_t; 126843e1988Sjohnlev #define NUM_KERNEL_RANGE 2 127843e1988Sjohnlev static hat_kernel_range_t kernel_ranges[NUM_KERNEL_RANGE]; 128843e1988Sjohnlev 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 13302bc52beSkchow /* export 1g page size to user applications if set */ 134*78b03d3aSkchow int enable_1gpg = 1; 13502bc52beSkchow 13602bc52beSkchow #ifdef DEBUG 13702bc52beSkchow uint_t map1gcnt; 13802bc52beSkchow #endif 13902bc52beSkchow 14002bc52beSkchow 1417c478bd9Sstevel@tonic-gate /* 1427c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 1437c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 1447c478bd9Sstevel@tonic-gate */ 1457c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 1467c478bd9Sstevel@tonic-gate 1477c478bd9Sstevel@tonic-gate /* 1487c478bd9Sstevel@tonic-gate * management stuff for hat structures 1497c478bd9Sstevel@tonic-gate */ 1507c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 1517c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 1527c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 1537c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 1547c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 1557c478bd9Sstevel@tonic-gate 1567c478bd9Sstevel@tonic-gate /* 1577c478bd9Sstevel@tonic-gate * Simple statistics 1587c478bd9Sstevel@tonic-gate */ 1597c478bd9Sstevel@tonic-gate struct hatstats hatstat; 1607c478bd9Sstevel@tonic-gate 161ab4a9bebSjohnlev /* 162ab4a9bebSjohnlev * Some earlier hypervisor versions do not emulate cmpxchg of PTEs 163ab4a9bebSjohnlev * correctly. For such hypervisors we must set PT_USER for kernel 164ab4a9bebSjohnlev * entries ourselves (normally the emulation would set PT_USER for 165ab4a9bebSjohnlev * kernel entries and PT_USER|PT_GLOBAL for user entries). pt_kern is 166ab4a9bebSjohnlev * thus set appropriately. Note that dboot/kbm is OK, as only the full 167ab4a9bebSjohnlev * HAT uses cmpxchg() and the other paths (hypercall etc.) were never 168ab4a9bebSjohnlev * incorrect. 169ab4a9bebSjohnlev */ 170ab4a9bebSjohnlev int pt_kern; 171ab4a9bebSjohnlev 1727c478bd9Sstevel@tonic-gate /* 1737c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 1747c478bd9Sstevel@tonic-gate */ 1757c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 1767c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 1777c478bd9Sstevel@tonic-gate 1787c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 1797c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 1807c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 1817c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 1827c478bd9Sstevel@tonic-gate 1837c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 1847c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 1857c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 1867c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 1877c478bd9Sstevel@tonic-gate 1887c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 1897c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 1907c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 1917c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 1927c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 1937c478bd9Sstevel@tonic-gate 1947c478bd9Sstevel@tonic-gate /* 1957c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 1967c478bd9Sstevel@tonic-gate */ 1977c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 1987c478bd9Sstevel@tonic-gate static int 1997c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 2007c478bd9Sstevel@tonic-gate { 2017c478bd9Sstevel@tonic-gate hat_t *hat = buf; 2027c478bd9Sstevel@tonic-gate 2037c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2047c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 2057c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 206250b7ff9Sjosephb hat->hat_ism_pgcnt = 0; 2077c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 2087c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 2097c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 2107c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2117c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 2127c478bd9Sstevel@tonic-gate return (0); 2137c478bd9Sstevel@tonic-gate } 2147c478bd9Sstevel@tonic-gate 2157c478bd9Sstevel@tonic-gate /* 2167c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 2177c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 2187c478bd9Sstevel@tonic-gate */ 2197c478bd9Sstevel@tonic-gate hat_t * 2207c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 2217c478bd9Sstevel@tonic-gate { 222843e1988Sjohnlev hat_t *hat; 223843e1988Sjohnlev htable_t *ht; /* top level htable */ 224843e1988Sjohnlev uint_t use_vlp; 225843e1988Sjohnlev uint_t r; 226843e1988Sjohnlev hat_kernel_range_t *rp; 227843e1988Sjohnlev uintptr_t va; 228843e1988Sjohnlev uintptr_t eva; 229843e1988Sjohnlev uint_t start; 230843e1988Sjohnlev uint_t cnt; 231843e1988Sjohnlev htable_t *src; 2327c478bd9Sstevel@tonic-gate 2337c478bd9Sstevel@tonic-gate /* 2347c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 2357c478bd9Sstevel@tonic-gate * the htable_steal() code. 2367c478bd9Sstevel@tonic-gate */ 2377c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 2387c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 2397c478bd9Sstevel@tonic-gate 2407c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 2417c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 2427c478bd9Sstevel@tonic-gate hat->hat_as = as; 2437c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 2447c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 2457c478bd9Sstevel@tonic-gate 246843e1988Sjohnlev #if defined(__xpv) 2477c478bd9Sstevel@tonic-gate /* 248843e1988Sjohnlev * No VLP stuff on the hypervisor due to the 64-bit split top level 249843e1988Sjohnlev * page tables. On 32-bit it's not needed as the hypervisor takes 250843e1988Sjohnlev * care of copying the top level PTEs to a below 4Gig page. 2517c478bd9Sstevel@tonic-gate */ 252843e1988Sjohnlev use_vlp = 0; 253843e1988Sjohnlev #else /* __xpv */ 254843e1988Sjohnlev /* 32 bit processes uses a VLP style hat when running with PAE */ 2557c478bd9Sstevel@tonic-gate #if defined(__amd64) 2567c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 2577c478bd9Sstevel@tonic-gate #elif defined(__i386) 2587c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 2597c478bd9Sstevel@tonic-gate #endif 260843e1988Sjohnlev #endif /* __xpv */ 2617c478bd9Sstevel@tonic-gate if (use_vlp) { 2627c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 2637c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 2647c478bd9Sstevel@tonic-gate } 2657c478bd9Sstevel@tonic-gate 2667c478bd9Sstevel@tonic-gate /* 2677c478bd9Sstevel@tonic-gate * Allocate the htable hash 2687c478bd9Sstevel@tonic-gate */ 2697c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 2707c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 2717c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 2727c478bd9Sstevel@tonic-gate } else { 2737c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 2747c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 2757c478bd9Sstevel@tonic-gate } 2767c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 2777c478bd9Sstevel@tonic-gate 2787c478bd9Sstevel@tonic-gate /* 2797c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 280843e1988Sjohnlev * tables for the new hat. 2817c478bd9Sstevel@tonic-gate */ 2827c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 2837c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 284843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2857c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 286843e1988Sjohnlev hat->hat_htable = ht; 287ae115bc7Smrj 288843e1988Sjohnlev #if defined(__amd64) 289843e1988Sjohnlev if (hat->hat_flags & HAT_VLP) 290843e1988Sjohnlev goto init_done; 2917c478bd9Sstevel@tonic-gate #endif 2927c478bd9Sstevel@tonic-gate 293843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 294843e1988Sjohnlev rp = &kernel_ranges[r]; 295843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 296843e1988Sjohnlev va += cnt * LEVEL_SIZE(rp->hkr_level)) { 297843e1988Sjohnlev 298843e1988Sjohnlev if (rp->hkr_level == TOP_LEVEL(hat)) 299843e1988Sjohnlev ht = hat->hat_htable; 300843e1988Sjohnlev else 301843e1988Sjohnlev ht = htable_create(hat, va, rp->hkr_level, 302843e1988Sjohnlev NULL); 303843e1988Sjohnlev 304843e1988Sjohnlev start = htable_va2entry(va, ht); 305843e1988Sjohnlev cnt = HTABLE_NUM_PTES(ht) - start; 306843e1988Sjohnlev eva = va + 307843e1988Sjohnlev ((uintptr_t)cnt << LEVEL_SHIFT(rp->hkr_level)); 308843e1988Sjohnlev if (rp->hkr_end_va != 0 && 309843e1988Sjohnlev (eva > rp->hkr_end_va || eva == 0)) 310843e1988Sjohnlev cnt = htable_va2entry(rp->hkr_end_va, ht) - 311843e1988Sjohnlev start; 312843e1988Sjohnlev 313843e1988Sjohnlev #if defined(__i386) && !defined(__xpv) 314843e1988Sjohnlev if (ht->ht_flags & HTABLE_VLP) { 315843e1988Sjohnlev bcopy(&vlp_page[start], 316843e1988Sjohnlev &hat->hat_vlp_ptes[start], 317843e1988Sjohnlev cnt * sizeof (x86pte_t)); 318843e1988Sjohnlev continue; 319843e1988Sjohnlev } 320843e1988Sjohnlev #endif 321843e1988Sjohnlev src = htable_lookup(kas.a_hat, va, rp->hkr_level); 322843e1988Sjohnlev ASSERT(src != NULL); 323843e1988Sjohnlev x86pte_copy(src, ht, start, cnt); 324843e1988Sjohnlev htable_release(src); 325843e1988Sjohnlev } 326843e1988Sjohnlev } 327843e1988Sjohnlev 328843e1988Sjohnlev init_done: 329843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 330843e1988Sjohnlev 331843e1988Sjohnlev #if defined(__xpv) 3327c478bd9Sstevel@tonic-gate /* 333843e1988Sjohnlev * Pin top level page tables after initializing them 3347c478bd9Sstevel@tonic-gate */ 335843e1988Sjohnlev xen_pin(hat->hat_htable->ht_pfn, mmu.max_level); 336843e1988Sjohnlev #if defined(__amd64) 337843e1988Sjohnlev xen_pin(hat->hat_user_ptable, mmu.max_level); 338843e1988Sjohnlev #endif 3397c478bd9Sstevel@tonic-gate #endif 3407c478bd9Sstevel@tonic-gate 3417c478bd9Sstevel@tonic-gate /* 342a85a6733Sjosephb * Put it at the start of the global list of all hats (used by stealing) 343a85a6733Sjosephb * 344a85a6733Sjosephb * kas.a_hat is not in the list but is instead used to find the 345a85a6733Sjosephb * first and last items in the list. 346a85a6733Sjosephb * 347a85a6733Sjosephb * - kas.a_hat->hat_next points to the start of the user hats. 348a85a6733Sjosephb * The list ends where hat->hat_next == NULL 349a85a6733Sjosephb * 350a85a6733Sjosephb * - kas.a_hat->hat_prev points to the last of the user hats. 351a85a6733Sjosephb * The list begins where hat->hat_prev == NULL 3527c478bd9Sstevel@tonic-gate */ 3537c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 354a85a6733Sjosephb hat->hat_prev = NULL; 355a85a6733Sjosephb hat->hat_next = kas.a_hat->hat_next; 356a85a6733Sjosephb if (hat->hat_next) 357a85a6733Sjosephb hat->hat_next->hat_prev = hat; 358a85a6733Sjosephb else 359a85a6733Sjosephb kas.a_hat->hat_prev = hat; 3607c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 3617c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3627c478bd9Sstevel@tonic-gate 3637c478bd9Sstevel@tonic-gate return (hat); 3647c478bd9Sstevel@tonic-gate } 3657c478bd9Sstevel@tonic-gate 3667c478bd9Sstevel@tonic-gate /* 3677c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 3687c478bd9Sstevel@tonic-gate */ 3697c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3707c478bd9Sstevel@tonic-gate void 3717c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 3727c478bd9Sstevel@tonic-gate { 3737c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock)); 374a85a6733Sjosephb 375a85a6733Sjosephb /* 376a85a6733Sjosephb * If the hat is currently a stealing victim, wait for the stealing 377a85a6733Sjosephb * to finish. Once we mark it as HAT_FREEING, htable_steal() 378a85a6733Sjosephb * won't look at its pagetables anymore. 379a85a6733Sjosephb */ 3807c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 381a85a6733Sjosephb while (hat->hat_flags & HAT_VICTIM) 382a85a6733Sjosephb cv_wait(&hat_list_cv, &hat_list_lock); 3837c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 3847c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 3857c478bd9Sstevel@tonic-gate } 3867c478bd9Sstevel@tonic-gate 3877c478bd9Sstevel@tonic-gate /* 3887c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 3897c478bd9Sstevel@tonic-gate */ 3907c478bd9Sstevel@tonic-gate void 3917c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 3927c478bd9Sstevel@tonic-gate { 3937c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 3947c478bd9Sstevel@tonic-gate 3957c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 3967c478bd9Sstevel@tonic-gate 3977c478bd9Sstevel@tonic-gate /* 3987c478bd9Sstevel@tonic-gate * must not be running on the given hat 3997c478bd9Sstevel@tonic-gate */ 4007c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 4017c478bd9Sstevel@tonic-gate 4027c478bd9Sstevel@tonic-gate /* 403a85a6733Sjosephb * Remove it from the list of HATs 4047c478bd9Sstevel@tonic-gate */ 4057c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 406a85a6733Sjosephb if (hat->hat_prev) 407a85a6733Sjosephb hat->hat_prev->hat_next = hat->hat_next; 408a85a6733Sjosephb else 4097c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 410a85a6733Sjosephb if (hat->hat_next) 411a85a6733Sjosephb hat->hat_next->hat_prev = hat->hat_prev; 412a85a6733Sjosephb else 413a85a6733Sjosephb kas.a_hat->hat_prev = hat->hat_prev; 4147c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 415a85a6733Sjosephb hat->hat_next = hat->hat_prev = NULL; 4167c478bd9Sstevel@tonic-gate 417843e1988Sjohnlev #if defined(__xpv) 418843e1988Sjohnlev /* 419843e1988Sjohnlev * On the hypervisor, unpin top level page table(s) 420843e1988Sjohnlev */ 421843e1988Sjohnlev xen_unpin(hat->hat_htable->ht_pfn); 422843e1988Sjohnlev #if defined(__amd64) 423843e1988Sjohnlev xen_unpin(hat->hat_user_ptable); 424843e1988Sjohnlev #endif 425843e1988Sjohnlev #endif 426843e1988Sjohnlev 4277c478bd9Sstevel@tonic-gate /* 4287c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 4297c478bd9Sstevel@tonic-gate */ 4307c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 4317c478bd9Sstevel@tonic-gate 4327c478bd9Sstevel@tonic-gate /* 4337c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 4347c478bd9Sstevel@tonic-gate */ 4357c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 4367c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 4377c478bd9Sstevel@tonic-gate else 4387c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 4397c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 4407c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 4417c478bd9Sstevel@tonic-gate 4427c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 4437c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 4447c478bd9Sstevel@tonic-gate } 4457c478bd9Sstevel@tonic-gate 4467c478bd9Sstevel@tonic-gate /* 4477c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 4487c478bd9Sstevel@tonic-gate * 4497c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 4507c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 4517c478bd9Sstevel@tonic-gate */ 4527c478bd9Sstevel@tonic-gate uintptr_t 4537c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 4547c478bd9Sstevel@tonic-gate { 4557c478bd9Sstevel@tonic-gate #if defined(__i386) 4567c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 4577c478bd9Sstevel@tonic-gate #endif 4587c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 4597c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 4607c478bd9Sstevel@tonic-gate return (va); 4617c478bd9Sstevel@tonic-gate } 4627c478bd9Sstevel@tonic-gate 4637c478bd9Sstevel@tonic-gate /* 4647c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 4657c478bd9Sstevel@tonic-gate */ 4667c478bd9Sstevel@tonic-gate void 4677c478bd9Sstevel@tonic-gate mmu_init(void) 4687c478bd9Sstevel@tonic-gate { 4697c478bd9Sstevel@tonic-gate uint_t max_htables; 4707c478bd9Sstevel@tonic-gate uint_t pa_bits; 4717c478bd9Sstevel@tonic-gate uint_t va_bits; 4727c478bd9Sstevel@tonic-gate int i; 4737c478bd9Sstevel@tonic-gate 4747c478bd9Sstevel@tonic-gate /* 475ae115bc7Smrj * If CPU enabled the page table global bit, use it for the kernel 476ae115bc7Smrj * This is bit 7 in CR4 (PGE - Page Global Enable). 4777c478bd9Sstevel@tonic-gate */ 478ae115bc7Smrj if ((x86_feature & X86_PGE) != 0 && (getcr4() & CR4_PGE) != 0) 4797c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 4807c478bd9Sstevel@tonic-gate 4817c478bd9Sstevel@tonic-gate /* 482ae115bc7Smrj * Detect NX and PAE usage. 4837c478bd9Sstevel@tonic-gate */ 484ae115bc7Smrj mmu.pae_hat = kbm_pae_support; 485ae115bc7Smrj if (kbm_nx_support) 4867c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 487ae115bc7Smrj else 4887c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 4897c478bd9Sstevel@tonic-gate 4907c478bd9Sstevel@tonic-gate /* 4917c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 4927c478bd9Sstevel@tonic-gate */ 4937c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 4947c478bd9Sstevel@tonic-gate 4957c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 4967c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 4977c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 4987c478bd9Sstevel@tonic-gate } else { 4997c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 5007c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 5017c478bd9Sstevel@tonic-gate } 5027c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 5037c478bd9Sstevel@tonic-gate /* 5047c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 5057c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 5067c478bd9Sstevel@tonic-gate */ 5077c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 5087c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 5097c478bd9Sstevel@tonic-gate #else 5107c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 5117c478bd9Sstevel@tonic-gate #endif 5127c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 5137c478bd9Sstevel@tonic-gate 5147c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 5157c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 5167c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 5177c478bd9Sstevel@tonic-gate 5187c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5197c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 5207c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 5217c478bd9Sstevel@tonic-gate } else { 5227c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 5237c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 5247c478bd9Sstevel@tonic-gate } 5257c478bd9Sstevel@tonic-gate 5267c478bd9Sstevel@tonic-gate if (mmu.pae_hat && (x86_feature & X86_PAE) == 0) 5277c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 5287c478bd9Sstevel@tonic-gate 5297c478bd9Sstevel@tonic-gate if ((x86_feature & X86_CX8) == 0) 5307c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 5317c478bd9Sstevel@tonic-gate 5327c478bd9Sstevel@tonic-gate /* 5337c478bd9Sstevel@tonic-gate * Initialize parameters based on the 64 or 32 bit kernels and 5347c478bd9Sstevel@tonic-gate * for the 32 bit kernel decide if we should use PAE. 5357c478bd9Sstevel@tonic-gate */ 53602bc52beSkchow if (kbm_largepage_support) { 53702bc52beSkchow if (x86_feature & X86_1GPG) { 53802bc52beSkchow mmu.max_page_level = 2; 53902bc52beSkchow mmu.umax_page_level = (enable_1gpg) ? 2 : 1; 54002bc52beSkchow } else { 54102bc52beSkchow mmu.max_page_level = 1; 54202bc52beSkchow mmu.umax_page_level = 1; 54302bc52beSkchow } 54402bc52beSkchow } else { 5457c478bd9Sstevel@tonic-gate mmu.max_page_level = 0; 54602bc52beSkchow mmu.umax_page_level = 0; 54702bc52beSkchow } 5487c478bd9Sstevel@tonic-gate mmu_page_sizes = mmu.max_page_level + 1; 54902bc52beSkchow mmu_exported_page_sizes = mmu.umax_page_level + 1; 55002bc52beSkchow 55102bc52beSkchow /* restrict legacy applications from using pagesizes 1g and above */ 55202bc52beSkchow mmu_legacy_page_sizes = 55302bc52beSkchow (mmu_exported_page_sizes > 2) ? 2 : mmu_exported_page_sizes; 55402bc52beSkchow 5557c478bd9Sstevel@tonic-gate 5567c478bd9Sstevel@tonic-gate #if defined(__amd64) 5577c478bd9Sstevel@tonic-gate 5587c478bd9Sstevel@tonic-gate mmu.num_level = 4; 5597c478bd9Sstevel@tonic-gate mmu.max_level = 3; 5607c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5617c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 5627c478bd9Sstevel@tonic-gate 5637c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5647c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5657c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5667c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 5677c478bd9Sstevel@tonic-gate 5687c478bd9Sstevel@tonic-gate #elif defined(__i386) 5697c478bd9Sstevel@tonic-gate 5707c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 5717c478bd9Sstevel@tonic-gate mmu.num_level = 3; 5727c478bd9Sstevel@tonic-gate mmu.max_level = 2; 5737c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 5747c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 5757c478bd9Sstevel@tonic-gate 5767c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5777c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 5787c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 5797c478bd9Sstevel@tonic-gate 5807c478bd9Sstevel@tonic-gate } else { 5817c478bd9Sstevel@tonic-gate mmu.num_level = 2; 5827c478bd9Sstevel@tonic-gate mmu.max_level = 1; 5837c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 5847c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 5857c478bd9Sstevel@tonic-gate 5867c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 5877c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 5887c478bd9Sstevel@tonic-gate } 5897c478bd9Sstevel@tonic-gate 5907c478bd9Sstevel@tonic-gate #endif /* __i386 */ 5917c478bd9Sstevel@tonic-gate 5927c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 5937c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 5947c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 5957c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 5967c478bd9Sstevel@tonic-gate } 5977c478bd9Sstevel@tonic-gate 598ae115bc7Smrj for (i = 0; i <= mmu.max_page_level; ++i) { 599ab4a9bebSjohnlev mmu.pte_bits[i] = PT_VALID | pt_kern; 600ae115bc7Smrj if (i > 0) 601ae115bc7Smrj mmu.pte_bits[i] |= PT_PAGESIZE; 602ae115bc7Smrj } 6037c478bd9Sstevel@tonic-gate 6047c478bd9Sstevel@tonic-gate /* 6057c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 6067c478bd9Sstevel@tonic-gate */ 6077c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 6087c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 609ae115bc7Smrj 6107c478bd9Sstevel@tonic-gate #if defined(__i386) 6117c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 6127c478bd9Sstevel@tonic-gate #endif 6137c478bd9Sstevel@tonic-gate 6147c478bd9Sstevel@tonic-gate /* 6157c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 6167c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 6177c478bd9Sstevel@tonic-gate * 6187c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 6197c478bd9Sstevel@tonic-gate */ 6207c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 6217c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 6227c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 6237c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 6247c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 6257c478bd9Sstevel@tonic-gate 6267c478bd9Sstevel@tonic-gate #if defined(__amd64) 6277c478bd9Sstevel@tonic-gate /* 6287c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 6297c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 6307c478bd9Sstevel@tonic-gate * a 64 bit process. 