1*7c478bd9Sstevel@tonic-gate /* 2*7c478bd9Sstevel@tonic-gate * CDDL HEADER START 3*7c478bd9Sstevel@tonic-gate * 4*7c478bd9Sstevel@tonic-gate * The contents of this file are subject to the terms of the 5*7c478bd9Sstevel@tonic-gate * Common Development and Distribution License, Version 1.0 only 6*7c478bd9Sstevel@tonic-gate * (the "License"). You may not use this file except in compliance 7*7c478bd9Sstevel@tonic-gate * with the License. 8*7c478bd9Sstevel@tonic-gate * 9*7c478bd9Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10*7c478bd9Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 11*7c478bd9Sstevel@tonic-gate * See the License for the specific language governing permissions 12*7c478bd9Sstevel@tonic-gate * and limitations under the License. 13*7c478bd9Sstevel@tonic-gate * 14*7c478bd9Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 15*7c478bd9Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16*7c478bd9Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 17*7c478bd9Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 18*7c478bd9Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 19*7c478bd9Sstevel@tonic-gate * 20*7c478bd9Sstevel@tonic-gate * CDDL HEADER END 21*7c478bd9Sstevel@tonic-gate */ 22*7c478bd9Sstevel@tonic-gate /* 23*7c478bd9Sstevel@tonic-gate * Copyright 2005 Sun Microsystems, Inc. All rights reserved. 24*7c478bd9Sstevel@tonic-gate * Use is subject to license terms. 25*7c478bd9Sstevel@tonic-gate */ 26*7c478bd9Sstevel@tonic-gate 27*7c478bd9Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 28*7c478bd9Sstevel@tonic-gate 29*7c478bd9Sstevel@tonic-gate /* 30*7c478bd9Sstevel@tonic-gate * VM - Hardware Address Translation management for i386 and amd64 31*7c478bd9Sstevel@tonic-gate * 32*7c478bd9Sstevel@tonic-gate * Implementation of the interfaces described in <common/vm/hat.h> 33*7c478bd9Sstevel@tonic-gate * 34*7c478bd9Sstevel@tonic-gate * Nearly all the details of how the hardware is managed should not be 35*7c478bd9Sstevel@tonic-gate * visible outside this layer except for misc. machine specific functions 36*7c478bd9Sstevel@tonic-gate * that work in conjunction with this code. 37*7c478bd9Sstevel@tonic-gate * 38*7c478bd9Sstevel@tonic-gate * Routines used only inside of i86pc/vm start with hati_ for HAT Internal. 39*7c478bd9Sstevel@tonic-gate */ 40*7c478bd9Sstevel@tonic-gate 41*7c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 42*7c478bd9Sstevel@tonic-gate #include <sys/machsystm.h> 43*7c478bd9Sstevel@tonic-gate #include <sys/mman.h> 44*7c478bd9Sstevel@tonic-gate #include <sys/types.h> 45*7c478bd9Sstevel@tonic-gate #include <sys/systm.h> 46*7c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h> 47*7c478bd9Sstevel@tonic-gate #include <sys/thread.h> 48*7c478bd9Sstevel@tonic-gate #include <sys/proc.h> 49*7c478bd9Sstevel@tonic-gate #include <sys/cpu.h> 50*7c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 51*7c478bd9Sstevel@tonic-gate #include <sys/disp.h> 52*7c478bd9Sstevel@tonic-gate #include <sys/shm.h> 53*7c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h> 54*7c478bd9Sstevel@tonic-gate #include <sys/machparam.h> 55*7c478bd9Sstevel@tonic-gate #include <sys/vmem.h> 56*7c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h> 57*7c478bd9Sstevel@tonic-gate #include <sys/promif.h> 58*7c478bd9Sstevel@tonic-gate #include <sys/var.h> 59*7c478bd9Sstevel@tonic-gate #include <sys/x86_archext.h> 60*7c478bd9Sstevel@tonic-gate #include <sys/atomic.h> 61*7c478bd9Sstevel@tonic-gate #include <sys/bitmap.h> 62*7c478bd9Sstevel@tonic-gate 63*7c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h> 64*7c478bd9Sstevel@tonic-gate #include <vm/hat_i86.h> 65*7c478bd9Sstevel@tonic-gate #include <vm/as.h> 66*7c478bd9Sstevel@tonic-gate #include <vm/seg.h> 67*7c478bd9Sstevel@tonic-gate #include <vm/page.h> 68*7c478bd9Sstevel@tonic-gate #include <vm/seg_kp.h> 69*7c478bd9Sstevel@tonic-gate #include <vm/seg_kpm.h> 70*7c478bd9Sstevel@tonic-gate #include <vm/vm_dep.h> 71*7c478bd9Sstevel@tonic-gate 72*7c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 73*7c478bd9Sstevel@tonic-gate 74*7c478bd9Sstevel@tonic-gate 75*7c478bd9Sstevel@tonic-gate /* 76*7c478bd9Sstevel@tonic-gate * Basic parameters for hat operation. 