6317c478bd9Sstevel@tonic-gate */ 6327c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 6337c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 6347c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 6357c478bd9Sstevel@tonic-gate #endif 6367c478bd9Sstevel@tonic-gate } 6377c478bd9Sstevel@tonic-gate 6387c478bd9Sstevel@tonic-gate 6397c478bd9Sstevel@tonic-gate /* 6407c478bd9Sstevel@tonic-gate * initialize hat data structures 6417c478bd9Sstevel@tonic-gate */ 6427c478bd9Sstevel@tonic-gate void 6437c478bd9Sstevel@tonic-gate hat_init() 6447c478bd9Sstevel@tonic-gate { 6457c478bd9Sstevel@tonic-gate #if defined(__i386) 6467c478bd9Sstevel@tonic-gate /* 6477c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 6487c478bd9Sstevel@tonic-gate */ 6497c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 6507c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 6517c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 6527c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 6537c478bd9Sstevel@tonic-gate } 6547c478bd9Sstevel@tonic-gate #endif 6557c478bd9Sstevel@tonic-gate 6567c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 6577c478bd9Sstevel@tonic-gate 6587c478bd9Sstevel@tonic-gate /* 6597c478bd9Sstevel@tonic-gate * initialize kmem caches 6607c478bd9Sstevel@tonic-gate */ 6617c478bd9Sstevel@tonic-gate htable_init(); 6627c478bd9Sstevel@tonic-gate hment_init(); 6637c478bd9Sstevel@tonic-gate 6647c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 6657c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 6667c478bd9Sstevel@tonic-gate NULL, 0, 0); 6677c478bd9Sstevel@tonic-gate 6687c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 6697c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6707c478bd9Sstevel@tonic-gate NULL, 0, 0); 6717c478bd9Sstevel@tonic-gate 6727c478bd9Sstevel@tonic-gate /* 6737c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 6747c478bd9Sstevel@tonic-gate */ 6757c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 6767c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 6777c478bd9Sstevel@tonic-gate } else { 6787c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 6797c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 6807c478bd9Sstevel@tonic-gate NULL, 0, 0); 6817c478bd9Sstevel@tonic-gate } 6827c478bd9Sstevel@tonic-gate 6837c478bd9Sstevel@tonic-gate /* 6847c478bd9Sstevel@tonic-gate * Set up the kernel's hat 6857c478bd9Sstevel@tonic-gate */ 6867c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER); 6877c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 6887c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 6897c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 6907c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 6917c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(&kas, &kas.a_lock); 6927c478bd9Sstevel@tonic-gate 6937c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 6947c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 6957c478bd9Sstevel@tonic-gate 6967c478bd9Sstevel@tonic-gate /* 6977c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 698a85a6733Sjosephb * The kernel hat's prev pointer tracks the last hat on the list for 699a85a6733Sjosephb * htable_steal() to use. 7007c478bd9Sstevel@tonic-gate */ 7017c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 702a85a6733Sjosephb kas.a_hat->hat_prev = NULL; 7037c478bd9Sstevel@tonic-gate 7047c478bd9Sstevel@tonic-gate /* 7057c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 7067c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 7077c478bd9Sstevel@tonic-gate */ 7087c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 7097c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 7107c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 7117c478bd9Sstevel@tonic-gate 7127c478bd9Sstevel@tonic-gate /* 7137c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 7147c478bd9Sstevel@tonic-gate */ 7157c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 7167c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 7179d9461f9Strevtom 7189d9461f9Strevtom /* 7199d9461f9Strevtom * Pre-allocate hrm_hashtab before enabling the collection of 7209d9461f9Strevtom * refmod statistics. Allocating on the fly would mean us 7219d9461f9Strevtom * running the risk of suffering recursive mutex enters or 7229d9461f9Strevtom * deadlocks. 7239d9461f9Strevtom */ 7249d9461f9Strevtom hrm_hashtab = kmem_zalloc(HRM_HASHSIZE * sizeof (struct hrmstat *), 7259d9461f9Strevtom KM_SLEEP); 7267c478bd9Sstevel@tonic-gate } 7277c478bd9Sstevel@tonic-gate 7287c478bd9Sstevel@tonic-gate /* 7297c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 7307c478bd9Sstevel@tonic-gate * 7317c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 7327c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 7337c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 7347c478bd9Sstevel@tonic-gate */ 7357c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 7367c478bd9Sstevel@tonic-gate static void 7377c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 7387c478bd9Sstevel@tonic-gate { 739843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 7407c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 7417c478bd9Sstevel@tonic-gate pfn_t pfn; 7427c478bd9Sstevel@tonic-gate 7437c478bd9Sstevel@tonic-gate /* 7447c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 7457c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 7467c478bd9Sstevel@tonic-gate */ 7477c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 7487c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7497c478bd9Sstevel@tonic-gate 7507c478bd9Sstevel@tonic-gate /* 7517c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 7527c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 7537c478bd9Sstevel@tonic-gate */ 7547c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 7557c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 7567c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 7577c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 758843e1988Sjohnlev bcopy(vlp_page, hci->hci_vlp_l3ptes, MMU_PAGESIZE); 7597c478bd9Sstevel@tonic-gate 7607c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 7617c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 7627c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 763843e1988Sjohnlev #endif /* __amd64 && !__xpv */ 7647c478bd9Sstevel@tonic-gate } 7657c478bd9Sstevel@tonic-gate 766ae115bc7Smrj /*ARGSUSED*/ 767ae115bc7Smrj static void 768ae115bc7Smrj hat_vlp_teardown(cpu_t *cpu) 769ae115bc7Smrj { 770843e1988Sjohnlev #if defined(__amd64) && !defined(__xpv) 771ae115bc7Smrj struct hat_cpu_info *hci; 772ae115bc7Smrj 773ae115bc7Smrj if ((hci = cpu->cpu_hat_info) == NULL) 774ae115bc7Smrj return; 775ae115bc7Smrj if (hci->hci_vlp_l2ptes) 776ae115bc7Smrj kmem_free(hci->hci_vlp_l2ptes, MMU_PAGESIZE); 777ae115bc7Smrj if (hci->hci_vlp_l3ptes) 778ae115bc7Smrj kmem_free(hci->hci_vlp_l3ptes, MMU_PAGESIZE); 779843e1988Sjohnlev #endif 780843e1988Sjohnlev } 781843e1988Sjohnlev 782843e1988Sjohnlev #define NEXT_HKR(r, l, s, e) { \ 783843e1988Sjohnlev kernel_ranges[r].hkr_level = l; \ 784843e1988Sjohnlev kernel_ranges[r].hkr_start_va = s; \ 785843e1988Sjohnlev kernel_ranges[r].hkr_end_va = e; \ 786843e1988Sjohnlev ++r; \ 787ae115bc7Smrj } 788ae115bc7Smrj 7897c478bd9Sstevel@tonic-gate /* 7907c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 7917c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 7927c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 7937c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 794ae115bc7Smrj * 795ae115bc7Smrj * create the kmap mappings. 7967c478bd9Sstevel@tonic-gate */ 7977c478bd9Sstevel@tonic-gate void 7987c478bd9Sstevel@tonic-gate hat_init_finish(void) 7997c478bd9Sstevel@tonic-gate { 800ae115bc7Smrj size_t size; 801843e1988Sjohnlev uint_t r = 0; 802843e1988Sjohnlev uintptr_t va; 803843e1988Sjohnlev hat_kernel_range_t *rp; 8047c478bd9Sstevel@tonic-gate 8057c478bd9Sstevel@tonic-gate 8067c478bd9Sstevel@tonic-gate /* 807843e1988Sjohnlev * We are now effectively running on the kernel hat. 808843e1988Sjohnlev * Clearing use_boot_reserve shuts off using the pre-allocated boot 809843e1988Sjohnlev * reserve for all HAT allocations. From here on, the reserves are 810843e1988Sjohnlev * only used when avoiding recursion in kmem_alloc(). 8117c478bd9Sstevel@tonic-gate */ 812843e1988Sjohnlev use_boot_reserve = 0; 813843e1988Sjohnlev htable_adjust_reserve(); 8147c478bd9Sstevel@tonic-gate 8157c478bd9Sstevel@tonic-gate /* 816843e1988Sjohnlev * User HATs are initialized with copies of all kernel mappings in 817843e1988Sjohnlev * higher level page tables. Ensure that those entries exist. 8187c478bd9Sstevel@tonic-gate */ 819843e1988Sjohnlev #if defined(__amd64) 820843e1988Sjohnlev 821843e1988Sjohnlev NEXT_HKR(r, 3, kernelbase, 0); 822843e1988Sjohnlev #if defined(__xpv) 823843e1988Sjohnlev NEXT_HKR(r, 3, HYPERVISOR_VIRT_START, HYPERVISOR_VIRT_END); 824843e1988Sjohnlev #endif 825843e1988Sjohnlev 826843e1988Sjohnlev #elif defined(__i386) 827843e1988Sjohnlev 828843e1988Sjohnlev #if !defined(__xpv) 829843e1988Sjohnlev if (mmu.pae_hat) { 830843e1988Sjohnlev va = kernelbase; 831843e1988Sjohnlev if ((va & LEVEL_MASK(2)) != va) { 832843e1988Sjohnlev va = P2ROUNDUP(va, LEVEL_SIZE(2)); 833843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, va); 834843e1988Sjohnlev } 835843e1988Sjohnlev if (va != 0) 836843e1988Sjohnlev NEXT_HKR(r, 2, va, 0); 837843e1988Sjohnlev } else 838843e1988Sjohnlev #endif /* __xpv */ 839843e1988Sjohnlev NEXT_HKR(r, 1, kernelbase, 0); 840843e1988Sjohnlev 841843e1988Sjohnlev #endif /* __i386 */ 842843e1988Sjohnlev 843843e1988Sjohnlev num_kernel_ranges = r; 8447c478bd9Sstevel@tonic-gate 8457c478bd9Sstevel@tonic-gate /* 846843e1988Sjohnlev * Create all the kernel pagetables that will have entries 847843e1988Sjohnlev * shared to user HATs. 8487c478bd9Sstevel@tonic-gate */ 849843e1988Sjohnlev for (r = 0; r < num_kernel_ranges; ++r) { 850843e1988Sjohnlev rp = &kernel_ranges[r]; 851843e1988Sjohnlev for (va = rp->hkr_start_va; va != rp->hkr_end_va; 852843e1988Sjohnlev va += LEVEL_SIZE(rp->hkr_level)) { 853843e1988Sjohnlev htable_t *ht; 854843e1988Sjohnlev 855843e1988Sjohnlev if (IN_HYPERVISOR_VA(va)) 856843e1988Sjohnlev continue; 857843e1988Sjohnlev 858843e1988Sjohnlev /* can/must skip if a page mapping already exists */ 859843e1988Sjohnlev if (rp->hkr_level <= mmu.max_page_level && 860843e1988Sjohnlev (ht = htable_getpage(kas.a_hat, va, NULL)) != 861843e1988Sjohnlev NULL) { 862843e1988Sjohnlev htable_release(ht); 863843e1988Sjohnlev continue; 864843e1988Sjohnlev } 865843e1988Sjohnlev 866843e1988Sjohnlev (void) htable_create(kas.a_hat, va, rp->hkr_level - 1, 867843e1988Sjohnlev NULL); 868843e1988Sjohnlev } 869843e1988Sjohnlev } 8707c478bd9Sstevel@tonic-gate 8717c478bd9Sstevel@tonic-gate /* 872843e1988Sjohnlev * 32 bit PAE metal kernels use only 4 of the 512 entries in the 873843e1988Sjohnlev * page holding the top level pagetable. We use the remainder for 874843e1988Sjohnlev * the "per CPU" page tables for VLP processes. 875843e1988Sjohnlev * Map the top level kernel pagetable into the kernel to make 876843e1988Sjohnlev * it easy to use bcopy access these tables. 8777c478bd9Sstevel@tonic-gate */ 8787c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 8797c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 8807c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 8817c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 882843e1988Sjohnlev #if !defined(__xpv) 883ae115bc7Smrj PROT_WRITE | 884843e1988Sjohnlev #endif 885ae115bc7Smrj PROT_READ | HAT_NOSYNC | HAT_UNORDERED_OK, 8867c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 8877c478bd9Sstevel@tonic-gate } 8887c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 889ae115bc7Smrj 890ae115bc7Smrj /* 891ae115bc7Smrj * Create kmap (cached mappings of kernel PTEs) 892ae115bc7Smrj * for 32 bit we map from segmap_start .. ekernelheap 893ae115bc7Smrj * for 64 bit we map from segmap_start .. segmap_start + segmapsize; 894ae115bc7Smrj */ 895ae115bc7Smrj #if defined(__i386) 896ae115bc7Smrj size = (uintptr_t)ekernelheap - segmap_start; 897ae115bc7Smrj #elif defined(__amd64) 898ae115bc7Smrj size = segmapsize; 899ae115bc7Smrj #endif 900ae115bc7Smrj hat_kmap_init((uintptr_t)segmap_start, size); 9017c478bd9Sstevel@tonic-gate } 9027c478bd9Sstevel@tonic-gate 9037c478bd9Sstevel@tonic-gate /* 9047c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 9057c478bd9Sstevel@tonic-gate * are 32 bit, so for safety we must use cas64() to install these. 9067c478bd9Sstevel@tonic-gate */ 9077c478bd9Sstevel@tonic-gate #ifdef __i386 9087c478bd9Sstevel@tonic-gate static void 9097c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 9107c478bd9Sstevel@tonic-gate { 9117c478bd9Sstevel@tonic-gate x86pte_t *src; 9127c478bd9Sstevel@tonic-gate x86pte_t *dest; 9137c478bd9Sstevel@tonic-gate x86pte_t pte; 9147c478bd9Sstevel@tonic-gate int i; 9157c478bd9Sstevel@tonic-gate 9167c478bd9Sstevel@tonic-gate /* 9177c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 9187c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 9197c478bd9Sstevel@tonic-gate */ 9207c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 9217c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 9227c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 9237c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 9247c478bd9Sstevel@tonic-gate for (;;) { 9257c478bd9Sstevel@tonic-gate pte = dest[i]; 9267c478bd9Sstevel@tonic-gate if (pte == src[i]) 9277c478bd9Sstevel@tonic-gate break; 9287c478bd9Sstevel@tonic-gate if (cas64(dest + i, pte, src[i]) != src[i]) 9297c478bd9Sstevel@tonic-gate break; 9307c478bd9Sstevel@tonic-gate } 9317c478bd9Sstevel@tonic-gate } 9327c478bd9Sstevel@tonic-gate } 9337c478bd9Sstevel@tonic-gate #endif 9347c478bd9Sstevel@tonic-gate 9357c478bd9Sstevel@tonic-gate /* 9367c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 937843e1988Sjohnlev * 938843e1988Sjohnlev * On the 32-bit PAE hypervisor, %cr3 is a 64-bit value, on metal it 939843e1988Sjohnlev * remains a 32-bit value. 9407c478bd9Sstevel@tonic-gate */ 9417c478bd9Sstevel@tonic-gate void 9427c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 9437c478bd9Sstevel@tonic-gate { 944843e1988Sjohnlev uint64_t newcr3; 9457c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 9467c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 9477c478bd9Sstevel@tonic-gate 9487c478bd9Sstevel@tonic-gate /* 9497c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 9507c478bd9Sstevel@tonic-gate */ 9517c478bd9Sstevel@tonic-gate if (old != NULL) { 9527c478bd9Sstevel@tonic-gate if (old == hat) 9537c478bd9Sstevel@tonic-gate return; 9547c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 9557c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 9567c478bd9Sstevel@tonic-gate } 9577c478bd9Sstevel@tonic-gate 9587c478bd9Sstevel@tonic-gate /* 95995c0a3c8Sjosephb * Add this CPU to the active set for this HAT. 9607c478bd9Sstevel@tonic-gate */ 9617c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 9627c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 9637c478bd9Sstevel@tonic-gate } 9647c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 9657c478bd9Sstevel@tonic-gate 9667c478bd9Sstevel@tonic-gate /* 9677c478bd9Sstevel@tonic-gate * now go ahead and load cr3 9687c478bd9Sstevel@tonic-gate */ 9697c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 9707c478bd9Sstevel@tonic-gate #if defined(__amd64) 9717c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 9727c478bd9Sstevel@tonic-gate 9737c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 9747c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 9757c478bd9Sstevel@tonic-gate #elif defined(__i386) 9767c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 9777c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 9787c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 9797c478bd9Sstevel@tonic-gate #endif 9807c478bd9Sstevel@tonic-gate } else { 981843e1988Sjohnlev newcr3 = MAKECR3((uint64_t)hat->hat_htable->ht_pfn); 9827c478bd9Sstevel@tonic-gate } 983843e1988Sjohnlev #ifdef __xpv 984843e1988Sjohnlev { 985843e1988Sjohnlev struct mmuext_op t[2]; 986843e1988Sjohnlev uint_t retcnt; 987843e1988Sjohnlev uint_t opcnt = 1; 988843e1988Sjohnlev 989843e1988Sjohnlev t[0].cmd = MMUEXT_NEW_BASEPTR; 990843e1988Sjohnlev t[0].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 991843e1988Sjohnlev #if defined(__amd64) 992843e1988Sjohnlev /* 993843e1988Sjohnlev * There's an interesting problem here, as to what to 994843e1988Sjohnlev * actually specify when switching to the kernel hat. 995843e1988Sjohnlev * For now we'll reuse the kernel hat again. 996843e1988Sjohnlev */ 997843e1988Sjohnlev t[1].cmd = MMUEXT_NEW_USER_BASEPTR; 998843e1988Sjohnlev if (hat == kas.a_hat) 999843e1988Sjohnlev t[1].arg1.mfn = mmu_btop(pa_to_ma(newcr3)); 1000843e1988Sjohnlev else 1001843e1988Sjohnlev t[1].arg1.mfn = pfn_to_mfn(hat->hat_user_ptable); 1002843e1988Sjohnlev ++opcnt; 1003843e1988Sjohnlev #endif /* __amd64 */ 1004843e1988Sjohnlev if (HYPERVISOR_mmuext_op(t, opcnt, &retcnt, DOMID_SELF) < 0) 1005843e1988Sjohnlev panic("HYPERVISOR_mmu_update() failed"); 1006843e1988Sjohnlev ASSERT(retcnt == opcnt); 1007843e1988Sjohnlev 1008843e1988Sjohnlev } 1009843e1988Sjohnlev #else 10107c478bd9Sstevel@tonic-gate setcr3(newcr3); 1011843e1988Sjohnlev #endif 10127c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 10137c478bd9Sstevel@tonic-gate } 10147c478bd9Sstevel@tonic-gate 10157c478bd9Sstevel@tonic-gate /* 10167c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 10177c478bd9Sstevel@tonic-gate */ 10187c478bd9Sstevel@tonic-gate static x86pte_t 10197c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 10207c478bd9Sstevel@tonic-gate { 10217c478bd9Sstevel@tonic-gate x86pte_t pte; 10227c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 10237c478bd9Sstevel@tonic-gate 10247c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 10257c478bd9Sstevel@tonic-gate 10267c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 10277c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 10287c478bd9Sstevel@tonic-gate 10297c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 10307c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 10317c478bd9Sstevel@tonic-gate 10327c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 10337c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 10347c478bd9Sstevel@tonic-gate 10357c478bd9Sstevel@tonic-gate /* 1036ae115bc7Smrj * Set the software bits used track ref/mod sync's and hments. 1037ae115bc7Smrj * If not using REF/MOD, set them to avoid h/w rewriting PTEs. 10387c478bd9Sstevel@tonic-gate */ 10397c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 1040ae115bc7Smrj PTE_SET(pte, PT_NOCONSIST | PT_REF | PT_MOD); 1041ae115bc7Smrj else if (attr & HAT_NOSYNC) 1042ae115bc7Smrj PTE_SET(pte, PT_NOSYNC | PT_REF | PT_MOD); 10437c478bd9Sstevel@tonic-gate 10447c478bd9Sstevel@tonic-gate /* 10457c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 10467c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 10477c478bd9Sstevel@tonic-gate * to them in the given order. 10487c478bd9Sstevel@tonic-gate * 10497c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 10507c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 10517c478bd9Sstevel@tonic-gate */ 10527c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 10537c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10547c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 10557c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 10567c478bd9Sstevel@tonic-gate /* nothing to set */; 10577c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 10587c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 10597c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) 10607c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 10617c478bd9Sstevel@tonic-gate else 10627c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 10637c478bd9Sstevel@tonic-gate } else { 10647c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 10657c478bd9Sstevel@tonic-gate } 10667c478bd9Sstevel@tonic-gate 10677c478bd9Sstevel@tonic-gate return (pte); 10687c478bd9Sstevel@tonic-gate } 10697c478bd9Sstevel@tonic-gate 10707c478bd9Sstevel@tonic-gate /* 10717c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 10727c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 10737c478bd9Sstevel@tonic-gate */ 10747c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 10757c478bd9Sstevel@tonic-gate int 10767c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 10777c478bd9Sstevel@tonic-gate { 10787c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 10797c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 10807c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 10817c478bd9Sstevel@tonic-gate return (0); 10827c478bd9Sstevel@tonic-gate } 10837c478bd9Sstevel@tonic-gate 10847c478bd9Sstevel@tonic-gate /* 10857c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 10867c478bd9Sstevel@tonic-gate */ 10877c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 10887c478bd9Sstevel@tonic-gate void 10897c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 10907c478bd9Sstevel@tonic-gate { 10917c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 10927c478bd9Sstevel@tonic-gate } 10937c478bd9Sstevel@tonic-gate 10947c478bd9Sstevel@tonic-gate /* 10957c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 10967c478bd9Sstevel@tonic-gate * that is being swapped out. 10977c478bd9Sstevel@tonic-gate */ 10987c478bd9Sstevel@tonic-gate void 10997c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 11007c478bd9Sstevel@tonic-gate { 11017c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 11027c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 11037c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 11047c478bd9Sstevel@tonic-gate level_t l; 11057c478bd9Sstevel@tonic-gate 1106843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 11077c478bd9Sstevel@tonic-gate /* 11087c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 11097c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 11107c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 11117c478bd9Sstevel@tonic-gate * 11127c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 11137c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 11147c478bd9Sstevel@tonic-gate */ 11157c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 11167c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 11177c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 11187c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 11197c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 11207c478bd9Sstevel@tonic-gate 11217c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 11227c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 11237c478bd9Sstevel@tonic-gate if (ht == NULL) 11247c478bd9Sstevel@tonic-gate break; 11257c478bd9Sstevel@tonic-gate 11267c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 11277c478bd9Sstevel@tonic-gate 11287c478bd9Sstevel@tonic-gate /* 11297c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 11307c478bd9Sstevel@tonic-gate * This code knows that only level 0 page tables are shared 11317c478bd9Sstevel@tonic-gate */ 11327c478bd9Sstevel@tonic-gate l = ht->ht_level; 11337c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 11347c478bd9Sstevel@tonic-gate ASSERT(l == 0); 11357c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 11367c478bd9Sstevel@tonic-gate htable_release(ht); 11377c478bd9Sstevel@tonic-gate ht = NULL; 11387c478bd9Sstevel@tonic-gate continue; 11397c478bd9Sstevel@tonic-gate } 11407c478bd9Sstevel@tonic-gate 11417c478bd9Sstevel@tonic-gate /* 11427c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 11437c478bd9Sstevel@tonic-gate */ 11447c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 11457c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 11467c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 11477c478bd9Sstevel@tonic-gate 11487c478bd9Sstevel@tonic-gate /* 11497c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 11507c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 11517c478bd9Sstevel@tonic-gate */ 11527c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 11537c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 11547c478bd9Sstevel@tonic-gate else 11557c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 11567c478bd9Sstevel@tonic-gate } 11577c478bd9Sstevel@tonic-gate if (ht) 11587c478bd9Sstevel@tonic-gate htable_release(ht); 11597c478bd9Sstevel@tonic-gate 11607c478bd9Sstevel@tonic-gate /* 11617c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 11627c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 11637c478bd9Sstevel@tonic-gate */ 11647c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 1165843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 11667c478bd9Sstevel@tonic-gate } 11677c478bd9Sstevel@tonic-gate 11687c478bd9Sstevel@tonic-gate /* 11697c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 11707c478bd9Sstevel@tonic-gate */ 11717c478bd9Sstevel@tonic-gate size_t 11727c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 11737c478bd9Sstevel@tonic-gate { 11747c478bd9Sstevel@tonic-gate size_t total = 0; 11757c478bd9Sstevel@tonic-gate int l; 11767c478bd9Sstevel@tonic-gate 11777c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 11787c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 1179250b7ff9Sjosephb total += hat->hat_ism_pgcnt; 11807c478bd9Sstevel@tonic-gate 11817c478bd9Sstevel@tonic-gate return (total); 11827c478bd9Sstevel@tonic-gate } 11837c478bd9Sstevel@tonic-gate 11847c478bd9Sstevel@tonic-gate /* 11857c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 11867c478bd9Sstevel@tonic-gate */ 11877c478bd9Sstevel@tonic-gate int 11887c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 11897c478bd9Sstevel@tonic-gate { 11907c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, 1); 11917c478bd9Sstevel@tonic-gate return (1); 11927c478bd9Sstevel@tonic-gate } 11937c478bd9Sstevel@tonic-gate 11947c478bd9Sstevel@tonic-gate void 11957c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 11967c478bd9Sstevel@tonic-gate { 11977c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, -1); 11987c478bd9Sstevel@tonic-gate } 11997c478bd9Sstevel@tonic-gate 12007c478bd9Sstevel@tonic-gate /* 12017c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 12027c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 12037c478bd9Sstevel@tonic-gate */ 12047c478bd9Sstevel@tonic-gate static void 12057c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 12067c478bd9Sstevel@tonic-gate { 12077c478bd9Sstevel@tonic-gate uint_t rm = 0; 12087c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 12097c478bd9Sstevel@tonic-gate 1210ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 12117c478bd9Sstevel@tonic-gate return; 12127c478bd9Sstevel@tonic-gate 12137c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 12147c478bd9Sstevel@tonic-gate rm |= P_REF; 12157c478bd9Sstevel@tonic-gate 12167c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 12177c478bd9Sstevel@tonic-gate rm |= P_MOD; 12187c478bd9Sstevel@tonic-gate 12197c478bd9Sstevel@tonic-gate if (rm == 0) 12207c478bd9Sstevel@tonic-gate return; 12217c478bd9Sstevel@tonic-gate 12227c478bd9Sstevel@tonic-gate /* 12237c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 12247c478bd9Sstevel@tonic-gate */ 12257c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 12267c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 12277c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 12287c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 12297c478bd9Sstevel@tonic-gate /* 12307c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 12317c478bd9Sstevel@tonic-gate * pszc below this mapping size 12327c478bd9Sstevel@tonic-gate * since this large mapping existed after we 12337c478bd9Sstevel@tonic-gate * took mlist lock. 12347c478bd9Sstevel@tonic-gate */ 12357c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 12367c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 12377c478bd9Sstevel@tonic-gate ++pp; 12387c478bd9Sstevel@tonic-gate } 12397c478bd9Sstevel@tonic-gate } 12407c478bd9Sstevel@tonic-gate 12417c478bd9Sstevel@tonic-gate /* 12427c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1243843e1988Sjohnlev * that are allowed to change on a HAT_LOAD_REMAP 12447c478bd9Sstevel@tonic-gate */ 12457c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 12467c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 1247843e1988Sjohnlev PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE | PT_IGNORE | PT_REF | PT_MOD) 12487c478bd9Sstevel@tonic-gate 1249b193e412Skchow #define REMAPASSERT(EX) if (!(EX)) panic("hati_pte_map: " #EX) 12507c478bd9Sstevel@tonic-gate /* 12517c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 12527c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 12537c478bd9Sstevel@tonic-gate */ 1254ae115bc7Smrj static int 12557c478bd9Sstevel@tonic-gate hati_pte_map( 12567c478bd9Sstevel@tonic-gate htable_t *ht, 12577c478bd9Sstevel@tonic-gate uint_t entry, 12587c478bd9Sstevel@tonic-gate page_t *pp, 12597c478bd9Sstevel@tonic-gate x86pte_t pte, 12607c478bd9Sstevel@tonic-gate int flags, 12617c478bd9Sstevel@tonic-gate void *pte_ptr) 12627c478bd9Sstevel@tonic-gate { 12637c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 12647c478bd9Sstevel@tonic-gate x86pte_t old_pte; 12657c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 12667c478bd9Sstevel@tonic-gate hment_t *hm; 12677c478bd9Sstevel@tonic-gate uint_t is_consist; 1268ae115bc7Smrj int rv = 0; 12697c478bd9Sstevel@tonic-gate 12707c478bd9Sstevel@tonic-gate /* 12717c478bd9Sstevel@tonic-gate * Is this a consistant (ie. need mapping list lock) mapping? 12727c478bd9Sstevel@tonic-gate */ 12737c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 12747c478bd9Sstevel@tonic-gate 12757c478bd9Sstevel@tonic-gate /* 12767c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 12777c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 12787c478bd9Sstevel@tonic-gate */ 12797c478bd9Sstevel@tonic-gate if ((flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat) 12807c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 12817c478bd9Sstevel@tonic-gate 12827c478bd9Sstevel@tonic-gate /* 12837c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 12847c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 12857c478bd9Sstevel@tonic-gate */ 12867c478bd9Sstevel@tonic-gate if (is_consist) { 12877c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 12887c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 12897c478bd9Sstevel@tonic-gate } 12907c478bd9Sstevel@tonic-gate 12917c478bd9Sstevel@tonic-gate /* 12927c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 12937c478bd9Sstevel@tonic-gate */ 12947c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 12957c478bd9Sstevel@tonic-gate 12967c478bd9Sstevel@tonic-gate /* 1297ae115bc7Smrj * did we get a large page / page table collision? 12987c478bd9Sstevel@tonic-gate */ 1299ae115bc7Smrj if (old_pte == LPAGE_ERROR) { 1300ae115bc7Smrj rv = -1; 1301ae115bc7Smrj goto done; 13027c478bd9Sstevel@tonic-gate } 13037c478bd9Sstevel@tonic-gate 1304ae115bc7Smrj /* 1305ae115bc7Smrj * If the mapping didn't change there is nothing more to do. 1306ae115bc7Smrj */ 1307ae115bc7Smrj if (PTE_EQUIV(pte, old_pte)) 1308ae115bc7Smrj goto done; 1309ae115bc7Smrj 13107c478bd9Sstevel@tonic-gate /* 13117c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 13127c478bd9Sstevel@tonic-gate */ 13137c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 13147c478bd9Sstevel@tonic-gate if (is_consist) { 13157c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 13167c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 13177c478bd9Sstevel@tonic-gate } else { 13187c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 13197c478bd9Sstevel@tonic-gate } 132002bc52beSkchow #if defined(__amd64) 132102bc52beSkchow if (ht->ht_flags & HTABLE_VLP) { 132202bc52beSkchow cpu_t *cpu = CPU; 132302bc52beSkchow x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 132402bc52beSkchow VLP_COPY(hat->hat_vlp_ptes, vlpptep); 132502bc52beSkchow } 132602bc52beSkchow #endif 13277c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 13287c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1329ae115bc7Smrj return (rv); 13307c478bd9Sstevel@tonic-gate } 13317c478bd9Sstevel@tonic-gate 13327c478bd9Sstevel@tonic-gate /* 13337c478bd9Sstevel@tonic-gate * Remap's are more complicated: 13347c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 13357c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 13367c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 13377c478bd9Sstevel@tonic-gate */ 13387c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 13397c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 13407c478bd9Sstevel@tonic-gate 13417c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1342b193e412Skchow REMAPASSERT(flags & HAT_LOAD_REMAP); 1343b193e412Skchow REMAPASSERT(flags & HAT_LOAD_NOCONSIST); 1344ae115bc7Smrj REMAPASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 1345b193e412Skchow REMAPASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 13467c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1347b193e412Skchow REMAPASSERT(!is_consist); 13487c478bd9Sstevel@tonic-gate } 13497c478bd9Sstevel@tonic-gate 13507c478bd9Sstevel@tonic-gate /* 1351843e1988Sjohnlev * We only let remaps change the certain bits in the PTE. 13527c478bd9Sstevel@tonic-gate */ 1353843e1988Sjohnlev if (PTE_GET(old_pte, ~PT_REMAP_BITS) != PTE_GET(pte, ~PT_REMAP_BITS)) 1354843e1988Sjohnlev panic("remap bits changed: old_pte="FMT_PTE", pte="FMT_PTE"\n", 1355843e1988Sjohnlev old_pte, pte); 13567c478bd9Sstevel@tonic-gate 13577c478bd9Sstevel@tonic-gate /* 13587c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 13597c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 13607c478bd9Sstevel@tonic-gate */ 1361ae115bc7Smrj done: 13627c478bd9Sstevel@tonic-gate if (is_consist) { 13637c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 13647c478bd9Sstevel@tonic-gate if (hm != NULL) 13657c478bd9Sstevel@tonic-gate hment_free(hm); 13667c478bd9Sstevel@tonic-gate } 1367ae115bc7Smrj return (rv); 13687c478bd9Sstevel@tonic-gate } 13697c478bd9Sstevel@tonic-gate 13707c478bd9Sstevel@tonic-gate /* 1371ae115bc7Smrj * Internal routine to load a single page table entry. This only fails if 1372ae115bc7Smrj * we attempt to overwrite a page table link with a large page. 13737c478bd9Sstevel@tonic-gate */ 1374ae115bc7Smrj static int 13757c478bd9Sstevel@tonic-gate hati_load_common( 13767c478bd9Sstevel@tonic-gate hat_t *hat, 13777c478bd9Sstevel@tonic-gate uintptr_t va, 13787c478bd9Sstevel@tonic-gate page_t *pp, 13797c478bd9Sstevel@tonic-gate uint_t attr, 13807c478bd9Sstevel@tonic-gate uint_t flags, 13817c478bd9Sstevel@tonic-gate level_t level, 13827c478bd9Sstevel@tonic-gate pfn_t pfn) 13837c478bd9Sstevel@tonic-gate { 13847c478bd9Sstevel@tonic-gate htable_t *ht; 13857c478bd9Sstevel@tonic-gate uint_t entry; 13867c478bd9Sstevel@tonic-gate x86pte_t pte; 1387ae115bc7Smrj int rv = 0; 13887c478bd9Sstevel@tonic-gate 1389aac11643Sjosephb /* 1390aac11643Sjosephb * The number 16 is arbitrary and here to catch a recursion problem 1391aac11643Sjosephb * early before we blow out the kernel stack. 1392aac11643Sjosephb */ 1393aac11643Sjosephb ++curthread->t_hatdepth; 1394aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1395aac11643Sjosephb 13967c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 13977c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 13987c478bd9Sstevel@tonic-gate 13997c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 14007c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 14017c478bd9Sstevel@tonic-gate 14027c478bd9Sstevel@tonic-gate /* 14037c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 14047c478bd9Sstevel@tonic-gate */ 14057c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 14067c478bd9Sstevel@tonic-gate 14077c478bd9Sstevel@tonic-gate /* 1408aac11643Sjosephb * We must have HAT_LOAD_NOCONSIST if page_t is NULL. 14097c478bd9Sstevel@tonic-gate */ 1410aac11643Sjosephb if (pp == NULL) 14117c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 14127c478bd9Sstevel@tonic-gate 14137c478bd9Sstevel@tonic-gate if (ht == NULL) { 14147c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 14157c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 14167c478bd9Sstevel@tonic-gate } 14177c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 14187c478bd9Sstevel@tonic-gate 14197c478bd9Sstevel@tonic-gate /* 14207c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 14217c478bd9Sstevel@tonic-gate */ 14227c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 14237c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 14247c478bd9Sstevel@tonic-gate panic("hati_load_common: bad htable %p, va %p", ht, (void *)va); 14257c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 14267c478bd9Sstevel@tonic-gate 14277c478bd9Sstevel@tonic-gate /* 14287c478bd9Sstevel@tonic-gate * construct the new PTE 14297c478bd9Sstevel@tonic-gate */ 14307c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 14317c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 14327c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 14337c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 14347c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 14357c478bd9Sstevel@tonic-gate 14367c478bd9Sstevel@tonic-gate /* 14377c478bd9Sstevel@tonic-gate * establish the mapping 14387c478bd9Sstevel@tonic-gate */ 1439ae115bc7Smrj rv = hati_pte_map(ht, entry, pp, pte, flags, NULL); 14407c478bd9Sstevel@tonic-gate 14417c478bd9Sstevel@tonic-gate /* 14427c478bd9Sstevel@tonic-gate * release the htable and any reserves 14437c478bd9Sstevel@tonic-gate */ 14447c478bd9Sstevel@tonic-gate htable_release(ht); 1445aac11643Sjosephb --curthread->t_hatdepth; 1446ae115bc7Smrj return (rv); 14477c478bd9Sstevel@tonic-gate } 14487c478bd9Sstevel@tonic-gate 14497c478bd9Sstevel@tonic-gate /* 14507c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 14517c478bd9Sstevel@tonic-gate */ 14527c478bd9Sstevel@tonic-gate static void 14537c478bd9Sstevel@tonic-gate hat_kmap_load( 14547c478bd9Sstevel@tonic-gate caddr_t addr, 14557c478bd9Sstevel@tonic-gate page_t *pp, 14567c478bd9Sstevel@tonic-gate uint_t attr, 14577c478bd9Sstevel@tonic-gate uint_t flags) 14587c478bd9Sstevel@tonic-gate { 14597c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 14607c478bd9Sstevel@tonic-gate x86pte_t pte; 14617c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 14627c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 14637c478bd9Sstevel@tonic-gate htable_t *ht; 14647c478bd9Sstevel@tonic-gate uint_t entry; 14657c478bd9Sstevel@tonic-gate void *pte_ptr; 14667c478bd9Sstevel@tonic-gate 14677c478bd9Sstevel@tonic-gate /* 14687c478bd9Sstevel@tonic-gate * construct the requested PTE 14697c478bd9Sstevel@tonic-gate */ 14707c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 14717c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 14727c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 14737c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 14747c478bd9Sstevel@tonic-gate 14757c478bd9Sstevel@tonic-gate /* 14767c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 14777c478bd9Sstevel@tonic-gate */ 14787c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 14797c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 14807c478bd9Sstevel@tonic-gate else 14817c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 14827c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 14837c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 14847c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1485aac11643Sjosephb ++curthread->t_hatdepth; 1486aac11643Sjosephb ASSERT(curthread->t_hatdepth < 16); 1487ae115bc7Smrj (void) hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1488aac11643Sjosephb --curthread->t_hatdepth; 14897c478bd9Sstevel@tonic-gate } 14907c478bd9Sstevel@tonic-gate 14917c478bd9Sstevel@tonic-gate /* 14927c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 14937c478bd9Sstevel@tonic-gate * 14947c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 14957c478bd9Sstevel@tonic-gate * 14967c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 14977c478bd9Sstevel@tonic-gate * 14987c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 14997c478bd9Sstevel@tonic-gate * and hat_devload(). 15007c478bd9Sstevel@tonic-gate * 15017c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1502ae115bc7Smrj * sets PT_NOCONSIST 15037c478bd9Sstevel@tonic-gate * 15047c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 15057c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 15067c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 15077c478bd9Sstevel@tonic-gate * 15087c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 15097c478bd9Sstevel@tonic-gate * 15107c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 15117c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 15127c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 15137c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 15147c478bd9Sstevel@tonic-gate * recursion. 15157c478bd9Sstevel@tonic-gate * 15167c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 15177c478bd9Sstevel@tonic-gate * 1518ae115bc7Smrj * HAT_NOSYNC set PT_NOSYNC - this mapping's ref/mod bits 15197c478bd9Sstevel@tonic-gate * are never cleared. 15207c478bd9Sstevel@tonic-gate * 15217c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 15227c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 15237c478bd9Sstevel@tonic-gate * page_t being mapped. 15247c478bd9Sstevel@tonic-gate */ 15257c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 15267c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 15277c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 15287c478bd9Sstevel@tonic-gate 15297c478bd9Sstevel@tonic-gate void 15307c478bd9Sstevel@tonic-gate hat_memload( 15317c478bd9Sstevel@tonic-gate hat_t *hat, 15327c478bd9Sstevel@tonic-gate caddr_t addr, 15337c478bd9Sstevel@tonic-gate page_t *pp, 15347c478bd9Sstevel@tonic-gate uint_t attr, 15357c478bd9Sstevel@tonic-gate uint_t flags) 15367c478bd9Sstevel@tonic-gate { 15377c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15387c478bd9Sstevel@tonic-gate level_t level = 0; 15397c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 15407c478bd9Sstevel@tonic-gate 1541843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 15427c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1543ae115bc7Smrj ASSERT(hat == kas.a_hat || va < _userlimit); 15447c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 15457c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 15467c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 15477c478bd9Sstevel@tonic-gate 15487c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 15497c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 15507c478bd9Sstevel@tonic-gate 15517c478bd9Sstevel@tonic-gate /* 15527c478bd9Sstevel@tonic-gate * kernel address special case for performance. 15537c478bd9Sstevel@tonic-gate */ 15547c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 15557c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 15567c478bd9Sstevel@tonic-gate hat_kmap_load(addr, pp, attr, flags); 1557843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 15587c478bd9Sstevel@tonic-gate return; 15597c478bd9Sstevel@tonic-gate } 15607c478bd9Sstevel@tonic-gate 15617c478bd9Sstevel@tonic-gate /* 15627c478bd9Sstevel@tonic-gate * This is used for memory with normal caching enabled, so 15637c478bd9Sstevel@tonic-gate * always set HAT_STORECACHING_OK. 15647c478bd9Sstevel@tonic-gate */ 15657c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1566ae115bc7Smrj if (hati_load_common(hat, va, pp, attr, flags, level, pfn) != 0) 1567ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1568843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 15697c478bd9Sstevel@tonic-gate } 15707c478bd9Sstevel@tonic-gate 157105d3dc4bSpaulsan /* ARGSUSED */ 157205d3dc4bSpaulsan void 157305d3dc4bSpaulsan hat_memload_region(struct hat *hat, caddr_t addr, struct page *pp, 157405d3dc4bSpaulsan uint_t attr, uint_t flags, hat_region_cookie_t rcookie) 157505d3dc4bSpaulsan { 157605d3dc4bSpaulsan hat_memload(hat, addr, pp, attr, flags); 157705d3dc4bSpaulsan } 157805d3dc4bSpaulsan 15797c478bd9Sstevel@tonic-gate /* 15807c478bd9Sstevel@tonic-gate * Load the given array of page structs using large pages when possible 15817c478bd9Sstevel@tonic-gate */ 15827c478bd9Sstevel@tonic-gate void 15837c478bd9Sstevel@tonic-gate hat_memload_array( 15847c478bd9Sstevel@tonic-gate hat_t *hat, 15857c478bd9Sstevel@tonic-gate caddr_t addr, 15867c478bd9Sstevel@tonic-gate size_t len, 15877c478bd9Sstevel@tonic-gate page_t **pages, 15887c478bd9Sstevel@tonic-gate uint_t attr, 15897c478bd9Sstevel@tonic-gate uint_t flags) 15907c478bd9Sstevel@tonic-gate { 15917c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 15927c478bd9Sstevel@tonic-gate uintptr_t eaddr = va + len; 15937c478bd9Sstevel@tonic-gate level_t level; 15947c478bd9Sstevel@tonic-gate size_t pgsize; 15957c478bd9Sstevel@tonic-gate pgcnt_t pgindx = 0; 15967c478bd9Sstevel@tonic-gate pfn_t pfn; 15977c478bd9Sstevel@tonic-gate pgcnt_t i; 15987c478bd9Sstevel@tonic-gate 1599843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 16007c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1601ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 16027c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 16037c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 16047c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 16057c478bd9Sstevel@tonic-gate 16067c478bd9Sstevel@tonic-gate /* 16077c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 16087c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 16097c478bd9Sstevel@tonic-gate */ 16107c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 16117c478bd9Sstevel@tonic-gate 16127c478bd9Sstevel@tonic-gate /* 16137c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 16147c478bd9Sstevel@tonic-gate */ 16157c478bd9Sstevel@tonic-gate while (va < eaddr) { 16167c478bd9Sstevel@tonic-gate /* 16177c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 16187c478bd9Sstevel@tonic-gate */ 16197c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 16207c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 16217c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 16227c478bd9Sstevel@tonic-gate if (level == 0) 16237c478bd9Sstevel@tonic-gate break; 1624ae115bc7Smrj 16257c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 16267c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1627ae115bc7Smrj !IS_P2ALIGNED(pfn_to_pa(pfn), pgsize)) 16287c478bd9Sstevel@tonic-gate continue; 16297c478bd9Sstevel@tonic-gate 16307c478bd9Sstevel@tonic-gate /* 16317c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 16327c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 16337c478bd9Sstevel@tonic-gate * of the large page. 16347c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 16357c478bd9Sstevel@tonic-gate * all pages are locked. 16367c478bd9Sstevel@tonic-gate */ 16377c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 16387c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 16397c478bd9Sstevel@tonic-gate if (pfn + i != 16407c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 16417c478bd9Sstevel@tonic-gate break; 16427c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 16437c478bd9Sstevel@tonic-gate level); 16447c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 16457c478bd9Sstevel@tonic-gate pages[pgindx + i]); 16467c478bd9Sstevel@tonic-gate } 164702bc52beSkchow if (i == mmu_btop(pgsize)) { 164802bc52beSkchow #ifdef DEBUG 164902bc52beSkchow if (level == 2) 165002bc52beSkchow map1gcnt++; 165102bc52beSkchow #endif 16527c478bd9Sstevel@tonic-gate break; 165302bc52beSkchow } 16547c478bd9Sstevel@tonic-gate } 16557c478bd9Sstevel@tonic-gate } 16567c478bd9Sstevel@tonic-gate 16577c478bd9Sstevel@tonic-gate /* 1658ae115bc7Smrj * Load this page mapping. If the load fails, try a smaller 1659ae115bc7Smrj * pagesize. 16607c478bd9Sstevel@tonic-gate */ 16617c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1662ae115bc7Smrj while (hati_load_common(hat, va, pages[pgindx], attr, 1663250b7ff9Sjosephb flags, level, pfn) != 0) { 1664ae115bc7Smrj if (level == 0) 1665ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1666ae115bc7Smrj --level; 1667ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1668ae115bc7Smrj } 16697c478bd9Sstevel@tonic-gate 16707c478bd9Sstevel@tonic-gate /* 16717c478bd9Sstevel@tonic-gate * move to next page 16727c478bd9Sstevel@tonic-gate */ 16737c478bd9Sstevel@tonic-gate va += pgsize; 16747c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 16757c478bd9Sstevel@tonic-gate } 1676843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 16777c478bd9Sstevel@tonic-gate } 16787c478bd9Sstevel@tonic-gate 167905d3dc4bSpaulsan /* ARGSUSED */ 168005d3dc4bSpaulsan void 168105d3dc4bSpaulsan hat_memload_array_region(struct hat *hat, caddr_t addr, size_t len, 168205d3dc4bSpaulsan struct page **pps, uint_t attr, uint_t flags, 168305d3dc4bSpaulsan hat_region_cookie_t rcookie) 168405d3dc4bSpaulsan { 168505d3dc4bSpaulsan hat_memload_array(hat, addr, len, pps, attr, flags); 168605d3dc4bSpaulsan } 168705d3dc4bSpaulsan 16887c478bd9Sstevel@tonic-gate /* 16897c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 16907c478bd9Sstevel@tonic-gate * load/lock the given page frame number 16917c478bd9Sstevel@tonic-gate * 16927c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 16937c478bd9Sstevel@tonic-gate * 16947c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 16957c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 16967c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 16977c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 16987c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 16997c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 17007c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 17017c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 17027c478bd9Sstevel@tonic-gate * This also implies re-ordering. 17037c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 17047c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 17057c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 17067c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 17077c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 17087c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 17097c478bd9Sstevel@tonic-gate * 17107c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 17117c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 17127c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 17137c478bd9Sstevel@tonic-gate */ 17147c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 17157c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 17167c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 17177c478bd9Sstevel@tonic-gate 17187c478bd9Sstevel@tonic-gate void 17197c478bd9Sstevel@tonic-gate hat_devload( 17207c478bd9Sstevel@tonic-gate hat_t *hat, 17217c478bd9Sstevel@tonic-gate caddr_t addr, 17227c478bd9Sstevel@tonic-gate size_t len, 17237c478bd9Sstevel@tonic-gate pfn_t pfn, 17247c478bd9Sstevel@tonic-gate uint_t attr, 17257c478bd9Sstevel@tonic-gate int flags) 17267c478bd9Sstevel@tonic-gate { 17277c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 17287c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 17297c478bd9Sstevel@tonic-gate level_t level; 17307c478bd9Sstevel@tonic-gate size_t pgsize; 17317c478bd9Sstevel@tonic-gate page_t *pp; 17327c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 17337c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 17347c478bd9Sstevel@tonic-gate 1735843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 17367c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1737ae115bc7Smrj ASSERT(hat == kas.a_hat || eva <= _userlimit); 17387c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 17397c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 17407c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 17417c478bd9Sstevel@tonic-gate 17427c478bd9Sstevel@tonic-gate /* 17437c478bd9Sstevel@tonic-gate * handle all pages 17447c478bd9Sstevel@tonic-gate */ 17457c478bd9Sstevel@tonic-gate while (va < eva) { 17467c478bd9Sstevel@tonic-gate 17477c478bd9Sstevel@tonic-gate /* 17487c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 17497c478bd9Sstevel@tonic-gate */ 17507c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 17517c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 17527c478bd9Sstevel@tonic-gate if (level == 0) 17537c478bd9Sstevel@tonic-gate break; 17547c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 17557c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 175602bc52beSkchow IS_P2ALIGNED(pfn, mmu_btop(pgsize))) { 175702bc52beSkchow #ifdef DEBUG 175802bc52beSkchow if (level == 2) 175902bc52beSkchow map1gcnt++; 176002bc52beSkchow #endif 17617c478bd9Sstevel@tonic-gate break; 176202bc52beSkchow } 17637c478bd9Sstevel@tonic-gate } 17647c478bd9Sstevel@tonic-gate 17657c478bd9Sstevel@tonic-gate /* 1766ae115bc7Smrj * If this is just memory then allow caching (this happens 17677c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1768ae115bc7Smrj * to override that. If we don't have a page_t then make sure 17697c478bd9Sstevel@tonic-gate * NOCONSIST is set. 17707c478bd9Sstevel@tonic-gate */ 17717c478bd9Sstevel@tonic-gate a = attr; 17727c478bd9Sstevel@tonic-gate f = flags; 1773843e1988Sjohnlev if (!pf_is_memory(pfn)) 1774843e1988Sjohnlev f |= HAT_LOAD_NOCONSIST; 1775843e1988Sjohnlev else if (!(a & HAT_PLAT_NOCACHE)) 1776843e1988Sjohnlev a |= HAT_STORECACHING_OK; 17777c478bd9Sstevel@tonic-gate 1778843e1988Sjohnlev if (f & HAT_LOAD_NOCONSIST) 17797c478bd9Sstevel@tonic-gate pp = NULL; 1780843e1988Sjohnlev else 1781843e1988Sjohnlev pp = page_numtopp_nolock(pfn); 17827c478bd9Sstevel@tonic-gate 17837c478bd9Sstevel@tonic-gate /* 17847c478bd9Sstevel@tonic-gate * load this page mapping 17857c478bd9Sstevel@tonic-gate */ 17867c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1787ae115bc7Smrj while (hati_load_common(hat, va, pp, a, f, level, pfn) != 0) { 1788ae115bc7Smrj if (level == 0) 1789ae115bc7Smrj panic("unexpected hati_load_common() failure"); 1790ae115bc7Smrj --level; 1791ae115bc7Smrj pgsize = LEVEL_SIZE(level); 1792ae115bc7Smrj } 17937c478bd9Sstevel@tonic-gate 17947c478bd9Sstevel@tonic-gate /* 17957c478bd9Sstevel@tonic-gate * move to next page 17967c478bd9Sstevel@tonic-gate */ 17977c478bd9Sstevel@tonic-gate va += pgsize; 17987c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 17997c478bd9Sstevel@tonic-gate } 1800843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 18017c478bd9Sstevel@tonic-gate } 18027c478bd9Sstevel@tonic-gate 18037c478bd9Sstevel@tonic-gate /* 18047c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 18057c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 18067c478bd9Sstevel@tonic-gate * 18077c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 18087c478bd9Sstevel@tonic-gate */ 18097c478bd9Sstevel@tonic-gate void 18107c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 18117c478bd9Sstevel@tonic-gate { 18127c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 18137c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 18147c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 18157c478bd9Sstevel@tonic-gate 18167c478bd9Sstevel@tonic-gate /* 18177c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 18187c478bd9Sstevel@tonic-gate */ 1819ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 18207c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 18217c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 18227c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 18237c478bd9Sstevel@tonic-gate return; 18247c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 18257c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 18267c478bd9Sstevel@tonic-gate 1827843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 18287c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 18297c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 18307c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 18317c478bd9Sstevel@tonic-gate if (ht == NULL) 18327c478bd9Sstevel@tonic-gate break; 18337c478bd9Sstevel@tonic-gate 18347c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 18357c478bd9Sstevel@tonic-gate 18367c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 18377c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 18387c478bd9Sstevel@tonic-gate "htable=%p, vaddr=%p\n", ht, (caddr_t)vaddr); 18397c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 18407c478bd9Sstevel@tonic-gate 18417c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 18427c478bd9Sstevel@tonic-gate } 18437c478bd9Sstevel@tonic-gate if (ht) 18447c478bd9Sstevel@tonic-gate htable_release(ht); 1845843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 18467c478bd9Sstevel@tonic-gate } 18477c478bd9Sstevel@tonic-gate 184805d3dc4bSpaulsan /* ARGSUSED */ 184905d3dc4bSpaulsan void 18507dacfc44Spaulsan hat_unlock_region(struct hat *hat, caddr_t addr, size_t len, 185105d3dc4bSpaulsan hat_region_cookie_t rcookie) 185205d3dc4bSpaulsan { 185305d3dc4bSpaulsan panic("No shared region support on x86"); 185405d3dc4bSpaulsan } 185505d3dc4bSpaulsan 1856843e1988Sjohnlev #if !defined(__xpv) 18577c478bd9Sstevel@tonic-gate /* 18587c478bd9Sstevel@tonic-gate * Cross call service routine to demap a virtual page on 18597c478bd9Sstevel@tonic-gate * the current CPU or flush all mappings in TLB. 18607c478bd9Sstevel@tonic-gate */ 18617c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 18627c478bd9Sstevel@tonic-gate static int 18637c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 18647c478bd9Sstevel@tonic-gate { 18657c478bd9Sstevel@tonic-gate hat_t *hat = (hat_t *)a1; 18667c478bd9Sstevel@tonic-gate caddr_t addr = (caddr_t)a2; 18677c478bd9Sstevel@tonic-gate 18687c478bd9Sstevel@tonic-gate /* 18697c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 18707c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 18717c478bd9Sstevel@tonic-gate */ 18727c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 18737c478bd9Sstevel@tonic-gate return (0); 18747c478bd9Sstevel@tonic-gate 18757c478bd9Sstevel@tonic-gate /* 18767c478bd9Sstevel@tonic-gate * For a normal address, we just flush one page mapping 18777c478bd9Sstevel@tonic-gate */ 18787c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1879ae115bc7Smrj mmu_tlbflush_entry(addr); 18807c478bd9Sstevel@tonic-gate return (0); 18817c478bd9Sstevel@tonic-gate } 18827c478bd9Sstevel@tonic-gate 18837c478bd9Sstevel@tonic-gate /* 18847c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 18857c478bd9Sstevel@tonic-gate * 18867c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 18877c478bd9Sstevel@tonic-gate * the pte values from the struct hat 18887c478bd9Sstevel@tonic-gate */ 18897c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 18907c478bd9Sstevel@tonic-gate #if defined(__amd64) 18917c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 18927c478bd9Sstevel@tonic-gate 18937c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 18947c478bd9Sstevel@tonic-gate #elif defined(__i386) 18957c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 18967c478bd9Sstevel@tonic-gate #endif 18977c478bd9Sstevel@tonic-gate } 18987c478bd9Sstevel@tonic-gate reload_cr3(); 18997c478bd9Sstevel@tonic-gate return (0); 19007c478bd9Sstevel@tonic-gate } 19017c478bd9Sstevel@tonic-gate 190295c0a3c8Sjosephb /* 190395c0a3c8Sjosephb * Flush all TLB entries, including global (ie. kernel) ones. 190495c0a3c8Sjosephb */ 190595c0a3c8Sjosephb static void 190695c0a3c8Sjosephb flush_all_tlb_entries(void) 190795c0a3c8Sjosephb { 190895c0a3c8Sjosephb ulong_t cr4 = getcr4(); 190995c0a3c8Sjosephb 191095c0a3c8Sjosephb if (cr4 & CR4_PGE) { 191195c0a3c8Sjosephb setcr4(cr4 & ~(ulong_t)CR4_PGE); 191295c0a3c8Sjosephb setcr4(cr4); 191395c0a3c8Sjosephb 191495c0a3c8Sjosephb /* 191595c0a3c8Sjosephb * 32 bit PAE also needs to always reload_cr3() 191695c0a3c8Sjosephb */ 191795c0a3c8Sjosephb if (mmu.max_level == 2) 191895c0a3c8Sjosephb reload_cr3(); 191995c0a3c8Sjosephb } else { 192095c0a3c8Sjosephb reload_cr3(); 192195c0a3c8Sjosephb } 192295c0a3c8Sjosephb } 192395c0a3c8Sjosephb 192495c0a3c8Sjosephb #define TLB_CPU_HALTED (01ul) 192595c0a3c8Sjosephb #define TLB_INVAL_ALL (02ul) 192695c0a3c8Sjosephb #define CAS_TLB_INFO(cpu, old, new) \ 192795c0a3c8Sjosephb caslong((ulong_t *)&(cpu)->cpu_m.mcpu_tlb_info, (old), (new)) 192895c0a3c8Sjosephb 192995c0a3c8Sjosephb /* 193095c0a3c8Sjosephb * Record that a CPU is going idle 193195c0a3c8Sjosephb */ 193295c0a3c8Sjosephb void 193395c0a3c8Sjosephb tlb_going_idle(void) 193495c0a3c8Sjosephb { 193595c0a3c8Sjosephb atomic_or_long((ulong_t *)&CPU->cpu_m.mcpu_tlb_info, TLB_CPU_HALTED); 193695c0a3c8Sjosephb } 193795c0a3c8Sjosephb 193895c0a3c8Sjosephb /* 193995c0a3c8Sjosephb * Service a delayed TLB flush if coming out of being idle. 194095c0a3c8Sjosephb */ 194195c0a3c8Sjosephb void 194295c0a3c8Sjosephb tlb_service(void) 194395c0a3c8Sjosephb { 194495c0a3c8Sjosephb ulong_t flags = getflags(); 194595c0a3c8Sjosephb ulong_t tlb_info; 194695c0a3c8Sjosephb ulong_t found; 194795c0a3c8Sjosephb 194895c0a3c8Sjosephb /* 194995c0a3c8Sjosephb * Be sure interrupts are off while doing this so that 195095c0a3c8Sjosephb * higher level interrupts correctly wait for flushes to finish. 195195c0a3c8Sjosephb */ 195295c0a3c8Sjosephb if (flags & PS_IE) 195395c0a3c8Sjosephb flags = intr_clear(); 195495c0a3c8Sjosephb 195595c0a3c8Sjosephb /* 195695c0a3c8Sjosephb * We only have to do something if coming out of being idle. 195795c0a3c8Sjosephb */ 195895c0a3c8Sjosephb tlb_info = CPU->cpu_m.mcpu_tlb_info; 195995c0a3c8Sjosephb if (tlb_info & TLB_CPU_HALTED) { 196095c0a3c8Sjosephb ASSERT(CPU->cpu_current_hat == kas.a_hat); 196195c0a3c8Sjosephb 196295c0a3c8Sjosephb /* 196395c0a3c8Sjosephb * Atomic clear and fetch of old state. 196495c0a3c8Sjosephb */ 196595c0a3c8Sjosephb while ((found = CAS_TLB_INFO(CPU, tlb_info, 0)) != tlb_info) { 196695c0a3c8Sjosephb ASSERT(found & TLB_CPU_HALTED); 196795c0a3c8Sjosephb tlb_info = found; 196895c0a3c8Sjosephb SMT_PAUSE(); 196995c0a3c8Sjosephb } 197095c0a3c8Sjosephb if (tlb_info & TLB_INVAL_ALL) 197195c0a3c8Sjosephb flush_all_tlb_entries(); 197295c0a3c8Sjosephb } 197395c0a3c8Sjosephb 197495c0a3c8Sjosephb /* 197595c0a3c8Sjosephb * Restore interrupt enable control bit. 197695c0a3c8Sjosephb */ 197795c0a3c8Sjosephb if (flags & PS_IE) 197895c0a3c8Sjosephb sti(); 197995c0a3c8Sjosephb } 1980843e1988Sjohnlev #endif /* !__xpv */ 198195c0a3c8Sjosephb 19827c478bd9Sstevel@tonic-gate /* 19837c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 19847c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 19857c478bd9Sstevel@tonic-gate */ 19867c478bd9Sstevel@tonic-gate void 1987ae115bc7Smrj hat_tlb_inval(hat_t *hat, uintptr_t va) 19887c478bd9Sstevel@tonic-gate { 19897c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 19907c478bd9Sstevel@tonic-gate cpuset_t justme; 1991ae115bc7Smrj cpuset_t cpus_to_shootdown; 1992843e1988Sjohnlev #ifndef __xpv 1993843e1988Sjohnlev cpuset_t check_cpus; 199495c0a3c8Sjosephb cpu_t *cpup; 199595c0a3c8Sjosephb int c; 1996843e1988Sjohnlev #endif 19977c478bd9Sstevel@tonic-gate 19987c478bd9Sstevel@tonic-gate /* 19997c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 20007c478bd9Sstevel@tonic-gate * demap need not do anything. 20017c478bd9Sstevel@tonic-gate */ 20027c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 20037c478bd9Sstevel@tonic-gate return; 20047c478bd9Sstevel@tonic-gate 20057c478bd9Sstevel@tonic-gate /* 20067c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 20077c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 20087c478bd9Sstevel@tonic-gate * these were shared at. 20097c478bd9Sstevel@tonic-gate */ 20107c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 20117c478bd9Sstevel@tonic-gate hat = kas.a_hat; 20127c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 20137c478bd9Sstevel@tonic-gate } 20147c478bd9Sstevel@tonic-gate 20157c478bd9Sstevel@tonic-gate /* 20167c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 20177c478bd9Sstevel@tonic-gate */ 20187c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 2019843e1988Sjohnlev #ifdef __xpv 2020843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2021843e1988Sjohnlev xen_flush_tlb(); 2022843e1988Sjohnlev else 2023843e1988Sjohnlev xen_flush_va((caddr_t)va); 2024843e1988Sjohnlev #else 20257c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 2026843e1988Sjohnlev #endif 20277c478bd9Sstevel@tonic-gate return; 20287c478bd9Sstevel@tonic-gate } 20297c478bd9Sstevel@tonic-gate 20307c478bd9Sstevel@tonic-gate 20317c478bd9Sstevel@tonic-gate /* 2032ae115bc7Smrj * Determine CPUs to shootdown. Kernel changes always do all CPUs. 2033ae115bc7Smrj * Otherwise it's just CPUs currently executing in this hat. 20347c478bd9Sstevel@tonic-gate */ 20357c478bd9Sstevel@tonic-gate kpreempt_disable(); 20367c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 2037ae115bc7Smrj if (hat == kas.a_hat) 2038ae115bc7Smrj cpus_to_shootdown = khat_cpuset; 20397c478bd9Sstevel@tonic-gate else 2040ae115bc7Smrj cpus_to_shootdown = hat->hat_cpus; 2041ae115bc7Smrj 2042843e1988Sjohnlev #ifndef __xpv 204395c0a3c8Sjosephb /* 204495c0a3c8Sjosephb * If any CPUs in the set are idle, just request a delayed flush 204595c0a3c8Sjosephb * and avoid waking them up. 204695c0a3c8Sjosephb */ 204795c0a3c8Sjosephb check_cpus = cpus_to_shootdown; 204895c0a3c8Sjosephb for (c = 0; c < NCPU && !CPUSET_ISNULL(check_cpus); ++c) { 204995c0a3c8Sjosephb ulong_t tlb_info; 205095c0a3c8Sjosephb 205195c0a3c8Sjosephb if (!CPU_IN_SET(check_cpus, c)) 205295c0a3c8Sjosephb continue; 205395c0a3c8Sjosephb CPUSET_DEL(check_cpus, c); 205495c0a3c8Sjosephb cpup = cpu[c]; 205595c0a3c8Sjosephb if (cpup == NULL) 205695c0a3c8Sjosephb continue; 205795c0a3c8Sjosephb 205895c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 205995c0a3c8Sjosephb while (tlb_info == TLB_CPU_HALTED) { 206095c0a3c8Sjosephb (void) CAS_TLB_INFO(cpup, TLB_CPU_HALTED, 2061250b7ff9Sjosephb TLB_CPU_HALTED | TLB_INVAL_ALL); 206295c0a3c8Sjosephb SMT_PAUSE(); 206395c0a3c8Sjosephb tlb_info = cpup->cpu_m.mcpu_tlb_info; 206495c0a3c8Sjosephb } 206595c0a3c8Sjosephb if (tlb_info == (TLB_CPU_HALTED | TLB_INVAL_ALL)) { 206695c0a3c8Sjosephb HATSTAT_INC(hs_tlb_inval_delayed); 206795c0a3c8Sjosephb CPUSET_DEL(cpus_to_shootdown, c); 206895c0a3c8Sjosephb } 206995c0a3c8Sjosephb } 2070843e1988Sjohnlev #endif 207195c0a3c8Sjosephb 2072ae115bc7Smrj if (CPUSET_ISNULL(cpus_to_shootdown) || 2073ae115bc7Smrj CPUSET_ISEQUAL(cpus_to_shootdown, justme)) { 2074ae115bc7Smrj 2075843e1988Sjohnlev #ifdef __xpv 2076843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2077843e1988Sjohnlev xen_flush_tlb(); 2078843e1988Sjohnlev else 2079843e1988Sjohnlev xen_flush_va((caddr_t)va); 2080843e1988Sjohnlev #else 2081ae115bc7Smrj (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 2082843e1988Sjohnlev #endif 2083ae115bc7Smrj 2084ae115bc7Smrj } else { 2085ae115bc7Smrj 2086ae115bc7Smrj CPUSET_ADD(cpus_to_shootdown, CPU->cpu_id); 2087843e1988Sjohnlev #ifdef __xpv 2088843e1988Sjohnlev if (va == DEMAP_ALL_ADDR) 2089843e1988Sjohnlev xen_gflush_tlb(cpus_to_shootdown); 2090843e1988Sjohnlev else 2091843e1988Sjohnlev xen_gflush_va((caddr_t)va, cpus_to_shootdown); 2092843e1988Sjohnlev #else 2093ae115bc7Smrj xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, X_CALL_HIPRI, 2094ae115bc7Smrj cpus_to_shootdown, hati_demap_func); 2095843e1988Sjohnlev #endif 2096ae115bc7Smrj 2097ae115bc7Smrj } 20987c478bd9Sstevel@tonic-gate kpreempt_enable(); 20997c478bd9Sstevel@tonic-gate } 21007c478bd9Sstevel@tonic-gate 21017c478bd9Sstevel@tonic-gate /* 21027c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 21037c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 21047c478bd9Sstevel@tonic-gate * handle releasing of the htables. 21057c478bd9Sstevel@tonic-gate */ 21067c478bd9Sstevel@tonic-gate void 21077c478bd9Sstevel@tonic-gate hat_pte_unmap( 21087c478bd9Sstevel@tonic-gate htable_t *ht, 21097c478bd9Sstevel@tonic-gate uint_t entry, 21107c478bd9Sstevel@tonic-gate uint_t flags, 21117c478bd9Sstevel@tonic-gate x86pte_t old_pte, 21127c478bd9Sstevel@tonic-gate void *pte_ptr) 21137c478bd9Sstevel@tonic-gate { 21147c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 21157c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 21167c478bd9Sstevel@tonic-gate page_t *pp = NULL; 21177c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 21187c478bd9Sstevel@tonic-gate pfn_t pfn; 21197c478bd9Sstevel@tonic-gate 21207c478bd9Sstevel@tonic-gate /* 21217c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 21227c478bd9Sstevel@tonic-gate */ 21237c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 21247c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 21257c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 21267c478bd9Sstevel@tonic-gate } 21277c478bd9Sstevel@tonic-gate 21287c478bd9Sstevel@tonic-gate /* 21297c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 21307c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 21317c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 21327c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 21337c478bd9Sstevel@tonic-gate */ 21347c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 21357c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 21367c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 2137ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST) { 21387c478bd9Sstevel@tonic-gate pp = NULL; 21397c478bd9Sstevel@tonic-gate } else { 2140843e1988Sjohnlev #ifdef __xpv 2141843e1988Sjohnlev if (pfn == PFN_INVALID) 2142843e1988Sjohnlev panic("Invalid PFN, but not PT_NOCONSIST"); 2143843e1988Sjohnlev #endif 21447c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 2145aa2ed9e5Sjosephb if (pp == NULL) { 2146aa2ed9e5Sjosephb panic("no page_t, not NOCONSIST: old_pte=" 2147aa2ed9e5Sjosephb FMT_PTE " ht=%lx entry=0x%x pte_ptr=%lx", 2148aa2ed9e5Sjosephb old_pte, (uintptr_t)ht, entry, 2149aa2ed9e5Sjosephb (uintptr_t)pte_ptr); 2150aa2ed9e5Sjosephb } 21517c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 21527c478bd9Sstevel@tonic-gate } 2153aa2ed9e5Sjosephb 2154aa2ed9e5Sjosephb /* 2155aa2ed9e5Sjosephb * If freeing the address space, check that the PTE 2156aa2ed9e5Sjosephb * hasn't changed, as the mappings are no longer in use by 2157aa2ed9e5Sjosephb * any thread, invalidation is unnecessary. 2158aa2ed9e5Sjosephb * If not freeing, do a full invalidate. 2159843e1988Sjohnlev * 2160843e1988Sjohnlev * On the hypervisor we must always remove mappings, as a 2161843e1988Sjohnlev * writable mapping left behind could cause a page table 2162843e1988Sjohnlev * allocation to fail. 2163aa2ed9e5Sjosephb */ 2164843e1988Sjohnlev #if !defined(__xpv) 2165aa2ed9e5Sjosephb if (hat->hat_flags & HAT_FREEING) 2166aa2ed9e5Sjosephb old_pte = x86pte_get(ht, entry); 2167aa2ed9e5Sjosephb else 2168843e1988Sjohnlev #endif 2169ae115bc7Smrj old_pte = x86pte_inval(ht, entry, old_pte, pte_ptr); 21707c478bd9Sstevel@tonic-gate 21717c478bd9Sstevel@tonic-gate /* 21727c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 21737c478bd9Sstevel@tonic-gate */ 21747c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 21757c478bd9Sstevel@tonic-gate break; 21767c478bd9Sstevel@tonic-gate 21777c478bd9Sstevel@tonic-gate /* 21787c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 21797c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 21807c478bd9Sstevel@tonic-gate */ 21817c478bd9Sstevel@tonic-gate if (pp != NULL) { 21827c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 21837c478bd9Sstevel@tonic-gate pp = NULL; 21847c478bd9Sstevel@tonic-gate } else { 2185ae115bc7Smrj ASSERT(PTE_GET(old_pte, PT_SOFTWARE) >= PT_NOCONSIST); 21867c478bd9Sstevel@tonic-gate } 21877c478bd9Sstevel@tonic-gate } 21887c478bd9Sstevel@tonic-gate 21897c478bd9Sstevel@tonic-gate /* 21907c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 21917c478bd9Sstevel@tonic-gate */ 21927c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 21937c478bd9Sstevel@tonic-gate if (pp != NULL) 21947c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 21957c478bd9Sstevel@tonic-gate return; 21967c478bd9Sstevel@tonic-gate } 21977c478bd9Sstevel@tonic-gate 21987c478bd9Sstevel@tonic-gate /* 21997c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 22007c478bd9Sstevel@tonic-gate */ 22017c478bd9Sstevel@tonic-gate if (pp != NULL) { 22027c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 22037c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 22047c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 22057c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 22067c478bd9Sstevel@tonic-gate if (hm != NULL) 22077c478bd9Sstevel@tonic-gate hment_free(hm); 22087c478bd9Sstevel@tonic-gate } 22097c478bd9Sstevel@tonic-gate 22107c478bd9Sstevel@tonic-gate /* 22117c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 22127c478bd9Sstevel@tonic-gate */ 22137c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 22147c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 22157c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 22167c478bd9Sstevel@tonic-gate } 22177c478bd9Sstevel@tonic-gate 22187c478bd9Sstevel@tonic-gate /* 22197c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 22207c478bd9Sstevel@tonic-gate */ 22217c478bd9Sstevel@tonic-gate static void 22227c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 22237c478bd9Sstevel@tonic-gate { 22247c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 22257c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2226ae115bc7Smrj pgcnt_t pg_index; 22277c478bd9Sstevel@tonic-gate htable_t *ht; 22287c478bd9Sstevel@tonic-gate uint_t entry; 2229ae115bc7Smrj x86pte_t *pte_ptr; 22307c478bd9Sstevel@tonic-gate x86pte_t old_pte; 22317c478bd9Sstevel@tonic-gate 22327c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 22337c478bd9Sstevel@tonic-gate /* 22347c478bd9Sstevel@tonic-gate * Get the PTE 22357c478bd9Sstevel@tonic-gate */ 2236ae115bc7Smrj pg_index = mmu_btop(va - mmu.kmap_addr); 2237ae115bc7Smrj pte_ptr = PT_INDEX_PTR(mmu.kmap_ptes, pg_index); 2238ae115bc7Smrj old_pte = GET_PTE(pte_ptr); 22397c478bd9Sstevel@tonic-gate 22407c478bd9Sstevel@tonic-gate /* 22417c478bd9Sstevel@tonic-gate * get the htable / entry 22427c478bd9Sstevel@tonic-gate */ 22437c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 22447c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 22457c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 22467c478bd9Sstevel@tonic-gate 22477c478bd9Sstevel@tonic-gate /* 22487c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 22497c478bd9Sstevel@tonic-gate */ 22507c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr); 22517c478bd9Sstevel@tonic-gate } 22527c478bd9Sstevel@tonic-gate } 22537c478bd9Sstevel@tonic-gate 22547c478bd9Sstevel@tonic-gate 22557c478bd9Sstevel@tonic-gate /* 22567c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 22577c478bd9Sstevel@tonic-gate */ 22587c478bd9Sstevel@tonic-gate void 22597c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 22607c478bd9Sstevel@tonic-gate { 22617c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2262ae115bc7Smrj 2263843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2264ae115bc7Smrj ASSERT(hat == kas.a_hat || va + len <= _userlimit); 22657c478bd9Sstevel@tonic-gate 22667c478bd9Sstevel@tonic-gate /* 22677c478bd9Sstevel@tonic-gate * special case for performance. 