77*7c478bd9Sstevel@tonic-gate */ 78*7c478bd9Sstevel@tonic-gate struct hat_mmu_info mmu; 79*7c478bd9Sstevel@tonic-gate uint_t force_pae_off = 0; /* for testing, change with kernel debugger */ 80*7c478bd9Sstevel@tonic-gate uint_t force_pae_on = 0; /* for testing, change with kernel debugger */ 81*7c478bd9Sstevel@tonic-gate 82*7c478bd9Sstevel@tonic-gate /* 83*7c478bd9Sstevel@tonic-gate * The page that is the kernel's top level pagetable. 84*7c478bd9Sstevel@tonic-gate * 85*7c478bd9Sstevel@tonic-gate * For 32 bit VLP support, the kernel hat will use the 1st 4 entries 86*7c478bd9Sstevel@tonic-gate * on this 4K page for its top level page table. The remaining groups of 87*7c478bd9Sstevel@tonic-gate * 4 entries are used for per processor copies of user VLP pagetables for 88*7c478bd9Sstevel@tonic-gate * running threads. See hat_switch() and reload_pae32() for details. 89*7c478bd9Sstevel@tonic-gate * 90*7c478bd9Sstevel@tonic-gate * vlp_page[0] - 0th level==2 PTE for kernel HAT (will be zero) 91*7c478bd9Sstevel@tonic-gate * vlp_page[1] - 1st level==2 PTE for kernel HAT (will be zero) 92*7c478bd9Sstevel@tonic-gate * vlp_page[2] - 2nd level==2 PTE for kernel HAT (zero for small memory) 93*7c478bd9Sstevel@tonic-gate * vlp_page[3] - 3rd level==2 PTE for kernel 94*7c478bd9Sstevel@tonic-gate * 95*7c478bd9Sstevel@tonic-gate * vlp_page[4] - 0th level==2 PTE for user thread on cpu 0 96*7c478bd9Sstevel@tonic-gate * vlp_page[5] - 1st level==2 PTE for user thread on cpu 0 97*7c478bd9Sstevel@tonic-gate * vlp_page[6] - 2nd level==2 PTE for user thread on cpu 0 98*7c478bd9Sstevel@tonic-gate * vlp_page[7] - probably copy of kernel PTE 99*7c478bd9Sstevel@tonic-gate * 100*7c478bd9Sstevel@tonic-gate * vlp_page[8] - 0th level==2 PTE for user thread on cpu 1 101*7c478bd9Sstevel@tonic-gate * vlp_page[9] - 1st level==2 PTE for user thread on cpu 1 102*7c478bd9Sstevel@tonic-gate * vlp_page[10] - 2nd level==2 PTE for user thread on cpu 1 103*7c478bd9Sstevel@tonic-gate * vlp_page[11] - probably copy of kernel PTE 104*7c478bd9Sstevel@tonic-gate * ... 105*7c478bd9Sstevel@tonic-gate * 106*7c478bd9Sstevel@tonic-gate * when / where the kernel PTE's are (entry 2 or 3 or none) depends 107*7c478bd9Sstevel@tonic-gate * on kernelbase. 108*7c478bd9Sstevel@tonic-gate */ 109*7c478bd9Sstevel@tonic-gate static x86pte_t *vlp_page; 110*7c478bd9Sstevel@tonic-gate 111*7c478bd9Sstevel@tonic-gate /* 112*7c478bd9Sstevel@tonic-gate * forward declaration of internal utility routines 113*7c478bd9Sstevel@tonic-gate */ 114*7c478bd9Sstevel@tonic-gate static x86pte_t hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, 115*7c478bd9Sstevel@tonic-gate x86pte_t new); 116*7c478bd9Sstevel@tonic-gate 117*7c478bd9Sstevel@tonic-gate /* 118*7c478bd9Sstevel@tonic-gate * The kernel address space exists in all HATs. To implement this the 119*7c478bd9Sstevel@tonic-gate * kernel reserves a fixed number of entries in every topmost level page 120*7c478bd9Sstevel@tonic-gate * table. The values are setup in hat_init() and then copied to every hat 121*7c478bd9Sstevel@tonic-gate * created by hat_alloc(). This means that kernelbase must be: 122*7c478bd9Sstevel@tonic-gate * 123*7c478bd9Sstevel@tonic-gate * 4Meg aligned for 32 bit kernels 124*7c478bd9Sstevel@tonic-gate * 512Gig aligned for x86_64 64 bit kernel 125*7c478bd9Sstevel@tonic-gate * 126*7c478bd9Sstevel@tonic-gate * The PAE 32 bit hat is handled as a special case. Otherwise requiring 1Gig 127*7c478bd9Sstevel@tonic-gate * alignment would use too much VA for the kernel. 128*7c478bd9Sstevel@tonic-gate * 129*7c478bd9Sstevel@tonic-gate */ 130*7c478bd9Sstevel@tonic-gate static uint_t khat_start; /* index of 1st entry in kernel's top ptable */ 131*7c478bd9Sstevel@tonic-gate static uint_t khat_entries; /* number of entries in kernel's top ptable */ 132*7c478bd9Sstevel@tonic-gate 133*7c478bd9Sstevel@tonic-gate #if defined(__i386) 134*7c478bd9Sstevel@tonic-gate 135*7c478bd9Sstevel@tonic-gate static htable_t *khat_pae32_htable = NULL; 136*7c478bd9Sstevel@tonic-gate static uint_t khat_pae32_start; 137*7c478bd9Sstevel@tonic-gate static uint_t khat_pae32_entries; 138*7c478bd9Sstevel@tonic-gate 139*7c478bd9Sstevel@tonic-gate #endif 140*7c478bd9Sstevel@tonic-gate 141*7c478bd9Sstevel@tonic-gate /* 142*7c478bd9Sstevel@tonic-gate * Locks, etc. to control use of the hat reserves when recursively 143*7c478bd9Sstevel@tonic-gate * allocating pagetables for the hat data structures. 144*7c478bd9Sstevel@tonic-gate */ 145*7c478bd9Sstevel@tonic-gate static kmutex_t hat_reserves_lock; 146*7c478bd9Sstevel@tonic-gate static kcondvar_t hat_reserves_cv; 147*7c478bd9Sstevel@tonic-gate kthread_t *hat_reserves_thread; 148*7c478bd9Sstevel@tonic-gate uint_t use_boot_reserve = 1; /* cleared after early boot process */ 149*7c478bd9Sstevel@tonic-gate uint_t can_steal_post_boot = 0; /* set late in boot to enable stealing */ 150*7c478bd9Sstevel@tonic-gate 151*7c478bd9Sstevel@tonic-gate /* 152*7c478bd9Sstevel@tonic-gate * A cpuset for all cpus. This is used for kernel address cross calls, since 153*7c478bd9Sstevel@tonic-gate * the kernel addresses apply to all cpus. 154*7c478bd9Sstevel@tonic-gate */ 155*7c478bd9Sstevel@tonic-gate cpuset_t khat_cpuset; 156*7c478bd9Sstevel@tonic-gate 157*7c478bd9Sstevel@tonic-gate /* 158*7c478bd9Sstevel@tonic-gate * management stuff for hat structures 159*7c478bd9Sstevel@tonic-gate */ 160*7c478bd9Sstevel@tonic-gate kmutex_t hat_list_lock; 161*7c478bd9Sstevel@tonic-gate kcondvar_t hat_list_cv; 162*7c478bd9Sstevel@tonic-gate kmem_cache_t *hat_cache; 163*7c478bd9Sstevel@tonic-gate kmem_cache_t *hat_hash_cache; 164*7c478bd9Sstevel@tonic-gate kmem_cache_t *vlp_hash_cache; 165*7c478bd9Sstevel@tonic-gate 166*7c478bd9Sstevel@tonic-gate /* 167*7c478bd9Sstevel@tonic-gate * Simple statistics 168*7c478bd9Sstevel@tonic-gate */ 169*7c478bd9Sstevel@tonic-gate struct hatstats hatstat; 170*7c478bd9Sstevel@tonic-gate 171*7c478bd9Sstevel@tonic-gate /* 172*7c478bd9Sstevel@tonic-gate * macros to detect addresses in use by kernel only during boot 173*7c478bd9Sstevel@tonic-gate */ 174*7c478bd9Sstevel@tonic-gate #if defined(__amd64) 175*7c478bd9Sstevel@tonic-gate 176*7c478bd9Sstevel@tonic-gate #define BOOT_VA(va) ((va) < kernelbase || \ 177*7c478bd9Sstevel@tonic-gate ((va) >= BOOT_DOUBLEMAP_BASE && \ 178*7c478bd9Sstevel@tonic-gate (va) < BOOT_DOUBLEMAP_BASE + BOOT_DOUBLEMAP_SIZE)) 179*7c478bd9Sstevel@tonic-gate 180*7c478bd9Sstevel@tonic-gate #elif defined(__i386) 181*7c478bd9Sstevel@tonic-gate 182*7c478bd9Sstevel@tonic-gate #define BOOT_VA(va) ((va) < kernelbase) 183*7c478bd9Sstevel@tonic-gate 184*7c478bd9Sstevel@tonic-gate #endif /* __i386 */ 185*7c478bd9Sstevel@tonic-gate 186*7c478bd9Sstevel@tonic-gate /* 187*7c478bd9Sstevel@tonic-gate * useful stuff for atomic access/clearing/setting REF/MOD/RO bits in page_t's. 188*7c478bd9Sstevel@tonic-gate */ 189*7c478bd9Sstevel@tonic-gate extern void atomic_orb(uchar_t *addr, uchar_t val); 190*7c478bd9Sstevel@tonic-gate extern void atomic_andb(uchar_t *addr, uchar_t val); 191*7c478bd9Sstevel@tonic-gate 192*7c478bd9Sstevel@tonic-gate #define PP_GETRM(pp, rmmask) (pp->p_nrm & rmmask) 193*7c478bd9Sstevel@tonic-gate #define PP_ISMOD(pp) PP_GETRM(pp, P_MOD) 194*7c478bd9Sstevel@tonic-gate #define PP_ISREF(pp) PP_GETRM(pp, P_REF) 195*7c478bd9Sstevel@tonic-gate #define PP_ISRO(pp) PP_GETRM(pp, P_RO) 196*7c478bd9Sstevel@tonic-gate 197*7c478bd9Sstevel@tonic-gate #define PP_SETRM(pp, rm) atomic_orb(&(pp->p_nrm), rm) 198*7c478bd9Sstevel@tonic-gate #define PP_SETMOD(pp) PP_SETRM(pp, P_MOD) 199*7c478bd9Sstevel@tonic-gate #define PP_SETREF(pp) PP_SETRM(pp, P_REF) 200*7c478bd9Sstevel@tonic-gate #define PP_SETRO(pp) PP_SETRM(pp, P_RO) 201*7c478bd9Sstevel@tonic-gate 202*7c478bd9Sstevel@tonic-gate #define PP_CLRRM(pp, rm) atomic_andb(&(pp->p_nrm), ~(rm)) 203*7c478bd9Sstevel@tonic-gate #define PP_CLRMOD(pp) PP_CLRRM(pp, P_MOD) 204*7c478bd9Sstevel@tonic-gate #define PP_CLRREF(pp) PP_CLRRM(pp, P_REF) 205*7c478bd9Sstevel@tonic-gate #define PP_CLRRO(pp) PP_CLRRM(pp, P_RO) 206*7c478bd9Sstevel@tonic-gate #define PP_CLRALL(pp) PP_CLRRM(pp, P_MOD | P_REF | P_RO) 207*7c478bd9Sstevel@tonic-gate 208*7c478bd9Sstevel@tonic-gate /* 209*7c478bd9Sstevel@tonic-gate * some useful tracing macros 210*7c478bd9Sstevel@tonic-gate */ 211*7c478bd9Sstevel@tonic-gate 212*7c478bd9Sstevel@tonic-gate int hattrace = 0; 213*7c478bd9Sstevel@tonic-gate #ifdef DEBUG 214*7c478bd9Sstevel@tonic-gate 215*7c478bd9Sstevel@tonic-gate #define HATIN(r, h, a, l) \ 216*7c478bd9Sstevel@tonic-gate if (hattrace) prom_printf("->%s hat=%p, adr=%p, len=%lx\n", #r, h, a, l) 217*7c478bd9Sstevel@tonic-gate 218*7c478bd9Sstevel@tonic-gate #define HATOUT(r, h, a) \ 219*7c478bd9Sstevel@tonic-gate if (hattrace) prom_printf("<-%s hat=%p, adr=%p\n", #r, h, a) 220*7c478bd9Sstevel@tonic-gate #else 221*7c478bd9Sstevel@tonic-gate 222*7c478bd9Sstevel@tonic-gate #define HATIN(r, h, a, l) 223*7c478bd9Sstevel@tonic-gate #define HATOUT(r, h, a) 224*7c478bd9Sstevel@tonic-gate 225*7c478bd9Sstevel@tonic-gate #endif 226*7c478bd9Sstevel@tonic-gate 227*7c478bd9Sstevel@tonic-gate 228*7c478bd9Sstevel@tonic-gate /* 229*7c478bd9Sstevel@tonic-gate * kmem cache constructor for struct hat 230*7c478bd9Sstevel@tonic-gate */ 231*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 232*7c478bd9Sstevel@tonic-gate static int 233*7c478bd9Sstevel@tonic-gate hati_constructor(void *buf, void *handle, int kmflags) 234*7c478bd9Sstevel@tonic-gate { 235*7c478bd9Sstevel@tonic-gate hat_t *hat = buf; 236*7c478bd9Sstevel@tonic-gate 237*7c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 238*7c478bd9Sstevel@tonic-gate bzero(hat->hat_pages_mapped, 239*7c478bd9Sstevel@tonic-gate sizeof (pgcnt_t) * (mmu.max_page_level + 1)); 240*7c478bd9Sstevel@tonic-gate hat->hat_stats = 0; 241*7c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 242*7c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_switch_mutex, NULL, MUTEX_DRIVER, 243*7c478bd9Sstevel@tonic-gate (void *)ipltospl(DISP_LEVEL)); 244*7c478bd9Sstevel@tonic-gate CPUSET_ZERO(hat->hat_cpus); 245*7c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 246*7c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 247*7c478bd9Sstevel@tonic-gate return (0); 248*7c478bd9Sstevel@tonic-gate } 249*7c478bd9Sstevel@tonic-gate 250*7c478bd9Sstevel@tonic-gate /* 251*7c478bd9Sstevel@tonic-gate * Allocate a hat structure for as. We also create the top level 252*7c478bd9Sstevel@tonic-gate * htable and initialize it to contain the kernel hat entries. 253*7c478bd9Sstevel@tonic-gate */ 254*7c478bd9Sstevel@tonic-gate hat_t * 255*7c478bd9Sstevel@tonic-gate hat_alloc(struct as *as) 256*7c478bd9Sstevel@tonic-gate { 257*7c478bd9Sstevel@tonic-gate hat_t *hat; 258*7c478bd9Sstevel@tonic-gate htable_t *ht; /* top level htable */ 259*7c478bd9Sstevel@tonic-gate uint_t use_vlp; 260*7c478bd9Sstevel@tonic-gate 261*7c478bd9Sstevel@tonic-gate /* 262*7c478bd9Sstevel@tonic-gate * Once we start creating user process HATs we can enable 263*7c478bd9Sstevel@tonic-gate * the htable_steal() code. 264*7c478bd9Sstevel@tonic-gate */ 265*7c478bd9Sstevel@tonic-gate if (can_steal_post_boot == 0) 266*7c478bd9Sstevel@tonic-gate can_steal_post_boot = 1; 267*7c478bd9Sstevel@tonic-gate 268*7c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(as, &as->a_lock)); 269*7c478bd9Sstevel@tonic-gate hat = kmem_cache_alloc(hat_cache, KM_SLEEP); 270*7c478bd9Sstevel@tonic-gate hat->hat_as = as; 271*7c478bd9Sstevel@tonic-gate mutex_init(&hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 272*7c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags == 0); 273*7c478bd9Sstevel@tonic-gate 274*7c478bd9Sstevel@tonic-gate /* 275*7c478bd9Sstevel@tonic-gate * a 32 bit process uses a VLP style hat when using PAE 276*7c478bd9Sstevel@tonic-gate */ 277*7c478bd9Sstevel@tonic-gate #if defined(__amd64) 278*7c478bd9Sstevel@tonic-gate use_vlp = (ttoproc(curthread)->p_model == DATAMODEL_ILP32); 279*7c478bd9Sstevel@tonic-gate #elif defined(__i386) 280*7c478bd9Sstevel@tonic-gate use_vlp = mmu.pae_hat; 281*7c478bd9Sstevel@tonic-gate #endif 282*7c478bd9Sstevel@tonic-gate if (use_vlp) { 283*7c478bd9Sstevel@tonic-gate hat->hat_flags = HAT_VLP; 284*7c478bd9Sstevel@tonic-gate bzero(hat->hat_vlp_ptes, VLP_SIZE); 285*7c478bd9Sstevel@tonic-gate } 286*7c478bd9Sstevel@tonic-gate 287*7c478bd9Sstevel@tonic-gate /* 288*7c478bd9Sstevel@tonic-gate * Allocate the htable hash 289*7c478bd9Sstevel@tonic-gate */ 290*7c478bd9Sstevel@tonic-gate if ((hat->hat_flags & HAT_VLP)) { 291*7c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.vlp_hash_cnt; 292*7c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(vlp_hash_cache, KM_SLEEP); 293*7c478bd9Sstevel@tonic-gate } else { 294*7c478bd9Sstevel@tonic-gate hat->hat_num_hash = mmu.hash_cnt; 295*7c478bd9Sstevel@tonic-gate hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_SLEEP); 296*7c478bd9Sstevel@tonic-gate } 297*7c478bd9Sstevel@tonic-gate bzero(hat->hat_ht_hash, hat->hat_num_hash * sizeof (htable_t *)); 298*7c478bd9Sstevel@tonic-gate 299*7c478bd9Sstevel@tonic-gate /* 300*7c478bd9Sstevel@tonic-gate * Initialize Kernel HAT entries at the top of the top level page 301*7c478bd9Sstevel@tonic-gate * table for the new hat. 302*7c478bd9Sstevel@tonic-gate * 303*7c478bd9Sstevel@tonic-gate * Note that we don't call htable_release() for the top level, that 304*7c478bd9Sstevel@tonic-gate * happens when the hat is destroyed in hat_free_end() 305*7c478bd9Sstevel@tonic-gate */ 306*7c478bd9Sstevel@tonic-gate hat->hat_htable = NULL; 307*7c478bd9Sstevel@tonic-gate hat->hat_ht_cached = NULL; 308*7c478bd9Sstevel@tonic-gate ht = htable_create(hat, (uintptr_t)0, TOP_LEVEL(hat), NULL); 309*7c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_VLP)) 310*7c478bd9Sstevel@tonic-gate x86pte_copy(kas.a_hat->hat_htable, ht, khat_start, 311*7c478bd9Sstevel@tonic-gate khat_entries); 312*7c478bd9Sstevel@tonic-gate #if defined(__i386) 313*7c478bd9Sstevel@tonic-gate else if (khat_entries > 0) 314*7c478bd9Sstevel@tonic-gate bcopy(vlp_page + khat_start, hat->hat_vlp_ptes + khat_start, 315*7c478bd9Sstevel@tonic-gate khat_entries * sizeof (x86pte_t)); 316*7c478bd9Sstevel@tonic-gate #endif 317*7c478bd9Sstevel@tonic-gate hat->hat_htable = ht; 318*7c478bd9Sstevel@tonic-gate 319*7c478bd9Sstevel@tonic-gate #if defined(__i386) 320*7c478bd9Sstevel@tonic-gate /* 321*7c478bd9Sstevel@tonic-gate * PAE32 HAT alignment is less restrictive than the others to keep 322*7c478bd9Sstevel@tonic-gate * the kernel from using too much VA. Because of this we may need 323*7c478bd9Sstevel@tonic-gate * one layer further down when kernelbase isn't 1Gig aligned. 324*7c478bd9Sstevel@tonic-gate * See hat_free_end() for the htable_release() that goes with this 325*7c478bd9Sstevel@tonic-gate * htable_create() 326*7c478bd9Sstevel@tonic-gate */ 327*7c478bd9Sstevel@tonic-gate if (khat_pae32_htable != NULL) { 328*7c478bd9Sstevel@tonic-gate ht = htable_create(hat, kernelbase, 329*7c478bd9Sstevel@tonic-gate khat_pae32_htable->ht_level, NULL); 330*7c478bd9Sstevel@tonic-gate x86pte_copy(khat_pae32_htable, ht, khat_pae32_start, 331*7c478bd9Sstevel@tonic-gate khat_pae32_entries); 332*7c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = khat_pae32_entries; 333*7c478bd9Sstevel@tonic-gate } 334*7c478bd9Sstevel@tonic-gate #endif 335*7c478bd9Sstevel@tonic-gate 336*7c478bd9Sstevel@tonic-gate /* 337*7c478bd9Sstevel@tonic-gate * Put it in the global list of all hats (used by stealing, etc.) 338*7c478bd9Sstevel@tonic-gate */ 339*7c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 340*7c478bd9Sstevel@tonic-gate if (kas.a_hat->hat_next != NULL) { 341*7c478bd9Sstevel@tonic-gate hat->hat_next = kas.a_hat->hat_next; 342*7c478bd9Sstevel@tonic-gate hat->hat_prev = kas.a_hat->hat_next->hat_prev; 343*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_next->hat_prev->hat_next = hat; 344*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_next->hat_prev = hat; 345*7c478bd9Sstevel@tonic-gate } else { 346*7c478bd9Sstevel@tonic-gate