22687c478bd9Sstevel@tonic-gate */ 22697c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 22707c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 22717c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2272ae115bc7Smrj } else { 2273ae115bc7Smrj hat_unload_callback(hat, addr, len, flags, NULL); 22747c478bd9Sstevel@tonic-gate } 2275843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 22767c478bd9Sstevel@tonic-gate } 22777c478bd9Sstevel@tonic-gate 22787c478bd9Sstevel@tonic-gate /* 22797c478bd9Sstevel@tonic-gate * Do the callbacks for ranges being unloaded. 22807c478bd9Sstevel@tonic-gate */ 22817c478bd9Sstevel@tonic-gate typedef struct range_info { 22827c478bd9Sstevel@tonic-gate uintptr_t rng_va; 22837c478bd9Sstevel@tonic-gate ulong_t rng_cnt; 22847c478bd9Sstevel@tonic-gate level_t rng_level; 22857c478bd9Sstevel@tonic-gate } range_info_t; 22867c478bd9Sstevel@tonic-gate 22877c478bd9Sstevel@tonic-gate static void 22887c478bd9Sstevel@tonic-gate handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range) 22897c478bd9Sstevel@tonic-gate { 22907c478bd9Sstevel@tonic-gate /* 22917c478bd9Sstevel@tonic-gate * do callbacks to upper level VM system 22927c478bd9Sstevel@tonic-gate */ 22937c478bd9Sstevel@tonic-gate while (cb != NULL && cnt > 0) { 22947c478bd9Sstevel@tonic-gate --cnt; 22957c478bd9Sstevel@tonic-gate cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 22967c478bd9Sstevel@tonic-gate cb->hcb_end_addr = cb->hcb_start_addr; 22977c478bd9Sstevel@tonic-gate cb->hcb_end_addr += 22987c478bd9Sstevel@tonic-gate range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level); 22997c478bd9Sstevel@tonic-gate cb->hcb_function(cb); 23007c478bd9Sstevel@tonic-gate } 23017c478bd9Sstevel@tonic-gate } 23027c478bd9Sstevel@tonic-gate 23037c478bd9Sstevel@tonic-gate /* 23047c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 23057c478bd9Sstevel@tonic-gate * 23067c478bd9Sstevel@tonic-gate * Flags: 23077c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 23087c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 23097c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 23107c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 23117c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 23127c478bd9Sstevel@tonic-gate */ 23137c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 23147c478bd9Sstevel@tonic-gate void 23157c478bd9Sstevel@tonic-gate hat_unload_callback( 23167c478bd9Sstevel@tonic-gate hat_t *hat, 23177c478bd9Sstevel@tonic-gate caddr_t addr, 23187c478bd9Sstevel@tonic-gate size_t len, 23197c478bd9Sstevel@tonic-gate uint_t flags, 23207c478bd9Sstevel@tonic-gate hat_callback_t *cb) 23217c478bd9Sstevel@tonic-gate { 23227c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 23237c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 23247c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 23257c478bd9Sstevel@tonic-gate uint_t entry; 2326aa2ed9e5Sjosephb uintptr_t contig_va = (uintptr_t)-1L; 23277c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 23287c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 23297c478bd9Sstevel@tonic-gate x86pte_t old_pte; 23307c478bd9Sstevel@tonic-gate 2331843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 2332ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 23337c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 23347c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 23357c478bd9Sstevel@tonic-gate 2336ae115bc7Smrj /* 2337ae115bc7Smrj * Special case a single page being unloaded for speed. This happens 2338ae115bc7Smrj * quite frequently, COW faults after a fork() for example. 2339ae115bc7Smrj */ 2340ae115bc7Smrj if (cb == NULL && len == MMU_PAGESIZE) { 2341ae115bc7Smrj ht = htable_getpte(hat, vaddr, &entry, &old_pte, 0); 2342ae115bc7Smrj if (ht != NULL) { 2343ae115bc7Smrj if (PTE_ISVALID(old_pte)) 2344ae115bc7Smrj hat_pte_unmap(ht, entry, flags, old_pte, NULL); 2345ae115bc7Smrj htable_release(ht); 2346ae115bc7Smrj } 2347843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2348ae115bc7Smrj return; 2349ae115bc7Smrj } 2350ae115bc7Smrj 23517c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 23527c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 23537c478bd9Sstevel@tonic-gate if (ht == NULL) 23547c478bd9Sstevel@tonic-gate break; 23557c478bd9Sstevel@tonic-gate 23567c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 23577c478bd9Sstevel@tonic-gate 23587c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 23597c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 23607c478bd9Sstevel@tonic-gate 23617c478bd9Sstevel@tonic-gate /* 23627c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 23637c478bd9Sstevel@tonic-gate */ 2364aa2ed9e5Sjosephb if (vaddr != contig_va || 23657c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 23667c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 23677c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 23687c478bd9Sstevel@tonic-gate r_cnt = 0; 23697c478bd9Sstevel@tonic-gate } 23707c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 23717c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 23727c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 23737c478bd9Sstevel@tonic-gate ++r_cnt; 23747c478bd9Sstevel@tonic-gate } 23757c478bd9Sstevel@tonic-gate 23767c478bd9Sstevel@tonic-gate /* 23777c478bd9Sstevel@tonic-gate * Unload one mapping from the page tables. 23787c478bd9Sstevel@tonic-gate */ 23797c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 23807c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, NULL); 23817c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 23827c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2383aa2ed9e5Sjosephb contig_va = vaddr; 23847c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 23857c478bd9Sstevel@tonic-gate } 23867c478bd9Sstevel@tonic-gate if (ht) 23877c478bd9Sstevel@tonic-gate htable_release(ht); 23887c478bd9Sstevel@tonic-gate 23897c478bd9Sstevel@tonic-gate /* 23907c478bd9Sstevel@tonic-gate * handle last range for callbacks 23917c478bd9Sstevel@tonic-gate */ 23927c478bd9Sstevel@tonic-gate if (r_cnt > 0) 23937c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 2394843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 23957c478bd9Sstevel@tonic-gate } 23967c478bd9Sstevel@tonic-gate 23977c478bd9Sstevel@tonic-gate /* 23987c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 23997c478bd9Sstevel@tonic-gate * 24007c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 24017c478bd9Sstevel@tonic-gate * driver. 24027c478bd9Sstevel@tonic-gate */ 24037c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24047c478bd9Sstevel@tonic-gate void 24057c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 24067c478bd9Sstevel@tonic-gate { 24077c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 24087c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 24097c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 24107c478bd9Sstevel@tonic-gate uint_t entry; 24117c478bd9Sstevel@tonic-gate x86pte_t pte; 24127c478bd9Sstevel@tonic-gate x86pte_t save_pte; 24137c478bd9Sstevel@tonic-gate x86pte_t new; 24147c478bd9Sstevel@tonic-gate page_t *pp; 24157c478bd9Sstevel@tonic-gate 24167c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 24177c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 24187c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2419ae115bc7Smrj ASSERT(hat == kas.a_hat || eaddr <= _userlimit); 24207c478bd9Sstevel@tonic-gate 2421843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 24227c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 24237c478bd9Sstevel@tonic-gate try_again: 24247c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 24257c478bd9Sstevel@tonic-gate if (ht == NULL) 24267c478bd9Sstevel@tonic-gate break; 24277c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 24287c478bd9Sstevel@tonic-gate 2429ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 24307c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 24317c478bd9Sstevel@tonic-gate continue; 24327c478bd9Sstevel@tonic-gate 24337c478bd9Sstevel@tonic-gate /* 24347c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 24357c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 24367c478bd9Sstevel@tonic-gate */ 24377c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 24387c478bd9Sstevel@tonic-gate if (pp == NULL) 24397c478bd9Sstevel@tonic-gate break; 24407c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 24417c478bd9Sstevel@tonic-gate save_pte = pte; 24427c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 24437c478bd9Sstevel@tonic-gate if (pte != save_pte) { 24447c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24457c478bd9Sstevel@tonic-gate goto try_again; 24467c478bd9Sstevel@tonic-gate } 2447ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC || 24487c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 24497c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24507c478bd9Sstevel@tonic-gate continue; 24517c478bd9Sstevel@tonic-gate } 24527c478bd9Sstevel@tonic-gate 24537c478bd9Sstevel@tonic-gate /* 24547c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 24557c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 24567c478bd9Sstevel@tonic-gate */ 24577c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 24587c478bd9Sstevel@tonic-gate new = pte; 24597c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 24607c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 24617c478bd9Sstevel@tonic-gate if (pte != 0) { 24627c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24637c478bd9Sstevel@tonic-gate goto try_again; 24647c478bd9Sstevel@tonic-gate } 24657c478bd9Sstevel@tonic-gate } else { 24667c478bd9Sstevel@tonic-gate /* 24677c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 24687c478bd9Sstevel@tonic-gate */ 24697c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 24707c478bd9Sstevel@tonic-gate } 24717c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 24727c478bd9Sstevel@tonic-gate } 24737c478bd9Sstevel@tonic-gate if (ht) 24747c478bd9Sstevel@tonic-gate htable_release(ht); 2475843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 24767c478bd9Sstevel@tonic-gate } 24777c478bd9Sstevel@tonic-gate 24787c478bd9Sstevel@tonic-gate /* 24797c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 24807c478bd9Sstevel@tonic-gate */ 24817c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 24827c478bd9Sstevel@tonic-gate void 24837c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 24847c478bd9Sstevel@tonic-gate { 24857c478bd9Sstevel@tonic-gate /* does nothing */ 24867c478bd9Sstevel@tonic-gate } 24877c478bd9Sstevel@tonic-gate 24887c478bd9Sstevel@tonic-gate /* 24897c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 24907c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 24917c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 24927c478bd9Sstevel@tonic-gate * *attr is not valid. 24937c478bd9Sstevel@tonic-gate */ 24947c478bd9Sstevel@tonic-gate uint_t 24957c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 24967c478bd9Sstevel@tonic-gate { 24977c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 24987c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 24997c478bd9Sstevel@tonic-gate x86pte_t pte; 25007c478bd9Sstevel@tonic-gate 2501ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 25027c478bd9Sstevel@tonic-gate 25037c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 25047c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25057c478bd9Sstevel@tonic-gate 2506ae115bc7Smrj ht = htable_getpte(hat, vaddr, NULL, &pte, mmu.max_page_level); 25077c478bd9Sstevel@tonic-gate if (ht == NULL) 25087c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25097c478bd9Sstevel@tonic-gate 25107c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 25117c478bd9Sstevel@tonic-gate htable_release(ht); 25127c478bd9Sstevel@tonic-gate return ((uint_t)-1); 25137c478bd9Sstevel@tonic-gate } 25147c478bd9Sstevel@tonic-gate 25157c478bd9Sstevel@tonic-gate *attr = PROT_READ; 25167c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 25177c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 25187c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 25197c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 25207c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 25217c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2522ae115bc7Smrj if (PTE_GET(pte, PT_SOFTWARE) >= PT_NOSYNC) 25237c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 25247c478bd9Sstevel@tonic-gate htable_release(ht); 25257c478bd9Sstevel@tonic-gate return (0); 25267c478bd9Sstevel@tonic-gate } 25277c478bd9Sstevel@tonic-gate 25287c478bd9Sstevel@tonic-gate /* 25297c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 25307c478bd9Sstevel@tonic-gate */ 25317c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 25327c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 25337c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 25347c478bd9Sstevel@tonic-gate 25357c478bd9Sstevel@tonic-gate static void 25367c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 25377c478bd9Sstevel@tonic-gate { 25387c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 25397c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 25407c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 25417c478bd9Sstevel@tonic-gate uint_t entry; 25427c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 25437c478bd9Sstevel@tonic-gate page_t *pp; 25447c478bd9Sstevel@tonic-gate 2545843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 25467c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 25477c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 25487c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 25497c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 25507c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 25517c478bd9Sstevel@tonic-gate try_again: 25527c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 25537c478bd9Sstevel@tonic-gate if (ht == NULL) 25547c478bd9Sstevel@tonic-gate break; 2555ae115bc7Smrj if (PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOCONSIST) 25567c478bd9Sstevel@tonic-gate continue; 25577c478bd9Sstevel@tonic-gate 25587c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 25597c478bd9Sstevel@tonic-gate if (pp == NULL) 25607c478bd9Sstevel@tonic-gate continue; 25617c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 25627c478bd9Sstevel@tonic-gate 25637c478bd9Sstevel@tonic-gate newpte = oldpte; 25647c478bd9Sstevel@tonic-gate /* 25657c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 25667c478bd9Sstevel@tonic-gate * figure out the new attributes. 25677c478bd9Sstevel@tonic-gate */ 25687c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 25697c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 25707c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 25717c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 25727c478bd9Sstevel@tonic-gate 2573ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2574ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) < PT_NOSYNC) 25757c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 25767c478bd9Sstevel@tonic-gate 25777c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 25787c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 25797c478bd9Sstevel@tonic-gate } 25807c478bd9Sstevel@tonic-gate 25817c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 25827c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 25837c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 25847c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 25857c478bd9Sstevel@tonic-gate 2586ae115bc7Smrj if (!(attr & HAT_NOSYNC) && 2587ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2588ae115bc7Smrj newpte &= ~PT_SOFTWARE; 25897c478bd9Sstevel@tonic-gate 25907c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 25917c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 25927c478bd9Sstevel@tonic-gate } 25937c478bd9Sstevel@tonic-gate 25947c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 25957c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 25967c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 25977c478bd9Sstevel@tonic-gate 2598ae115bc7Smrj if ((attr & HAT_NOSYNC) && 2599ae115bc7Smrj PTE_GET(oldpte, PT_SOFTWARE) >= PT_NOSYNC) 2600ae115bc7Smrj newpte &= ~PT_SOFTWARE; 26017c478bd9Sstevel@tonic-gate 26027c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 26037c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 26047c478bd9Sstevel@tonic-gate } 26057c478bd9Sstevel@tonic-gate 2606ae115bc7Smrj /* 2607ae115bc7Smrj * Ensure NOSYNC/NOCONSIST mappings have REF and MOD set. 2608ae115bc7Smrj * x86pte_set() depends on this. 2609ae115bc7Smrj */ 2610ae115bc7Smrj if (PTE_GET(newpte, PT_SOFTWARE) >= PT_NOSYNC) 2611ae115bc7Smrj newpte |= PT_REF | PT_MOD; 2612ae115bc7Smrj 26137c478bd9Sstevel@tonic-gate /* 26147c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 26157c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 26167c478bd9Sstevel@tonic-gate */ 26177c478bd9Sstevel@tonic-gate 26187c478bd9Sstevel@tonic-gate /* 26197c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 26207c478bd9Sstevel@tonic-gate */ 26217c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 26227c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 26237c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 26247c478bd9Sstevel@tonic-gate if (oldpte != 0) { 26257c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 26267c478bd9Sstevel@tonic-gate goto try_again; 26277c478bd9Sstevel@tonic-gate } 26287c478bd9Sstevel@tonic-gate } 26297c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 26307c478bd9Sstevel@tonic-gate } 26317c478bd9Sstevel@tonic-gate if (ht) 26327c478bd9Sstevel@tonic-gate htable_release(ht); 2633843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 26347c478bd9Sstevel@tonic-gate } 26357c478bd9Sstevel@tonic-gate 26367c478bd9Sstevel@tonic-gate /* 26377c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 26387c478bd9Sstevel@tonic-gate */ 26397c478bd9Sstevel@tonic-gate void 26407c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26417c478bd9Sstevel@tonic-gate { 2642ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26437c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 26447c478bd9Sstevel@tonic-gate } 26457c478bd9Sstevel@tonic-gate 26467c478bd9Sstevel@tonic-gate void 26477c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26487c478bd9Sstevel@tonic-gate { 2649ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26507c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 26517c478bd9Sstevel@tonic-gate } 26527c478bd9Sstevel@tonic-gate 26537c478bd9Sstevel@tonic-gate void 26547c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 26557c478bd9Sstevel@tonic-gate { 2656ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26577c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 26587c478bd9Sstevel@tonic-gate } 26597c478bd9Sstevel@tonic-gate 26607c478bd9Sstevel@tonic-gate void 26617c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 26627c478bd9Sstevel@tonic-gate { 2663ae115bc7Smrj ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= _userlimit); 26647c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 26657c478bd9Sstevel@tonic-gate } 26667c478bd9Sstevel@tonic-gate 26677c478bd9Sstevel@tonic-gate /* 26687c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 26697c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 26707c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 26717c478bd9Sstevel@tonic-gate */ 26727c478bd9Sstevel@tonic-gate ssize_t 26737c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 26747c478bd9Sstevel@tonic-gate { 26757c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 26767c478bd9Sstevel@tonic-gate htable_t *ht; 26777c478bd9Sstevel@tonic-gate size_t pagesize; 26787c478bd9Sstevel@tonic-gate 2679ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 26807c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 26817c478bd9Sstevel@tonic-gate return (-1); 26827c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 26837c478bd9Sstevel@tonic-gate if (ht == NULL) 26847c478bd9Sstevel@tonic-gate return (-1); 26857c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 26867c478bd9Sstevel@tonic-gate htable_release(ht); 26877c478bd9Sstevel@tonic-gate return (pagesize); 26887c478bd9Sstevel@tonic-gate } 26897c478bd9Sstevel@tonic-gate 26907c478bd9Sstevel@tonic-gate 26917c478bd9Sstevel@tonic-gate 26927c478bd9Sstevel@tonic-gate /* 26937c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 26947c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 26957c478bd9Sstevel@tonic-gate */ 26967c478bd9Sstevel@tonic-gate pfn_t 26977c478bd9Sstevel@tonic-gate hat_getpfnum(hat_t *hat, caddr_t addr) 26987c478bd9Sstevel@tonic-gate { 26997c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 27007c478bd9Sstevel@tonic-gate htable_t *ht; 27017c478bd9Sstevel@tonic-gate uint_t entry; 27027c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 27037c478bd9Sstevel@tonic-gate 2704ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27057c478bd9Sstevel@tonic-gate if (khat_running == 0) 2706ae115bc7Smrj return (PFN_INVALID); 27077c478bd9Sstevel@tonic-gate 27087c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 27097c478bd9Sstevel@tonic-gate return (PFN_INVALID); 27107c478bd9Sstevel@tonic-gate 2711843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 27127c478bd9Sstevel@tonic-gate /* 27137c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 27147c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 27157c478bd9Sstevel@tonic-gate * this up. 27167c478bd9Sstevel@tonic-gate */ 27177c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 27187c478bd9Sstevel@tonic-gate x86pte_t pte; 2719ae115bc7Smrj pgcnt_t pg_index; 27207c478bd9Sstevel@tonic-gate 2721ae115bc7Smrj pg_index = mmu_btop(vaddr - mmu.kmap_addr); 2722ae115bc7Smrj pte = GET_PTE(PT_INDEX_PTR(mmu.kmap_ptes, pg_index)); 2723843e1988Sjohnlev if (PTE_ISVALID(pte)) 2724843e1988Sjohnlev /*LINTED [use of constant 0 causes a lint warning] */ 2725843e1988Sjohnlev pfn = PTE2PFN(pte, 0); 2726843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 2727843e1988Sjohnlev return (pfn); 27287c478bd9Sstevel@tonic-gate } 27297c478bd9Sstevel@tonic-gate 27307c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 2731843e1988Sjohnlev if (ht == NULL) { 2732843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27337c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2734843e1988Sjohnlev } 27357c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 27367c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 27377c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 27387c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 27397c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 27407c478bd9Sstevel@tonic-gate htable_release(ht); 2741843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27427c478bd9Sstevel@tonic-gate return (pfn); 27437c478bd9Sstevel@tonic-gate } 27447c478bd9Sstevel@tonic-gate 27457c478bd9Sstevel@tonic-gate /* 27467c478bd9Sstevel@tonic-gate * hat_getkpfnum() is an obsolete DDI routine, and its use is discouraged. 27477c478bd9Sstevel@tonic-gate * Use hat_getpfnum(kas.a_hat, ...) instead. 27487c478bd9Sstevel@tonic-gate * 27497c478bd9Sstevel@tonic-gate * We'd like to return PFN_INVALID if the mappings have underlying page_t's 27507c478bd9Sstevel@tonic-gate * but can't right now due to the fact that some software has grown to use 27517c478bd9Sstevel@tonic-gate * this interface incorrectly. So for now when the interface is misused, 27527c478bd9Sstevel@tonic-gate * return a warning to the user that in the future it won't work in the 27537c478bd9Sstevel@tonic-gate * way they're abusing it, and carry on. 27547c478bd9Sstevel@tonic-gate * 27557c478bd9Sstevel@tonic-gate * Note that hat_getkpfnum() is never supported on amd64. 27567c478bd9Sstevel@tonic-gate */ 27577c478bd9Sstevel@tonic-gate #if !defined(__amd64) 27587c478bd9Sstevel@tonic-gate pfn_t 27597c478bd9Sstevel@tonic-gate hat_getkpfnum(caddr_t addr) 27607c478bd9Sstevel@tonic-gate { 27617c478bd9Sstevel@tonic-gate pfn_t pfn; 27627c478bd9Sstevel@tonic-gate int badcaller = 0; 27637c478bd9Sstevel@tonic-gate 27647c478bd9Sstevel@tonic-gate if (khat_running == 0) 27657c478bd9Sstevel@tonic-gate panic("hat_getkpfnum(): called too early\n"); 27667c478bd9Sstevel@tonic-gate if ((uintptr_t)addr < kernelbase) 27677c478bd9Sstevel@tonic-gate return (PFN_INVALID); 27687c478bd9Sstevel@tonic-gate 2769843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 27707c478bd9Sstevel@tonic-gate if (segkpm && IS_KPM_ADDR(addr)) { 27717c478bd9Sstevel@tonic-gate badcaller = 1; 27727c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(addr); 27737c478bd9Sstevel@tonic-gate } else { 27747c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, addr); 27757c478bd9Sstevel@tonic-gate badcaller = pf_is_memory(pfn); 27767c478bd9Sstevel@tonic-gate } 27777c478bd9Sstevel@tonic-gate 27787c478bd9Sstevel@tonic-gate if (badcaller) 27797c478bd9Sstevel@tonic-gate hat_getkpfnum_badcall(caller()); 2780843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 27817c478bd9Sstevel@tonic-gate return (pfn); 27827c478bd9Sstevel@tonic-gate } 27837c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 27847c478bd9Sstevel@tonic-gate 27857c478bd9Sstevel@tonic-gate /* 27867c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 27877c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 27887c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 27897c478bd9Sstevel@tonic-gate */ 27907c478bd9Sstevel@tonic-gate int 27917c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 27927c478bd9Sstevel@tonic-gate { 27937c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 27947c478bd9Sstevel@tonic-gate uint_t entry; 27957c478bd9Sstevel@tonic-gate htable_t *ht; 27967c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 27977c478bd9Sstevel@tonic-gate 2798ae115bc7Smrj ASSERT(hat == kas.a_hat || vaddr <= _userlimit); 27997c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 28007c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 28017c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 28027c478bd9Sstevel@tonic-gate return (0); 28037c478bd9Sstevel@tonic-gate 28047c478bd9Sstevel@tonic-gate /* 28057c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 28067c478bd9Sstevel@tonic-gate * for performance. 28077c478bd9Sstevel@tonic-gate */ 28087c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 28097c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 28107c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 28117c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 28127c478bd9Sstevel@tonic-gate else 28137c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 28147c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 28157c478bd9Sstevel@tonic-gate } 28167c478bd9Sstevel@tonic-gate 28177c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 28187c478bd9Sstevel@tonic-gate htable_release(ht); 2819843e1988Sjohnlev return (ht != NULL); 28207c478bd9Sstevel@tonic-gate } 28217c478bd9Sstevel@tonic-gate 28227c478bd9Sstevel@tonic-gate /* 2823250b7ff9Sjosephb * Find out if the segment for hat_share()/hat_unshare() is DISM or locked ISM. 2824250b7ff9Sjosephb */ 2825250b7ff9Sjosephb static int 2826250b7ff9Sjosephb is_it_dism(hat_t *hat, caddr_t va) 2827250b7ff9Sjosephb { 2828250b7ff9Sjosephb struct seg *seg; 2829250b7ff9Sjosephb struct shm_data *shmd; 2830250b7ff9Sjosephb struct spt_data *sptd; 2831250b7ff9Sjosephb 2832250b7ff9Sjosephb seg = as_findseg(hat->hat_as, va, 0); 2833250b7ff9Sjosephb ASSERT(seg != NULL); 2834250b7ff9Sjosephb ASSERT(seg->s_base <= va); 2835250b7ff9Sjosephb shmd = (struct shm_data *)seg->s_data; 2836250b7ff9Sjosephb ASSERT(shmd != NULL); 2837250b7ff9Sjosephb sptd = (struct spt_data *)shmd->shm_sptseg->s_data; 2838250b7ff9Sjosephb ASSERT(sptd != NULL); 2839250b7ff9Sjosephb if (sptd->spt_flags & SHM_PAGEABLE) 2840250b7ff9Sjosephb return (1); 2841250b7ff9Sjosephb return (0); 2842250b7ff9Sjosephb } 2843250b7ff9Sjosephb 2844250b7ff9Sjosephb /* 2845250b7ff9Sjosephb * Simple implementation of ISM. hat_share() is similar to hat_memload_array(), 28467c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 2847250b7ff9Sjosephb * and protections to use for this hat. If we find a full properly aligned 2848250b7ff9Sjosephb * and sized pagetable, we will attempt to share the pagetable itself. 28497c478bd9Sstevel@tonic-gate */ 28507c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 28517c478bd9Sstevel@tonic-gate int 28527c478bd9Sstevel@tonic-gate hat_share( 28537c478bd9Sstevel@tonic-gate hat_t *hat, 28547c478bd9Sstevel@tonic-gate caddr_t addr, 28557c478bd9Sstevel@tonic-gate hat_t *ism_hat, 28567c478bd9Sstevel@tonic-gate caddr_t src_addr, 28577c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 28587c478bd9Sstevel@tonic-gate uint_t ismszc) 28597c478bd9Sstevel@tonic-gate { 28607c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 28617c478bd9Sstevel@tonic-gate uintptr_t vaddr; 28627c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 28637c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 28647c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 28657c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 28667c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 28677c478bd9Sstevel@tonic-gate htable_t *ht; 28687c478bd9Sstevel@tonic-gate x86pte_t pte; 28697c478bd9Sstevel@tonic-gate page_t *pp; 28707c478bd9Sstevel@tonic-gate pfn_t pfn; 28717c478bd9Sstevel@tonic-gate level_t l; 28727c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 28737c478bd9Sstevel@tonic-gate uint_t prot; 2874250b7ff9Sjosephb int is_dism; 2875250b7ff9Sjosephb int flags; 28767c478bd9Sstevel@tonic-gate 28777c478bd9Sstevel@tonic-gate /* 28787c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 28797c478bd9Sstevel@tonic-gate */ 28807c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2881ae115bc7Smrj ASSERT(eaddr <= _userlimit); 28827c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 28837c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 28847c478bd9Sstevel@tonic-gate return (0); 28857c478bd9Sstevel@tonic-gate } 2886843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 28877c478bd9Sstevel@tonic-gate 28887c478bd9Sstevel@tonic-gate /* 28897c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 28907c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 28917c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 28927c478bd9Sstevel@tonic-gate */ 28937c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 28947c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 28957c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 2896250b7ff9Sjosephb is_dism = is_it_dism(hat, addr); 28977c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 28987c478bd9Sstevel@tonic-gate /* 28997c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 29007c478bd9Sstevel@tonic-gate */ 29017c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 29027c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 29037c478bd9Sstevel@tonic-gate break; 29047c478bd9Sstevel@tonic-gate 29057c478bd9Sstevel@tonic-gate /* 2906250b7ff9Sjosephb * First check to see if we already share the page table. 29077c478bd9Sstevel@tonic-gate */ 2908250b7ff9Sjosephb l = ism_ht->ht_level; 29097c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 2910250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 2911250b7ff9Sjosephb if (ht != NULL) { 2912250b7ff9Sjosephb if (ht->ht_flags & HTABLE_SHARED_PFN) 2913250b7ff9Sjosephb goto shared; 2914250b7ff9Sjosephb htable_release(ht); 2915250b7ff9Sjosephb goto not_shared; 29167c478bd9Sstevel@tonic-gate } 29177c478bd9Sstevel@tonic-gate 29187c478bd9Sstevel@tonic-gate /* 2919250b7ff9Sjosephb * Can't ever share top table. 29207c478bd9Sstevel@tonic-gate */ 2921250b7ff9Sjosephb if (l == mmu.max_level) 2922250b7ff9Sjosephb goto not_shared; 29237c478bd9Sstevel@tonic-gate 2924250b7ff9Sjosephb /* 2925250b7ff9Sjosephb * Avoid level mismatches later due to DISM faults. 2926250b7ff9Sjosephb */ 2927250b7ff9Sjosephb if (is_dism && l > 0) 2928250b7ff9Sjosephb goto not_shared; 29297c478bd9Sstevel@tonic-gate 2930250b7ff9Sjosephb /* 2931250b7ff9Sjosephb * addresses and lengths must align 2932250b7ff9Sjosephb * table must be fully populated 2933250b7ff9Sjosephb * no lower level page tables 2934250b7ff9Sjosephb */ 2935250b7ff9Sjosephb if (ism_addr != ism_ht->ht_vaddr || 2936250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l + 1)) != 0) 2937250b7ff9Sjosephb goto not_shared; 29387c478bd9Sstevel@tonic-gate 2939250b7ff9Sjosephb /* 2940250b7ff9Sjosephb * The range of address space must cover a full table. 2941250b7ff9Sjosephb */ 29421d03c31eSjohnlev if (e_ism_addr - ism_addr < LEVEL_SIZE(l + 1)) 2943250b7ff9Sjosephb goto not_shared; 29447c478bd9Sstevel@tonic-gate 2945250b7ff9Sjosephb /* 2946250b7ff9Sjosephb * All entries in the ISM page table must be leaf PTEs. 2947250b7ff9Sjosephb */ 2948250b7ff9Sjosephb if (l > 0) { 2949250b7ff9Sjosephb int e; 29507c478bd9Sstevel@tonic-gate 2951250b7ff9Sjosephb /* 2952250b7ff9Sjosephb * We know the 0th is from htable_walk() above. 2953250b7ff9Sjosephb */ 2954250b7ff9Sjosephb for (e = 1; e < HTABLE_NUM_PTES(ism_ht); ++e) { 2955250b7ff9Sjosephb x86pte_t pte; 2956250b7ff9Sjosephb pte = x86pte_get(ism_ht, e); 2957250b7ff9Sjosephb if (!PTE_ISPAGE(pte, l)) 2958250b7ff9Sjosephb goto not_shared; 29597c478bd9Sstevel@tonic-gate } 29607c478bd9Sstevel@tonic-gate } 29617c478bd9Sstevel@tonic-gate 2962250b7ff9Sjosephb /* 2963250b7ff9Sjosephb * share the page table 2964250b7ff9Sjosephb */ 2965250b7ff9Sjosephb ht = htable_create(hat, vaddr, l, ism_ht); 2966250b7ff9Sjosephb shared: 2967250b7ff9Sjosephb ASSERT(ht->ht_flags & HTABLE_SHARED_PFN); 2968250b7ff9Sjosephb ASSERT(ht->ht_shares == ism_ht); 2969250b7ff9Sjosephb hat->hat_ism_pgcnt += 2970250b7ff9Sjosephb (ism_ht->ht_valid_cnt - ht->ht_valid_cnt) << 2971250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 2972250b7ff9Sjosephb ht->ht_valid_cnt = ism_ht->ht_valid_cnt; 2973250b7ff9Sjosephb htable_release(ht); 2974250b7ff9Sjosephb ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(l + 1); 2975250b7ff9Sjosephb htable_release(ism_ht); 2976250b7ff9Sjosephb ism_ht = NULL; 2977250b7ff9Sjosephb continue; 2978250b7ff9Sjosephb 2979250b7ff9Sjosephb not_shared: 29807c478bd9Sstevel@tonic-gate /* 29817c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 29827c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 2983250b7ff9Sjosephb * Figure out what level size mappings to use; 2984250b7ff9Sjosephb */ 2985250b7ff9Sjosephb for (l = ism_ht->ht_level; l > 0; --l) { 2986250b7ff9Sjosephb if (LEVEL_SIZE(l) <= eaddr - vaddr && 2987250b7ff9Sjosephb (vaddr & LEVEL_OFFSET(l)) == 0) 2988250b7ff9Sjosephb break; 2989250b7ff9Sjosephb } 2990250b7ff9Sjosephb 2991250b7ff9Sjosephb /* 29927c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 2993250b7ff9Sjosephb * be careful to truncate it if needed. 29947c478bd9Sstevel@tonic-gate */ 29957c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 29967c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 29977c478bd9Sstevel@tonic-gate } else { 29987c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 29997c478bd9Sstevel@tonic-gate l = 0; 30007c478bd9Sstevel@tonic-gate } 30017c478bd9Sstevel@tonic-gate 30027c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 30037c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 30047c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 30057c478bd9Sstevel@tonic-gate /* 30067c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 30077c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 30087c478bd9Sstevel@tonic-gate */ 30097c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 30107c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 30117c478bd9Sstevel@tonic-gate 30127c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 30137c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 30147c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 30157c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 30167c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 30177c478bd9Sstevel@tonic-gate 3018250b7ff9Sjosephb flags = HAT_LOAD; 3019250b7ff9Sjosephb if (!is_dism) 3020250b7ff9Sjosephb flags |= HAT_LOAD_LOCK | HAT_LOAD_NOCONSIST; 3021250b7ff9Sjosephb while (hati_load_common(hat, vaddr, pp, prot, flags, 3022ae115bc7Smrj l, pfn) != 0) { 3023ae115bc7Smrj if (l == 0) 3024ae115bc7Smrj panic("hati_load_common() failure"); 3025ae115bc7Smrj --l; 3026ae115bc7Smrj } 30277c478bd9Sstevel@tonic-gate 30287c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 30297c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 30307c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 30317c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 30327c478bd9Sstevel@tonic-gate } 30337c478bd9Sstevel@tonic-gate } 30347c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 30357c478bd9Sstevel@tonic-gate htable_release(ism_ht); 3036843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 30377c478bd9Sstevel@tonic-gate return (0); 30387c478bd9Sstevel@tonic-gate } 30397c478bd9Sstevel@tonic-gate 30407c478bd9Sstevel@tonic-gate 30417c478bd9Sstevel@tonic-gate /* 30427c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 30437c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 30447c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 30457c478bd9Sstevel@tonic-gate */ 30467c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 30477c478bd9Sstevel@tonic-gate void 30487c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 30497c478bd9Sstevel@tonic-gate { 30507173d045Sjosephb uint64_t vaddr = (uintptr_t)addr; 30517c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 30527c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 30537c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 3054250b7ff9Sjosephb int flags = HAT_UNLOAD_UNMAP; 3055250b7ff9Sjosephb level_t l; 30567c478bd9Sstevel@tonic-gate 30577c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 3058ae115bc7Smrj ASSERT(eaddr <= _userlimit); 30597c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 30607c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 3061843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 30627c478bd9Sstevel@tonic-gate 30637c478bd9Sstevel@tonic-gate /* 30647c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 30657c478bd9Sstevel@tonic-gate * 3066ae115bc7Smrj * Note that it's ok to delay the TLB shootdown till the entire range is 30677c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 3068ae115bc7Smrj * pagetable page, its hat_tlb_inval() will do a global TLB invalidate. 30697c478bd9Sstevel@tonic-gate */ 3070250b7ff9Sjosephb l = mmu.max_page_level; 3071250b7ff9Sjosephb if (l == mmu.max_level) 3072250b7ff9Sjosephb --l; 3073250b7ff9Sjosephb for (; l >= 0; --l) { 3074250b7ff9Sjosephb for (vaddr = (uintptr_t)addr; vaddr < eaddr; 3075250b7ff9Sjosephb vaddr = (vaddr & LEVEL_MASK(l + 1)) + LEVEL_SIZE(l + 1)) { 3076250b7ff9Sjosephb ASSERT(!IN_VA_HOLE(vaddr)); 3077250b7ff9Sjosephb /* 3078250b7ff9Sjosephb * find a pagetable that maps the current address 3079250b7ff9Sjosephb */ 3080250b7ff9Sjosephb ht = htable_lookup(hat, vaddr, l); 3081250b7ff9Sjosephb if (ht == NULL) 3082250b7ff9Sjosephb continue; 30837c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 30847c478bd9Sstevel@tonic-gate /* 3085250b7ff9Sjosephb * clear page count, set valid_cnt to 0, 3086250b7ff9Sjosephb * let htable_release() finish the job 30877c478bd9Sstevel@tonic-gate */ 3088250b7ff9Sjosephb hat->hat_ism_pgcnt -= ht->ht_valid_cnt << 3089250b7ff9Sjosephb (LEVEL_SHIFT(ht->ht_level) - MMU_PAGESHIFT); 30907c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 30917c478bd9Sstevel@tonic-gate need_demaps = 1; 30927c478bd9Sstevel@tonic-gate } 30937c478bd9Sstevel@tonic-gate htable_release(ht); 30947c478bd9Sstevel@tonic-gate } 30957c478bd9Sstevel@tonic-gate } 30967c478bd9Sstevel@tonic-gate 30977c478bd9Sstevel@tonic-gate /* 30987c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 30997c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 31007c478bd9Sstevel@tonic-gate */ 31017c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 3102ae115bc7Smrj hat_tlb_inval(hat, DEMAP_ALL_ADDR); 31037c478bd9Sstevel@tonic-gate 31047c478bd9Sstevel@tonic-gate /* 31057c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 31067c478bd9Sstevel@tonic-gate * couldn't share pagetables. 31077c478bd9Sstevel@tonic-gate */ 3108250b7ff9Sjosephb if (!is_it_dism(hat, addr)) 3109250b7ff9Sjosephb flags |= HAT_UNLOAD_UNLOCK; 3110250b7ff9Sjosephb hat_unload(hat, addr, len, flags); 3111843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 31127c478bd9Sstevel@tonic-gate } 31137c478bd9Sstevel@tonic-gate 31147c478bd9Sstevel@tonic-gate 31157c478bd9Sstevel@tonic-gate /* 31167c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 31177c478bd9Sstevel@tonic-gate */ 31187c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 31197c478bd9Sstevel@tonic-gate void 31207c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 31217c478bd9Sstevel@tonic-gate { 31227c478bd9Sstevel@tonic-gate } 31237c478bd9Sstevel@tonic-gate 31247c478bd9Sstevel@tonic-gate 31257c478bd9Sstevel@tonic-gate /* 31267c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 31277c478bd9Sstevel@tonic-gate * Mostly stolem from hat_pagesync() 31287c478bd9Sstevel@tonic-gate */ 31297c478bd9Sstevel@tonic-gate static void 31307c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 31317c478bd9Sstevel@tonic-gate { 31327c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 31337c478bd9Sstevel@tonic-gate htable_t *ht; 31347c478bd9Sstevel@tonic-gate uint_t entry; 31357c478bd9Sstevel@tonic-gate x86pte_t old; 31367c478bd9Sstevel@tonic-gate x86pte_t new; 31377c478bd9Sstevel@tonic-gate uint_t pszc = 0; 31387c478bd9Sstevel@tonic-gate 3139843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 31407c478bd9Sstevel@tonic-gate next_size: 31417c478bd9Sstevel@tonic-gate /* 31427c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 31437c478bd9Sstevel@tonic-gate */ 31447c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 31457c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 31467c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 31477c478bd9Sstevel@tonic-gate continue; 31487c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 31497c478bd9Sstevel@tonic-gate 31507c478bd9Sstevel@tonic-gate for (;;) { 31517c478bd9Sstevel@tonic-gate /* 31527c478bd9Sstevel@tonic-gate * Is this mapping of interest? 31537c478bd9Sstevel@tonic-gate */ 31547c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 31557c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 31567c478bd9Sstevel@tonic-gate break; 31577c478bd9Sstevel@tonic-gate 31587c478bd9Sstevel@tonic-gate /* 31597c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 31607c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 31617c478bd9Sstevel@tonic-gate */ 31627c478bd9Sstevel@tonic-gate new = old; 31637c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 31647c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 31657c478bd9Sstevel@tonic-gate if (old != 0) 31667c478bd9Sstevel@tonic-gate continue; 31677c478bd9Sstevel@tonic-gate 31687c478bd9Sstevel@tonic-gate break; 31697c478bd9Sstevel@tonic-gate } 31707c478bd9Sstevel@tonic-gate } 31717c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 31727c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 31737c478bd9Sstevel@tonic-gate page_t *tpp; 31747c478bd9Sstevel@tonic-gate pszc++; 31757c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 31767c478bd9Sstevel@tonic-gate if (pp != tpp) { 31777c478bd9Sstevel@tonic-gate pp = tpp; 31787c478bd9Sstevel@tonic-gate goto next_size; 31797c478bd9Sstevel@tonic-gate } 31807c478bd9Sstevel@tonic-gate } 3181843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 31827c478bd9Sstevel@tonic-gate } 31837c478bd9Sstevel@tonic-gate 31847c478bd9Sstevel@tonic-gate /* 31857c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 31867c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 31877c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 31887c478bd9Sstevel@tonic-gate */ 31897c478bd9Sstevel@tonic-gate void 31907c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 31917c478bd9Sstevel@tonic-gate { 31927c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 31937c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 31947c478bd9Sstevel@tonic-gate page_t **listp; 3195d9615970Sqiao int noshuffle; 3196d9615970Sqiao 3197d9615970Sqiao noshuffle = flag & P_NSH; 3198d9615970Sqiao flag &= ~P_NSH; 31997c478bd9Sstevel@tonic-gate 32007c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 32017c478bd9Sstevel@tonic-gate return; 32027c478bd9Sstevel@tonic-gate 3203d9615970Sqiao if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp) && 3204d9615970Sqiao !noshuffle) { 32057c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 32067c478bd9Sstevel@tonic-gate mutex_enter(vphm); 32077c478bd9Sstevel@tonic-gate } 32087c478bd9Sstevel@tonic-gate 32097c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 32107c478bd9Sstevel@tonic-gate 32117c478bd9Sstevel@tonic-gate if (vphm != NULL) { 32127c478bd9Sstevel@tonic-gate 32137c478bd9Sstevel@tonic-gate /* 32147c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 32157c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 32167c478bd9Sstevel@tonic-gate * pp is the only page on the list. 32177c478bd9Sstevel@tonic-gate */ 32187c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 32197c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 32207c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 32217c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 32227c478bd9Sstevel@tonic-gate else 32237c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 32247c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 32257c478bd9Sstevel@tonic-gate } 32267c478bd9Sstevel@tonic-gate mutex_exit(vphm); 32277c478bd9Sstevel@tonic-gate } 32287c478bd9Sstevel@tonic-gate } 32297c478bd9Sstevel@tonic-gate 32307c478bd9Sstevel@tonic-gate void 32317c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 32327c478bd9Sstevel@tonic-gate { 32337c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 32347c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 32357c478bd9Sstevel@tonic-gate 32367c478bd9Sstevel@tonic-gate /* 3237a71e32b6Sstans * Caller is expected to hold page's io lock for VMODSORT to work 3238a71e32b6Sstans * correctly with pvn_vplist_dirty() and pvn_getdirty() when mod 3239a71e32b6Sstans * bit is cleared. 