hat->hat_next = hat; 347*7c478bd9Sstevel@tonic-gate hat->hat_prev = hat; 348*7c478bd9Sstevel@tonic-gate } 349*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat; 350*7c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 351*7c478bd9Sstevel@tonic-gate 352*7c478bd9Sstevel@tonic-gate 353*7c478bd9Sstevel@tonic-gate return (hat); 354*7c478bd9Sstevel@tonic-gate } 355*7c478bd9Sstevel@tonic-gate 356*7c478bd9Sstevel@tonic-gate /* 357*7c478bd9Sstevel@tonic-gate * process has finished executing but as has not been cleaned up yet. 358*7c478bd9Sstevel@tonic-gate */ 359*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 360*7c478bd9Sstevel@tonic-gate void 361*7c478bd9Sstevel@tonic-gate hat_free_start(hat_t *hat) 362*7c478bd9Sstevel@tonic-gate { 363*7c478bd9Sstevel@tonic-gate ASSERT(AS_WRITE_HELD(hat->hat_as, &hat->hat_as->a_lock)); 364*7c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 365*7c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_FREEING; 366*7c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 367*7c478bd9Sstevel@tonic-gate } 368*7c478bd9Sstevel@tonic-gate 369*7c478bd9Sstevel@tonic-gate /* 370*7c478bd9Sstevel@tonic-gate * An address space is being destroyed, so we destroy the associated hat. 371*7c478bd9Sstevel@tonic-gate */ 372*7c478bd9Sstevel@tonic-gate void 373*7c478bd9Sstevel@tonic-gate hat_free_end(hat_t *hat) 374*7c478bd9Sstevel@tonic-gate { 375*7c478bd9Sstevel@tonic-gate int i; 376*7c478bd9Sstevel@tonic-gate kmem_cache_t *cache; 377*7c478bd9Sstevel@tonic-gate 378*7c478bd9Sstevel@tonic-gate #ifdef DEBUG 379*7c478bd9Sstevel@tonic-gate for (i = 0; i <= mmu.max_page_level; i++) 380*7c478bd9Sstevel@tonic-gate ASSERT(hat->hat_pages_mapped[i] == 0); 381*7c478bd9Sstevel@tonic-gate #endif 382*7c478bd9Sstevel@tonic-gate ASSERT(hat->hat_flags & HAT_FREEING); 383*7c478bd9Sstevel@tonic-gate 384*7c478bd9Sstevel@tonic-gate /* 385*7c478bd9Sstevel@tonic-gate * must not be running on the given hat 386*7c478bd9Sstevel@tonic-gate */ 387*7c478bd9Sstevel@tonic-gate ASSERT(CPU->cpu_current_hat != hat); 388*7c478bd9Sstevel@tonic-gate 389*7c478bd9Sstevel@tonic-gate /* 390*7c478bd9Sstevel@tonic-gate * If the hat is currently a stealing victim, wait for the stealing 391*7c478bd9Sstevel@tonic-gate * to finish. Once we've removed it from the list, nobody can 392*7c478bd9Sstevel@tonic-gate * find these htables anymore. 393*7c478bd9Sstevel@tonic-gate */ 394*7c478bd9Sstevel@tonic-gate mutex_enter(&hat_list_lock); 395*7c478bd9Sstevel@tonic-gate while (hat->hat_flags & HAT_VICTIM) 396*7c478bd9Sstevel@tonic-gate cv_wait(&hat_list_cv, &hat_list_lock); 397*7c478bd9Sstevel@tonic-gate hat->hat_next->hat_prev = hat->hat_prev; 398*7c478bd9Sstevel@tonic-gate hat->hat_prev->hat_next = hat->hat_next; 399*7c478bd9Sstevel@tonic-gate if (kas.a_hat->hat_next == hat) { 400*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = hat->hat_next; 401*7c478bd9Sstevel@tonic-gate if (kas.a_hat->hat_next == hat) 402*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 403*7c478bd9Sstevel@tonic-gate } 404*7c478bd9Sstevel@tonic-gate mutex_exit(&hat_list_lock); 405*7c478bd9Sstevel@tonic-gate 406*7c478bd9Sstevel@tonic-gate /* 407*7c478bd9Sstevel@tonic-gate * Make a pass through the htables freeing them all up. 408*7c478bd9Sstevel@tonic-gate */ 409*7c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 410*7c478bd9Sstevel@tonic-gate 411*7c478bd9Sstevel@tonic-gate /* 412*7c478bd9Sstevel@tonic-gate * Decide which kmem cache the hash table came from, then free it. 413*7c478bd9Sstevel@tonic-gate */ 414*7c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) 415*7c478bd9Sstevel@tonic-gate cache = vlp_hash_cache; 416*7c478bd9Sstevel@tonic-gate else 417*7c478bd9Sstevel@tonic-gate cache = hat_hash_cache; 418*7c478bd9Sstevel@tonic-gate kmem_cache_free(cache, hat->hat_ht_hash); 419*7c478bd9Sstevel@tonic-gate hat->hat_ht_hash = NULL; 420*7c478bd9Sstevel@tonic-gate 421*7c478bd9Sstevel@tonic-gate hat->hat_flags = 0; 422*7c478bd9Sstevel@tonic-gate kmem_cache_free(hat_cache, hat); 423*7c478bd9Sstevel@tonic-gate } 424*7c478bd9Sstevel@tonic-gate 425*7c478bd9Sstevel@tonic-gate /* 426*7c478bd9Sstevel@tonic-gate * round kernelbase down to a supported value to use for _userlimit 427*7c478bd9Sstevel@tonic-gate * 428*7c478bd9Sstevel@tonic-gate * userlimit must be aligned down to an entry in the top level htable. 