3240a71e32b6Sstans * We don't have assert to avoid tripping some existing third party 3241a71e32b6Sstans * code. The dirty page is moved back to top of the v_page list 3242a71e32b6Sstans * after IO is done in pvn_write_done(). 32437c478bd9Sstevel@tonic-gate */ 32447c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 32457c478bd9Sstevel@tonic-gate 3246a71e32b6Sstans if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 32477c478bd9Sstevel@tonic-gate 32487c478bd9Sstevel@tonic-gate /* 32497c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 32507c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 32517c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 32527c478bd9Sstevel@tonic-gate * to this page. 32537c478bd9Sstevel@tonic-gate */ 32547c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 32557c478bd9Sstevel@tonic-gate } 32567c478bd9Sstevel@tonic-gate } 32577c478bd9Sstevel@tonic-gate 32587c478bd9Sstevel@tonic-gate /* 32597c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 32607c478bd9Sstevel@tonic-gate * and non zero if enabled. If flag specifes multiple attributs 32617c478bd9Sstevel@tonic-gate * then returns 0 if ALL atriibutes are disabled. This is an advisory 32627c478bd9Sstevel@tonic-gate * call. 32637c478bd9Sstevel@tonic-gate */ 32647c478bd9Sstevel@tonic-gate uint_t 32657c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 32667c478bd9Sstevel@tonic-gate { 32677c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 32687c478bd9Sstevel@tonic-gate } 32697c478bd9Sstevel@tonic-gate 32707c478bd9Sstevel@tonic-gate 32717c478bd9Sstevel@tonic-gate /* 32727c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 32737c478bd9Sstevel@tonic-gate */ 32747c478bd9Sstevel@tonic-gate hment_t * 32757c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 32767c478bd9Sstevel@tonic-gate { 32777c478bd9Sstevel@tonic-gate x86pte_t old_pte; 32787c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 32797c478bd9Sstevel@tonic-gate hment_t *hm; 32807c478bd9Sstevel@tonic-gate 32817c478bd9Sstevel@tonic-gate /* 32827c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 32837c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 32847c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 32857c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 32867c478bd9Sstevel@tonic-gate */ 32877c478bd9Sstevel@tonic-gate htable_acquire(ht); 32887c478bd9Sstevel@tonic-gate 32897c478bd9Sstevel@tonic-gate /* 32907c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 32917c478bd9Sstevel@tonic-gate */ 3292ae115bc7Smrj old_pte = x86pte_inval(ht, entry, 0, NULL); 3293aa2ed9e5Sjosephb if (PTE2PFN(old_pte, ht->ht_level) != pfn) { 3294ae115bc7Smrj panic("x86pte_inval() failure found PTE = " FMT_PTE 3295aa2ed9e5Sjosephb " pfn being unmapped is %lx ht=0x%lx entry=0x%x", 3296aa2ed9e5Sjosephb old_pte, pfn, (uintptr_t)ht, entry); 3297aa2ed9e5Sjosephb } 32987c478bd9Sstevel@tonic-gate 32997c478bd9Sstevel@tonic-gate /* 33007c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 33017c478bd9Sstevel@tonic-gate */ 33027c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 33037c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 33047c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 33057c478bd9Sstevel@tonic-gate 33067c478bd9Sstevel@tonic-gate /* 33077c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 33087c478bd9Sstevel@tonic-gate */ 3309ae115bc7Smrj if (PTE_GET(old_pte, PT_SOFTWARE) < PT_NOSYNC) 33107c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 33117c478bd9Sstevel@tonic-gate 33127c478bd9Sstevel@tonic-gate /* 33137c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 33147c478bd9Sstevel@tonic-gate */ 33157c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 33167c478bd9Sstevel@tonic-gate 33177c478bd9Sstevel@tonic-gate /* 33187c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 33197c478bd9Sstevel@tonic-gate * hment and htable. 33207c478bd9Sstevel@tonic-gate */ 33217c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 33227c478bd9Sstevel@tonic-gate htable_release(ht); 33237c478bd9Sstevel@tonic-gate return (hm); 33247c478bd9Sstevel@tonic-gate } 33257c478bd9Sstevel@tonic-gate 3326a5652762Spraks extern int vpm_enable; 33277c478bd9Sstevel@tonic-gate /* 33287c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 33297c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 33307c478bd9Sstevel@tonic-gate * 33317c478bd9Sstevel@tonic-gate * The forceflags are unused. 33327c478bd9Sstevel@tonic-gate */ 33337c478bd9Sstevel@tonic-gate 33347c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 33357c478bd9Sstevel@tonic-gate static int 33367c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 33377c478bd9Sstevel@tonic-gate { 33387c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 33397c478bd9Sstevel@tonic-gate hment_t *hm; 33407c478bd9Sstevel@tonic-gate hment_t *prev; 33417c478bd9Sstevel@tonic-gate htable_t *ht; 33427c478bd9Sstevel@tonic-gate uint_t entry; 33437c478bd9Sstevel@tonic-gate level_t level; 33447c478bd9Sstevel@tonic-gate 3345843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 3346a5652762Spraks #if defined(__amd64) 3347a5652762Spraks /* 3348a5652762Spraks * clear the vpm ref. 3349a5652762Spraks */ 3350a5652762Spraks if (vpm_enable) { 3351a5652762Spraks pp->p_vpmref = 0; 3352a5652762Spraks } 3353a5652762Spraks #endif 33547c478bd9Sstevel@tonic-gate /* 33557c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 33567c478bd9Sstevel@tonic-gate */ 33577c478bd9Sstevel@tonic-gate next_size: 33587c478bd9Sstevel@tonic-gate for (;;) { 33597c478bd9Sstevel@tonic-gate 33607c478bd9Sstevel@tonic-gate /* 33617c478bd9Sstevel@tonic-gate * Get a mapping list entry 33627c478bd9Sstevel@tonic-gate */ 33637c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 33647c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 33657c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 33667c478bd9Sstevel@tonic-gate if (hm == NULL) { 33677c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 33687c478bd9Sstevel@tonic-gate 33697c478bd9Sstevel@tonic-gate /* 33707c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 33717c478bd9Sstevel@tonic-gate */ 3372ae115bc7Smrj if (cur_pp->p_szc <= pg_szcd) { 3373843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 33747c478bd9Sstevel@tonic-gate return (0); 3375ae115bc7Smrj } 33767c478bd9Sstevel@tonic-gate 33777c478bd9Sstevel@tonic-gate /* 33787c478bd9Sstevel@tonic-gate * Else check the next larger page size. 33797c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 33807c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 33817c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 33827c478bd9Sstevel@tonic-gate * and return. 33837c478bd9Sstevel@tonic-gate */ 33847c478bd9Sstevel@tonic-gate ++pg_szcd; 33857c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 33867c478bd9Sstevel@tonic-gate goto next_size; 33877c478bd9Sstevel@tonic-gate } 33887c478bd9Sstevel@tonic-gate 33897c478bd9Sstevel@tonic-gate /* 33907c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 33917c478bd9Sstevel@tonic-gate */ 33927c478bd9Sstevel@tonic-gate level = ht->ht_level; 33937c478bd9Sstevel@tonic-gate if (level == pg_szcd) 33947c478bd9Sstevel@tonic-gate break; 33957c478bd9Sstevel@tonic-gate } 33967c478bd9Sstevel@tonic-gate 33977c478bd9Sstevel@tonic-gate /* 33987c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 33997c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 34007c478bd9Sstevel@tonic-gate */ 34017c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 34027c478bd9Sstevel@tonic-gate if (hm != NULL) 34037c478bd9Sstevel@tonic-gate hment_free(hm); 34047c478bd9Sstevel@tonic-gate } 34057c478bd9Sstevel@tonic-gate } 34067c478bd9Sstevel@tonic-gate 34077c478bd9Sstevel@tonic-gate int 34087c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 34097c478bd9Sstevel@tonic-gate { 34107c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 34117c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 34127c478bd9Sstevel@tonic-gate } 34137c478bd9Sstevel@tonic-gate 34147c478bd9Sstevel@tonic-gate /* 34157c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 34167c478bd9Sstevel@tonic-gate * page level that included pp. 34177c478bd9Sstevel@tonic-gate * 34187c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 34197c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 34207c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 34217c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 34227c478bd9Sstevel@tonic-gate * locked. 34237c478bd9Sstevel@tonic-gate * 34247c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 34257c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 34267c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 34277c478bd9Sstevel@tonic-gate * 34287c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 34297c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 34307c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 34317c478bd9Sstevel@tonic-gate */ 34327c478bd9Sstevel@tonic-gate void 34337c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 34347c478bd9Sstevel@tonic-gate { 34357c478bd9Sstevel@tonic-gate uint_t pszc; 34367c478bd9Sstevel@tonic-gate uint_t rszc; 34377c478bd9Sstevel@tonic-gate uint_t szc; 34387c478bd9Sstevel@tonic-gate page_t *rootpp; 34397c478bd9Sstevel@tonic-gate page_t *firstpp; 34407c478bd9Sstevel@tonic-gate page_t *lastpp; 34417c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 34427c478bd9Sstevel@tonic-gate 34437c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 34447c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 34457c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 34467c478bd9Sstevel@tonic-gate 34477c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 34487c478bd9Sstevel@tonic-gate return; 34497c478bd9Sstevel@tonic-gate 34507c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 34517c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 34527c478bd9Sstevel@tonic-gate 34537c478bd9Sstevel@tonic-gate /* 34547c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 34557c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 34567c478bd9Sstevel@tonic-gate * 34577c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 34587c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 34597c478bd9Sstevel@tonic-gate * 34607c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 34617c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 34627c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 34637c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 34647c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 34657c478bd9Sstevel@tonic-gate * page). 34667c478bd9Sstevel@tonic-gate */ 34677c478bd9Sstevel@tonic-gate again: 34687c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 34697c478bd9Sstevel@tonic-gate if (pszc == 0) 34707c478bd9Sstevel@tonic-gate return; 34717c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 34727c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 34737c478bd9Sstevel@tonic-gate /* 34747c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 34757c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 34767c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 34777c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 34787c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 34797c478bd9Sstevel@tonic-gate */ 34807c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 34817c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 34827c478bd9Sstevel@tonic-gate if (rszc > pszc) { 34837c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 34847c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 34857c478bd9Sstevel@tonic-gate 34867c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 34877c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 34887c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 34897c478bd9Sstevel@tonic-gate } 34907c478bd9Sstevel@tonic-gate goto again; 34917c478bd9Sstevel@tonic-gate } 34927c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 34937c478bd9Sstevel@tonic-gate 34947c478bd9Sstevel@tonic-gate /* 34957c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 34967c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 34977c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 34987c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 34997c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 35007c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 35017c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 35027c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 35037c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 35047c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 35057c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 35067c478bd9Sstevel@tonic-gate * 35077c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 35087c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 35097c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 35107c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 35117c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 35127c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 35137c478bd9Sstevel@tonic-gate * 35147c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 35157c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 35167c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 35177c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 35187c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 35197c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 35207c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 35217c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 35227c478bd9Sstevel@tonic-gate */ 35237c478bd9Sstevel@tonic-gate szc = pszc; 35247c478bd9Sstevel@tonic-gate while (szc > 1) { 35257c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 35267c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 35277c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 35287c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 35297c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 35307c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 35317c478bd9Sstevel@tonic-gate szc--; 35327c478bd9Sstevel@tonic-gate continue; 35337c478bd9Sstevel@tonic-gate } 35347c478bd9Sstevel@tonic-gate firstpp += pgcnt; 35357c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 35367c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 35377c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 35387c478bd9Sstevel@tonic-gate lastpp--; 35397c478bd9Sstevel@tonic-gate } 35407c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 35417c478bd9Sstevel@tonic-gate szc--; 35427c478bd9Sstevel@tonic-gate } 35437c478bd9Sstevel@tonic-gate 35447c478bd9Sstevel@tonic-gate /* 35457c478bd9Sstevel@tonic-gate * Second loop description: 35467c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 35477c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 35487c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 35497c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 35507c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 35517c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 35527c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 35537c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 35547c478bd9Sstevel@tonic-gate * skip to the next iteration. 35557c478bd9Sstevel@tonic-gate * 35567c478bd9Sstevel@tonic-gate */ 35577c478bd9Sstevel@tonic-gate szc = 0; 35587c478bd9Sstevel@tonic-gate while (szc < pszc) { 35597c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 35607c478bd9Sstevel@tonic-gate if (szc == 0) { 35617c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 35627c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 35637c478bd9Sstevel@tonic-gate } else { 35647c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 35657c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 35667c478bd9Sstevel@tonic-gate szc++; 35677c478bd9Sstevel@tonic-gate continue; 35687c478bd9Sstevel@tonic-gate } 35697c478bd9Sstevel@tonic-gate lastpp--; 35707c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 35717c478bd9Sstevel@tonic-gate } 35727c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 35737c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 35747c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 35757c478bd9Sstevel@tonic-gate lastpp--; 35767c478bd9Sstevel@tonic-gate } 35777c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 35787c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 35797c478bd9Sstevel@tonic-gate break; 35807c478bd9Sstevel@tonic-gate szc++; 35817c478bd9Sstevel@tonic-gate } 35827c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 35837c478bd9Sstevel@tonic-gate } 35847c478bd9Sstevel@tonic-gate 35857c478bd9Sstevel@tonic-gate /* 35867c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 35877c478bd9Sstevel@tonic-gate * returns attributes of page 35887c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 35897c478bd9Sstevel@tonic-gate * 35907c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 35917c478bd9Sstevel@tonic-gate * 35927c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 35937c478bd9Sstevel@tonic-gate * 35947c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 35957c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 35967c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 35977c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 35987c478bd9Sstevel@tonic-gate */ 35997c478bd9Sstevel@tonic-gate uint_t 36007c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 36017c478bd9Sstevel@tonic-gate { 36027c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 36037c478bd9Sstevel@tonic-gate htable_t *ht; 36047c478bd9Sstevel@tonic-gate uint_t entry; 36057c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 36067c478bd9Sstevel@tonic-gate x86pte_t new; 36077c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 36087c478bd9Sstevel@tonic-gate extern ulong_t po_share; 36097c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 36107c478bd9Sstevel@tonic-gate uint_t pszc = 0; 36117c478bd9Sstevel@tonic-gate 36127c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 36137c478bd9Sstevel@tonic-gate 36147c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 36157c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36167c478bd9Sstevel@tonic-gate 36177c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 36187c478bd9Sstevel@tonic-gate 36197c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 36207c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36217c478bd9Sstevel@tonic-gate 36227c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 36237c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36247c478bd9Sstevel@tonic-gate 36257c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 36267c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 36277c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 36287c478bd9Sstevel@tonic-gate PP_SETREF(pp); 36297c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 36307c478bd9Sstevel@tonic-gate } 36317c478bd9Sstevel@tonic-gate } 36327c478bd9Sstevel@tonic-gate 3633843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 36347c478bd9Sstevel@tonic-gate next_size: 36357c478bd9Sstevel@tonic-gate /* 36367c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 36377c478bd9Sstevel@tonic-gate */ 36387c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 36397c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 36407c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 36417c478bd9Sstevel@tonic-gate continue; 36427c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 36437c478bd9Sstevel@tonic-gate try_again: 36447c478bd9Sstevel@tonic-gate 36457c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 36467c478bd9Sstevel@tonic-gate 36477c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 36487c478bd9Sstevel@tonic-gate continue; 36497c478bd9Sstevel@tonic-gate 36507c478bd9Sstevel@tonic-gate save_old = old; 36517c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 36527c478bd9Sstevel@tonic-gate 36537c478bd9Sstevel@tonic-gate /* 36547c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 36557c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 36567c478bd9Sstevel@tonic-gate */ 36577c478bd9Sstevel@tonic-gate new = old; 36587c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 36597c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 36607c478bd9Sstevel@tonic-gate if (old != 0) 36617c478bd9Sstevel@tonic-gate goto try_again; 36627c478bd9Sstevel@tonic-gate 36637c478bd9Sstevel@tonic-gate old = save_old; 36647c478bd9Sstevel@tonic-gate } 36657c478bd9Sstevel@tonic-gate 36667c478bd9Sstevel@tonic-gate /* 36677c478bd9Sstevel@tonic-gate * Sync the PTE 36687c478bd9Sstevel@tonic-gate */ 3669ae115bc7Smrj if (!(flags & HAT_SYNC_ZERORM) && 3670ae115bc7Smrj PTE_GET(old, PT_SOFTWARE) <= PT_NOSYNC) 36717c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 36727c478bd9Sstevel@tonic-gate 36737c478bd9Sstevel@tonic-gate /* 36747c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 36757c478bd9Sstevel@tonic-gate */ 36767c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 36777c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 36787c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3679ae115bc7Smrj goto done; 36807c478bd9Sstevel@tonic-gate } 36817c478bd9Sstevel@tonic-gate } 36827c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 36837c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 36847c478bd9Sstevel@tonic-gate page_t *tpp; 36857c478bd9Sstevel@tonic-gate pszc++; 36867c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 36877c478bd9Sstevel@tonic-gate if (pp != tpp) { 36887c478bd9Sstevel@tonic-gate pp = tpp; 36897c478bd9Sstevel@tonic-gate goto next_size; 36907c478bd9Sstevel@tonic-gate } 36917c478bd9Sstevel@tonic-gate } 3692ae115bc7Smrj done: 3693843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 36947c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 36957c478bd9Sstevel@tonic-gate } 36967c478bd9Sstevel@tonic-gate 36977c478bd9Sstevel@tonic-gate /* 36987c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 36997c478bd9Sstevel@tonic-gate * there are no mappings to the page. 37007c478bd9Sstevel@tonic-gate */ 37017c478bd9Sstevel@tonic-gate ulong_t 37027c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 37037c478bd9Sstevel@tonic-gate { 37047c478bd9Sstevel@tonic-gate uint_t cnt; 37057c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3706a5652762Spraks #if defined(__amd64) 3707a5652762Spraks if (vpm_enable && pp->p_vpmref) { 3708a5652762Spraks cnt += 1; 3709a5652762Spraks } 3710a5652762Spraks #endif 37117c478bd9Sstevel@tonic-gate return (cnt); 37127c478bd9Sstevel@tonic-gate } 37137c478bd9Sstevel@tonic-gate 371405d3dc4bSpaulsan /* 371505d3dc4bSpaulsan * Return 1 the number of mappings exceeds sh_thresh. Return 0 371605d3dc4bSpaulsan * otherwise. 371705d3dc4bSpaulsan */ 371805d3dc4bSpaulsan int 371905d3dc4bSpaulsan hat_page_checkshare(page_t *pp, ulong_t sh_thresh) 372005d3dc4bSpaulsan { 372105d3dc4bSpaulsan return (hat_page_getshare(pp) > sh_thresh); 372205d3dc4bSpaulsan } 372305d3dc4bSpaulsan 37247c478bd9Sstevel@tonic-gate /* 37257c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 37267c478bd9Sstevel@tonic-gate */ 37277c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37287c478bd9Sstevel@tonic-gate faultcode_t 37297c478bd9Sstevel@tonic-gate hat_softlock( 37307c478bd9Sstevel@tonic-gate hat_t *hat, 37317c478bd9Sstevel@tonic-gate caddr_t addr, 37327c478bd9Sstevel@tonic-gate size_t *len, 37337c478bd9Sstevel@tonic-gate struct page **page_array, 37347c478bd9Sstevel@tonic-gate uint_t flags) 37357c478bd9Sstevel@tonic-gate { 37367c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 37377c478bd9Sstevel@tonic-gate } 37387c478bd9Sstevel@tonic-gate 37397c478bd9Sstevel@tonic-gate 37407c478bd9Sstevel@tonic-gate 37417c478bd9Sstevel@tonic-gate /* 37427c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 37437c478bd9Sstevel@tonic-gate */ 37447c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37457c478bd9Sstevel@tonic-gate int 37467c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 37477c478bd9Sstevel@tonic-gate { 37487c478bd9Sstevel@tonic-gate switch (feature) { 37497c478bd9Sstevel@tonic-gate 37507c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 37517c478bd9Sstevel@tonic-gate return (1); 37527c478bd9Sstevel@tonic-gate 37537c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 37547c478bd9Sstevel@tonic-gate return (0); 37557c478bd9Sstevel@tonic-gate 37567c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 37577c478bd9Sstevel@tonic-gate return (1); 37587c478bd9Sstevel@tonic-gate 375905d3dc4bSpaulsan case HAT_SHARED_REGIONS: 376005d3dc4bSpaulsan return (0); 376105d3dc4bSpaulsan 37627c478bd9Sstevel@tonic-gate default: 37637c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 37647c478bd9Sstevel@tonic-gate } 37657c478bd9Sstevel@tonic-gate return (0); 37667c478bd9Sstevel@tonic-gate } 37677c478bd9Sstevel@tonic-gate 37687c478bd9Sstevel@tonic-gate /* 37697c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 37707c478bd9Sstevel@tonic-gate */ 37717c478bd9Sstevel@tonic-gate void 37727c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 37737c478bd9Sstevel@tonic-gate { 37747c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 3775843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 37767c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 3777843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 37787c478bd9Sstevel@tonic-gate } 37797c478bd9Sstevel@tonic-gate 37807c478bd9Sstevel@tonic-gate /* 37817c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 37827c478bd9Sstevel@tonic-gate */ 37837c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 37847c478bd9Sstevel@tonic-gate void 37857c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 37867c478bd9Sstevel@tonic-gate { 3787843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 37887c478bd9Sstevel@tonic-gate kpreempt_disable(); 37897c478bd9Sstevel@tonic-gate 37907c478bd9Sstevel@tonic-gate hat_switch(hat); 37917c478bd9Sstevel@tonic-gate 37927c478bd9Sstevel@tonic-gate kpreempt_enable(); 3793843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 37947c478bd9Sstevel@tonic-gate } 37957c478bd9Sstevel@tonic-gate 37967c478bd9Sstevel@tonic-gate /* 37977c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 37987c478bd9Sstevel@tonic-gate * 37997c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 38007c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 38017c478bd9Sstevel@tonic-gate * 38027c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 38037c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 38047c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 38057c478bd9Sstevel@tonic-gate */ 3806ae115bc7Smrj hat_mempte_t 3807ae115bc7Smrj hat_mempte_setup(caddr_t addr) 38087c478bd9Sstevel@tonic-gate { 38097c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 38107c478bd9Sstevel@tonic-gate htable_t *ht; 38117c478bd9Sstevel@tonic-gate uint_t entry; 38127c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3813ae115bc7Smrj hat_mempte_t p; 38147c478bd9Sstevel@tonic-gate 38157c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 38167c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3817aac11643Sjosephb ++curthread->t_hatdepth; 38187c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 38197c478bd9Sstevel@tonic-gate if (ht == NULL) { 38207c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 38217c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 38227c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 38237c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 38247c478bd9Sstevel@tonic-gate } 38257c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 38267c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 38277c478bd9Sstevel@tonic-gate "ht=%p, entry=%d, pte=" FMT_PTE, ht, entry, oldpte); 38287c478bd9Sstevel@tonic-gate 38297c478bd9Sstevel@tonic-gate /* 38307c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 38317c478bd9Sstevel@tonic-gate */ 38327c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 38337c478bd9Sstevel@tonic-gate 38347c478bd9Sstevel@tonic-gate /* 3835ae115bc7Smrj * return the PTE physical address to the caller. 38367c478bd9Sstevel@tonic-gate */ 38377c478bd9Sstevel@tonic-gate htable_release(ht); 3838ae115bc7Smrj p = PT_INDEX_PHYSADDR(pfn_to_pa(ht->ht_pfn), entry); 3839aac11643Sjosephb --curthread->t_hatdepth; 3840ae115bc7Smrj return (p); 38417c478bd9Sstevel@tonic-gate } 38427c478bd9Sstevel@tonic-gate 38437c478bd9Sstevel@tonic-gate /* 38447c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 38457c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 38467c478bd9Sstevel@tonic-gate */ 3847ae115bc7Smrj /*ARGSUSED1*/ 38487c478bd9Sstevel@tonic-gate void 3849ae115bc7Smrj hat_mempte_release(caddr_t addr, hat_mempte_t pte_pa) 38507c478bd9Sstevel@tonic-gate { 38517c478bd9Sstevel@tonic-gate htable_t *ht; 38527c478bd9Sstevel@tonic-gate 38537c478bd9Sstevel@tonic-gate /* 3854ae115bc7Smrj * invalidate any left over mapping and decrement the htable valid count 38557c478bd9Sstevel@tonic-gate */ 3856843e1988Sjohnlev #ifdef __xpv 3857843e1988Sjohnlev if (HYPERVISOR_update_va_mapping((uintptr_t)addr, 0, 3858843e1988Sjohnlev UVMF_INVLPG | UVMF_LOCAL)) 3859843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3860843e1988Sjohnlev #else 3861ae115bc7Smrj { 3862ae115bc7Smrj x86pte_t *pteptr; 3863ae115bc7Smrj 3864ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3865ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3866ae115bc7Smrj if (mmu.pae_hat) 3867ae115bc7Smrj *pteptr = 0; 3868ae115bc7Smrj else 3869ae115bc7Smrj *(x86pte32_t *)pteptr = 0; 3870ae115bc7Smrj mmu_tlbflush_entry(addr); 3871ae115bc7Smrj x86pte_mapout(); 3872ae115bc7Smrj } 3873843e1988Sjohnlev #endif 3874ae115bc7Smrj 38757c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 38767c478bd9Sstevel@tonic-gate if (ht == NULL) 38777c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 38787c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 38797c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 38807c478bd9Sstevel@tonic-gate htable_release(ht); 38817c478bd9Sstevel@tonic-gate } 38827c478bd9Sstevel@tonic-gate 38837c478bd9Sstevel@tonic-gate /* 38847c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 38857c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 38867c478bd9Sstevel@tonic-gate */ 38877c478bd9Sstevel@tonic-gate void 38887c478bd9Sstevel@tonic-gate hat_mempte_remap( 3889ae115bc7Smrj pfn_t pfn, 3890ae115bc7Smrj caddr_t addr, 3891ae115bc7Smrj hat_mempte_t pte_pa, 3892ae115bc7Smrj uint_t attr, 3893ae115bc7Smrj uint_t flags) 38947c478bd9Sstevel@tonic-gate { 38957c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 38967c478bd9Sstevel@tonic-gate x86pte_t pte; 38977c478bd9Sstevel@tonic-gate 38987c478bd9Sstevel@tonic-gate /* 38997c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 39007c478bd9Sstevel@tonic-gate * on this CPU. 39017c478bd9Sstevel@tonic-gate */ 39027c478bd9Sstevel@tonic-gate #ifdef DEBUG 39037c478bd9Sstevel@tonic-gate htable_t *ht; 39047c478bd9Sstevel@tonic-gate uint_t entry; 39057c478bd9Sstevel@tonic-gate 39067c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 39077c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 39087c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 39097c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 39107c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 39117c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3912ae115bc7Smrj ASSERT(ht->ht_pfn == mmu_btop(pte_pa)); 39137c478bd9Sstevel@tonic-gate htable_release(ht); 39147c478bd9Sstevel@tonic-gate #endif 3915843e1988Sjohnlev XPV_DISALLOW_MIGRATE(); 39167c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3917843e1988Sjohnlev #ifdef __xpv 3918843e1988Sjohnlev if (HYPERVISOR_update_va_mapping(va, pte, UVMF_INVLPG | UVMF_LOCAL)) 3919843e1988Sjohnlev panic("HYPERVISOR_update_va_mapping() failed"); 3920843e1988Sjohnlev #else 3921ae115bc7Smrj { 3922ae115bc7Smrj x86pte_t *pteptr; 3923ae115bc7Smrj 3924ae115bc7Smrj pteptr = x86pte_mapin(mmu_btop(pte_pa), 3925ae115bc7Smrj (pte_pa & MMU_PAGEOFFSET) >> mmu.pte_size_shift, NULL); 3926ae115bc7Smrj if (mmu.pae_hat) 3927ae115bc7Smrj *(x86pte_t *)pteptr = pte; 3928ae115bc7Smrj else 3929ae115bc7Smrj *(x86pte32_t *)pteptr = (x86pte32_t)pte; 3930ae115bc7Smrj mmu_tlbflush_entry(addr); 3931ae115bc7Smrj x86pte_mapout(); 3932ae115bc7Smrj } 3933843e1988Sjohnlev #endif 3934843e1988Sjohnlev XPV_ALLOW_MIGRATE(); 39357c478bd9Sstevel@tonic-gate } 39367c478bd9Sstevel@tonic-gate 39377c478bd9Sstevel@tonic-gate 39387c478bd9Sstevel@tonic-gate 39397c478bd9Sstevel@tonic-gate /* 39407c478bd9Sstevel@tonic-gate * Hat locking functions 39417c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 39427c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 39437c478bd9Sstevel@tonic-gate */ 39447c478bd9Sstevel@tonic-gate void 39457c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 39467c478bd9Sstevel@tonic-gate { 39477c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 39487c478bd9Sstevel@tonic-gate } 39497c478bd9Sstevel@tonic-gate 39507c478bd9Sstevel@tonic-gate void 39517c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 39527c478bd9Sstevel@tonic-gate { 39537c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 39547c478bd9Sstevel@tonic-gate } 39557c478bd9Sstevel@tonic-gate 39567c478bd9Sstevel@tonic-gate /* 3957ae115bc7Smrj * HAT part of cpu initialization. 39587c478bd9Sstevel@tonic-gate */ 39597c478bd9Sstevel@tonic-gate void 3960ae115bc7Smrj hat_cpu_online(struct cpu *cpup) 39617c478bd9Sstevel@tonic-gate { 3962ae115bc7Smrj if (cpup != CPU) { 3963ae115bc7Smrj x86pte_cpu_init(cpup); 3964ae115bc7Smrj hat_vlp_setup(cpup); 39657c478bd9Sstevel@tonic-gate } 3966ae115bc7Smrj CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 39677c478bd9Sstevel@tonic-gate } 39687c478bd9Sstevel@tonic-gate 39697c478bd9Sstevel@tonic-gate /* 3970ae115bc7Smrj * HAT part of cpu deletion. 3971ae115bc7Smrj * (currently, we only call this after the cpu is safely passivated.) 39727c478bd9Sstevel@tonic-gate */ 39737c478bd9Sstevel@tonic-gate void 3974ae115bc7Smrj hat_cpu_offline(struct cpu *cpup) 39757c478bd9Sstevel@tonic-gate { 3976ae115bc7Smrj ASSERT(cpup != CPU); 3977ae115bc7Smrj 3978ae115bc7Smrj CPUSET_ATOMIC_DEL(khat_cpuset, cpup->cpu_id); 3979ae115bc7Smrj x86pte_cpu_fini(cpup); 3980ae115bc7Smrj hat_vlp_teardown(cpup); 39817c478bd9Sstevel@tonic-gate } 39827c478bd9Sstevel@tonic-gate 39837c478bd9Sstevel@tonic-gate /* 39847c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 39857c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 39867c478bd9Sstevel@tonic-gate */ 39877c478bd9Sstevel@tonic-gate void 39887c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 39897c478bd9Sstevel@tonic-gate { 39907c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 39917c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 39927c478bd9Sstevel@tonic-gate level_t level; 39937c478bd9Sstevel@tonic-gate uint_t entry; 39947c478bd9Sstevel@tonic-gate x86pte_t pte; 39957c478bd9Sstevel@tonic-gate 39967c478bd9Sstevel@tonic-gate /* 39977c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 3998ae115bc7Smrj * all virtual mappings under _userlimit. 39997c478bd9Sstevel@tonic-gate */ 40007c478bd9Sstevel@tonic-gate while (vaddr < high) { 40017c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 40027c478bd9Sstevel@tonic-gate if (ht == NULL) 40037c478bd9Sstevel@tonic-gate break; 40047c478bd9Sstevel@tonic-gate 40057c478bd9Sstevel@tonic-gate level = ht->ht_level; 40067c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 40077c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 40087c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 40097c478bd9Sstevel@tonic-gate 40107c478bd9Sstevel@tonic-gate /* 40117c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 40127c478bd9Sstevel@tonic-gate */ 4013ae115bc7Smrj (void) x86pte_inval(ht, entry, 0, NULL); 40147c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 40157c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 40167c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 40177c478bd9Sstevel@tonic-gate 40187c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 40197c478bd9Sstevel@tonic-gate } 40207c478bd9Sstevel@tonic-gate if (ht) 40217c478bd9Sstevel@tonic-gate htable_release(ht); 40227c478bd9Sstevel@tonic-gate } 40237c478bd9Sstevel@tonic-gate 40247c478bd9Sstevel@tonic-gate /* 40257c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 40267c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 40277c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 40287c478bd9Sstevel@tonic-gate * 40297c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 40307c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 40317c478bd9Sstevel@tonic-gate */ 40327c478bd9Sstevel@tonic-gate static x86pte_t 40337c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 40347c478bd9Sstevel@tonic-gate { 40357c478bd9Sstevel@tonic-gate page_t *pp; 40367c478bd9Sstevel@tonic-gate uint_t rm = 0; 40377c478bd9Sstevel@tonic-gate x86pte_t replaced; 40387c478bd9Sstevel@tonic-gate 4039ae115bc7Smrj if (PTE_GET(expected, PT_SOFTWARE) < PT_NOSYNC && 40407c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 40417c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 4042250b7ff9Sjosephb !PTE_GET(new, PT_MOD | PT_REF))) { 40437c478bd9Sstevel@tonic-gate 4044ae115bc7Smrj ASSERT(!pfn_is_foreign(PTE2PFN(expected, ht->ht_level))); 40457c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 40467c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 40477c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 40487c478bd9Sstevel@tonic-gate rm |= P_MOD; 40497c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 40507c478bd9Sstevel@tonic-gate rm |= P_REF; 40517c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 40527c478bd9Sstevel@tonic-gate } 40537c478bd9Sstevel@tonic-gate 40547c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 40557c478bd9Sstevel@tonic-gate if (replaced != expected) 40567c478bd9Sstevel@tonic-gate return (replaced); 40577c478bd9Sstevel@tonic-gate 40587c478bd9Sstevel@tonic-gate if (rm) { 40597c478bd9Sstevel@tonic-gate /* 40607c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 40617c478bd9Sstevel@tonic-gate */ 40627c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 40637c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 40647c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 40657c478bd9Sstevel@tonic-gate /* 40667c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 40677c478bd9Sstevel@tonic-gate * pszc below this mapping size 40687c478bd9Sstevel@tonic-gate * since large mapping existed after we 40697c478bd9Sstevel@tonic-gate * took mlist lock. 40707c478bd9Sstevel@tonic-gate */ 40717c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 40727c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 40737c478bd9Sstevel@tonic-gate ++pp; 40747c478bd9Sstevel@tonic-gate } 40757c478bd9Sstevel@tonic-gate } 40767c478bd9Sstevel@tonic-gate 40777c478bd9Sstevel@tonic-gate return (0); 40787c478bd9Sstevel@tonic-gate } 40797c478bd9Sstevel@tonic-gate 408005d3dc4bSpaulsan /* ARGSUSED */ 408105d3dc4bSpaulsan void 40827dacfc44Spaulsan hat_join_srd(struct hat *hat, vnode_t *evp) 408305d3dc4bSpaulsan { 408405d3dc4bSpaulsan } 408505d3dc4bSpaulsan 408605d3dc4bSpaulsan /* ARGSUSED */ 408705d3dc4bSpaulsan hat_region_cookie_t 40887dacfc44Spaulsan hat_join_region(struct hat *hat, 408905d3dc4bSpaulsan caddr_t r_saddr, 409005d3dc4bSpaulsan size_t r_size, 409105d3dc4bSpaulsan void *r_obj, 409205d3dc4bSpaulsan u_offset_t r_objoff, 409305d3dc4bSpaulsan uchar_t r_perm, 409405d3dc4bSpaulsan uchar_t r_pgszc, 409505d3dc4bSpaulsan hat_rgn_cb_func_t r_cb_function, 409605d3dc4bSpaulsan uint_t flags) 409705d3dc4bSpaulsan { 409805d3dc4bSpaulsan panic("No shared region support on x86"); 409905d3dc4bSpaulsan return (HAT_INVALID_REGION_COOKIE); 410005d3dc4bSpaulsan } 410105d3dc4bSpaulsan 410205d3dc4bSpaulsan /* ARGSUSED */ 410305d3dc4bSpaulsan void 41047dacfc44Spaulsan hat_leave_region(struct hat *hat, hat_region_cookie_t rcookie, uint_t flags) 410505d3dc4bSpaulsan { 410605d3dc4bSpaulsan panic("No shared region support on x86"); 410705d3dc4bSpaulsan } 410805d3dc4bSpaulsan 410905d3dc4bSpaulsan /* ARGSUSED */ 411005d3dc4bSpaulsan void 41117dacfc44Spaulsan hat_dup_region(struct hat *hat, hat_region_cookie_t rcookie) 411205d3dc4bSpaulsan { 411305d3dc4bSpaulsan panic("No shared region support on x86"); 411405d3dc4bSpaulsan } 411505d3dc4bSpaulsan 411605d3dc4bSpaulsan 41177c478bd9Sstevel@tonic-gate /* 41187c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 41197c478bd9Sstevel@tonic-gate * 41207c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 41217c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 41227c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 41237c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 41247c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 41257c478bd9Sstevel@tonic-gate * caches. 41267c478bd9Sstevel@tonic-gate * 41277c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 41287c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 41297c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 41307c478bd9Sstevel@tonic-gate * upper levels. 41317c478bd9Sstevel@tonic-gate */ 41327c478bd9Sstevel@tonic-gate 41337c478bd9Sstevel@tonic-gate /* 41347c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 41357c478bd9Sstevel@tonic-gate */ 41367c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 41377c478bd9Sstevel@tonic-gate caddr_t 41387c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 41397c478bd9Sstevel@tonic-gate { 41407c478bd9Sstevel@tonic-gate caddr_t vaddr; 41417c478bd9Sstevel@tonic-gate 41427c478bd9Sstevel@tonic-gate #ifdef DEBUG 41437c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 41447c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 41457c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41467c478bd9Sstevel@tonic-gate } 41477c478bd9Sstevel@tonic-gate 41487c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 41497c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 41507c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 41517c478bd9Sstevel@tonic-gate } 41527c478bd9Sstevel@tonic-gate #endif 41537c478bd9Sstevel@tonic-gate 41547c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 41557c478bd9Sstevel@tonic-gate 41567c478bd9Sstevel@tonic-gate return (vaddr); 41577c478bd9Sstevel@tonic-gate } 41587c478bd9Sstevel@tonic-gate 41597c478bd9Sstevel@tonic-gate /* 41607c478bd9Sstevel@tonic-gate * Mapout a locked page. 41617c478bd9Sstevel@tonic-gate */ 41627c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 41637c478bd9Sstevel@tonic-gate void 41647c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 41657c478bd9Sstevel@tonic-gate { 41667c478bd9Sstevel@tonic-gate #ifdef DEBUG 41677c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 41687c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 41697c478bd9Sstevel@tonic-gate return; 41707c478bd9Sstevel@tonic-gate } 41717c478bd9Sstevel@tonic-gate 41727c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 41737c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 41747c478bd9Sstevel@tonic-gate return; 41757c478bd9Sstevel@tonic-gate } 41767c478bd9Sstevel@tonic-gate 41777c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 41787c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 41797c478bd9Sstevel@tonic-gate return; 41807c478bd9Sstevel@tonic-gate } 41817c478bd9Sstevel@tonic-gate #endif 41827c478bd9Sstevel@tonic-gate } 41837c478bd9Sstevel@tonic-gate 41847c478bd9Sstevel@tonic-gate /* 41857c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 41867c478bd9Sstevel@tonic-gate */ 41877c478bd9Sstevel@tonic-gate caddr_t 41887c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 41897c478bd9Sstevel@tonic-gate { 4190ae115bc7Smrj uintptr_t vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 41917c478bd9Sstevel@tonic-gate 4192d2b85481Srscott ASSERT(!pfn_is_foreign(pfn)); 41937c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 41947c478bd9Sstevel@tonic-gate } 41957c478bd9Sstevel@tonic-gate 41967c478bd9Sstevel@tonic-gate /* 41977c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 41987c478bd9Sstevel@tonic-gate */ 41997c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42007c478bd9Sstevel@tonic-gate caddr_t 42017c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 42027c478bd9Sstevel@tonic-gate { 42037c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 42047c478bd9Sstevel@tonic-gate } 42057c478bd9Sstevel@tonic-gate 42067c478bd9Sstevel@tonic-gate /* 42077c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 42087c478bd9Sstevel@tonic-gate */ 42097c478bd9Sstevel@tonic-gate pfn_t 42107c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 42117c478bd9Sstevel@tonic-gate { 42127c478bd9Sstevel@tonic-gate pfn_t pfn; 42137c478bd9Sstevel@tonic-gate 42147c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 42157c478bd9Sstevel@tonic-gate 42167c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 42177c478bd9Sstevel@tonic-gate 42187c478bd9Sstevel@tonic-gate return (pfn); 42197c478bd9Sstevel@tonic-gate } 42207c478bd9Sstevel@tonic-gate 42217c478bd9Sstevel@tonic-gate 42227c478bd9Sstevel@tonic-gate /* 42237c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 42247c478bd9Sstevel@tonic-gate */ 42257c478bd9Sstevel@tonic-gate page_t * 42267c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 42277c478bd9Sstevel@tonic-gate { 42287c478bd9Sstevel@tonic-gate pfn_t pfn; 42297c478bd9Sstevel@tonic-gate 42307c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 42317c478bd9Sstevel@tonic-gate 42327c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 42337c478bd9Sstevel@tonic-gate 42347c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 42357c478bd9Sstevel@tonic-gate } 42367c478bd9Sstevel@tonic-gate 42377c478bd9Sstevel@tonic-gate /* 42387c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 42397c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 42407c478bd9Sstevel@tonic-gate */ 42417c478bd9Sstevel@tonic-gate int 42427c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 42437c478bd9Sstevel@tonic-gate { 42447c478bd9Sstevel@tonic-gate panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", hat, vaddr); 42457c478bd9Sstevel@tonic-gate 42467c478bd9Sstevel@tonic-gate return (0); 42477c478bd9Sstevel@tonic-gate } 42487c478bd9Sstevel@tonic-gate 42497c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42507c478bd9Sstevel@tonic-gate void 42517c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 42527c478bd9Sstevel@tonic-gate {} 42537c478bd9Sstevel@tonic-gate 42547c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 42557c478bd9Sstevel@tonic-gate void 42567c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 42577c478bd9Sstevel@tonic-gate {} 4258843e1988Sjohnlev 4259843e1988Sjohnlev #ifdef __xpv 4260843e1988Sjohnlev /* 4261843e1988Sjohnlev * There are specific Hypervisor calls to establish and remove mappings 4262843e1988Sjohnlev * to grant table references and the privcmd driver. We have to ensure 4263843e1988Sjohnlev * that a page table actually exists. 4264843e1988Sjohnlev */ 4265843e1988Sjohnlev void 4266843e1988Sjohnlev hat_prepare_mapping(hat_t *hat, caddr_t addr) 4267843e1988Sjohnlev { 4268843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4269843e1988Sjohnlev (void) htable_create(hat, (uintptr_t)addr, 0, NULL); 4270843e1988Sjohnlev } 4271843e1988Sjohnlev 4272843e1988Sjohnlev void 4273843e1988Sjohnlev hat_release_mapping(hat_t *hat, caddr_t addr) 4274843e1988Sjohnlev { 4275843e1988Sjohnlev htable_t *ht; 4276843e1988Sjohnlev 4277843e1988Sjohnlev ASSERT(IS_P2ALIGNED((uintptr_t)addr, MMU_PAGESIZE)); 4278843e1988Sjohnlev ht = htable_lookup(hat, (uintptr_t)addr, 0); 4279843e1988Sjohnlev ASSERT(ht != NULL); 4280843e1988Sjohnlev ASSERT(ht->ht_busy >= 2); 4281843e1988Sjohnlev htable_release(ht); 4282843e1988Sjohnlev htable_release(ht); 4283843e1988Sjohnlev } 4284843e1988Sjohnlev #endif 4285