429*7c478bd9Sstevel@tonic-gate * The one exception is for 32 bit HAT's running PAE. 430*7c478bd9Sstevel@tonic-gate */ 431*7c478bd9Sstevel@tonic-gate uintptr_t 432*7c478bd9Sstevel@tonic-gate hat_kernelbase(uintptr_t va) 433*7c478bd9Sstevel@tonic-gate { 434*7c478bd9Sstevel@tonic-gate #if defined(__i386) 435*7c478bd9Sstevel@tonic-gate va &= LEVEL_MASK(1); 436*7c478bd9Sstevel@tonic-gate #endif 437*7c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(va)) 438*7c478bd9Sstevel@tonic-gate panic("_userlimit %p will fall in VA hole\n", (void *)va); 439*7c478bd9Sstevel@tonic-gate return (va); 440*7c478bd9Sstevel@tonic-gate } 441*7c478bd9Sstevel@tonic-gate 442*7c478bd9Sstevel@tonic-gate /* 443*7c478bd9Sstevel@tonic-gate * Initialize hat data structures based on processor MMU information. 444*7c478bd9Sstevel@tonic-gate */ 445*7c478bd9Sstevel@tonic-gate void 446*7c478bd9Sstevel@tonic-gate mmu_init(void) 447*7c478bd9Sstevel@tonic-gate { 448*7c478bd9Sstevel@tonic-gate uint_t max_htables; 449*7c478bd9Sstevel@tonic-gate uint_t pa_bits; 450*7c478bd9Sstevel@tonic-gate uint_t va_bits; 451*7c478bd9Sstevel@tonic-gate int i; 452*7c478bd9Sstevel@tonic-gate 453*7c478bd9Sstevel@tonic-gate /* 454*7c478bd9Sstevel@tonic-gate * if CPU enabled the page table global bit, use it for the kernel 455*7c478bd9Sstevel@tonic-gate * This is bit 7 in CR4 (PGE - Page Global Enable) 456*7c478bd9Sstevel@tonic-gate */ 457*7c478bd9Sstevel@tonic-gate if ((x86_feature & X86_PGE) != 0 && (getcr4() & 0x80) != 0) 458*7c478bd9Sstevel@tonic-gate mmu.pt_global = PT_GLOBAL; 459*7c478bd9Sstevel@tonic-gate 460*7c478bd9Sstevel@tonic-gate /* 461*7c478bd9Sstevel@tonic-gate * We use PAE except when we aren't on an AMD64 and this is 462*7c478bd9Sstevel@tonic-gate * a 32 bit kernel with all physical addresses less than 4 Gig. 463*7c478bd9Sstevel@tonic-gate */ 464*7c478bd9Sstevel@tonic-gate mmu.pae_hat = 1; 465*7c478bd9Sstevel@tonic-gate if (x86_feature & X86_NX) { 466*7c478bd9Sstevel@tonic-gate mmu.pt_nx = PT_NX; 467*7c478bd9Sstevel@tonic-gate } else { 468*7c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 469*7c478bd9Sstevel@tonic-gate #if defined(__i386) 470*7c478bd9Sstevel@tonic-gate if (!PFN_ABOVE4G(physmax)) 471*7c478bd9Sstevel@tonic-gate mmu.pae_hat = 0; 472*7c478bd9Sstevel@tonic-gate #endif 473*7c478bd9Sstevel@tonic-gate } 474*7c478bd9Sstevel@tonic-gate 475*7c478bd9Sstevel@tonic-gate #if defined(__i386) 476*7c478bd9Sstevel@tonic-gate /* 477*7c478bd9Sstevel@tonic-gate * Setting one of these two lets you force testing of the different 478*7c478bd9Sstevel@tonic-gate * hat modes for 32 bit, regardless of the hardware setup. 479*7c478bd9Sstevel@tonic-gate */ 480*7c478bd9Sstevel@tonic-gate if (force_pae_on) { 481*7c478bd9Sstevel@tonic-gate mmu.pae_hat = 1; 482*7c478bd9Sstevel@tonic-gate } else if (force_pae_off) { 483*7c478bd9Sstevel@tonic-gate mmu.pae_hat = 0; 484*7c478bd9Sstevel@tonic-gate mmu.pt_nx = 0; 485*7c478bd9Sstevel@tonic-gate } 486*7c478bd9Sstevel@tonic-gate #endif 487*7c478bd9Sstevel@tonic-gate 488*7c478bd9Sstevel@tonic-gate /* 489*7c478bd9Sstevel@tonic-gate * Use CPU info to set various MMU parameters 490*7c478bd9Sstevel@tonic-gate */ 491*7c478bd9Sstevel@tonic-gate cpuid_get_addrsize(CPU, &pa_bits, &va_bits); 492*7c478bd9Sstevel@tonic-gate 493*7c478bd9Sstevel@tonic-gate if (va_bits < sizeof (void *) * NBBY) { 494*7c478bd9Sstevel@tonic-gate mmu.hole_start = (1ul << (va_bits - 1)); 495*7c478bd9Sstevel@tonic-gate mmu.hole_end = 0ul - mmu.hole_start - 1; 496*7c478bd9Sstevel@tonic-gate } else { 497*7c478bd9Sstevel@tonic-gate mmu.hole_end = 0; 498*7c478bd9Sstevel@tonic-gate mmu.hole_start = mmu.hole_end - 1; 499*7c478bd9Sstevel@tonic-gate } 500*7c478bd9Sstevel@tonic-gate #if defined(OPTERON_ERRATUM_121) 501*7c478bd9Sstevel@tonic-gate /* 502*7c478bd9Sstevel@tonic-gate * If erratum 121 has already been detected at this time, hole_start 503*7c478bd9Sstevel@tonic-gate * contains the value to be subtracted from mmu.hole_start. 504*7c478bd9Sstevel@tonic-gate */ 505*7c478bd9Sstevel@tonic-gate ASSERT(hole_start == 0 || opteron_erratum_121 != 0); 506*7c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start - hole_start; 507*7c478bd9Sstevel@tonic-gate #else 508*7c478bd9Sstevel@tonic-gate hole_start = mmu.hole_start; 509*7c478bd9Sstevel@tonic-gate #endif 510*7c478bd9Sstevel@tonic-gate hole_end = mmu.hole_end; 511*7c478bd9Sstevel@tonic-gate 512*7c478bd9Sstevel@tonic-gate mmu.highest_pfn = mmu_btop((1ull << pa_bits) - 1); 513*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat == 0 && pa_bits > 32) 514*7c478bd9Sstevel@tonic-gate mmu.highest_pfn = PFN_4G - 1; 515*7c478bd9Sstevel@tonic-gate 516*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 517*7c478bd9Sstevel@tonic-gate mmu.pte_size = 8; /* 8 byte PTEs */ 518*7c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 3; 519*7c478bd9Sstevel@tonic-gate } else { 520*7c478bd9Sstevel@tonic-gate mmu.pte_size = 4; /* 4 byte PTEs */ 521*7c478bd9Sstevel@tonic-gate mmu.pte_size_shift = 2; 522*7c478bd9Sstevel@tonic-gate } 523*7c478bd9Sstevel@tonic-gate 524*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat && (x86_feature & X86_PAE) == 0) 525*7c478bd9Sstevel@tonic-gate panic("Processor does not support PAE"); 526*7c478bd9Sstevel@tonic-gate 527*7c478bd9Sstevel@tonic-gate if ((x86_feature & X86_CX8) == 0) 528*7c478bd9Sstevel@tonic-gate panic("Processor does not support cmpxchg8b instruction"); 529*7c478bd9Sstevel@tonic-gate 530*7c478bd9Sstevel@tonic-gate /* 531*7c478bd9Sstevel@tonic-gate * Initialize parameters based on the 64 or 32 bit kernels and 532*7c478bd9Sstevel@tonic-gate * for the 32 bit kernel decide if we should use PAE. 533*7c478bd9Sstevel@tonic-gate */ 534*7c478bd9Sstevel@tonic-gate if (x86_feature & X86_LARGEPAGE) 535*7c478bd9Sstevel@tonic-gate mmu.max_page_level = 1; 536*7c478bd9Sstevel@tonic-gate else 537*7c478bd9Sstevel@tonic-gate mmu.max_page_level = 0; 538*7c478bd9Sstevel@tonic-gate mmu_page_sizes = mmu.max_page_level + 1; 539*7c478bd9Sstevel@tonic-gate mmu_exported_page_sizes = mmu_page_sizes; 540*7c478bd9Sstevel@tonic-gate 541*7c478bd9Sstevel@tonic-gate #if defined(__amd64) 542*7c478bd9Sstevel@tonic-gate 543*7c478bd9Sstevel@tonic-gate mmu.num_level = 4; 544*7c478bd9Sstevel@tonic-gate mmu.max_level = 3; 545*7c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 546*7c478bd9Sstevel@tonic-gate mmu.top_level_count = 512; 547*7c478bd9Sstevel@tonic-gate 548*7c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 549*7c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 550*7c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 551*7c478bd9Sstevel@tonic-gate mmu.level_shift[3] = 39; 552*7c478bd9Sstevel@tonic-gate 553*7c478bd9Sstevel@tonic-gate #elif defined(__i386) 554*7c478bd9Sstevel@tonic-gate 555*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 556*7c478bd9Sstevel@tonic-gate mmu.num_level = 3; 557*7c478bd9Sstevel@tonic-gate mmu.max_level = 2; 558*7c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 512; 559*7c478bd9Sstevel@tonic-gate mmu.top_level_count = 4; 560*7c478bd9Sstevel@tonic-gate 561*7c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 562*7c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 21; 563*7c478bd9Sstevel@tonic-gate mmu.level_shift[2] = 30; 564*7c478bd9Sstevel@tonic-gate 565*7c478bd9Sstevel@tonic-gate } else { 566*7c478bd9Sstevel@tonic-gate mmu.num_level = 2; 567*7c478bd9Sstevel@tonic-gate mmu.max_level = 1; 568*7c478bd9Sstevel@tonic-gate mmu.ptes_per_table = 1024; 569*7c478bd9Sstevel@tonic-gate mmu.top_level_count = 1024; 570*7c478bd9Sstevel@tonic-gate 571*7c478bd9Sstevel@tonic-gate mmu.level_shift[0] = 12; 572*7c478bd9Sstevel@tonic-gate mmu.level_shift[1] = 22; 573*7c478bd9Sstevel@tonic-gate } 574*7c478bd9Sstevel@tonic-gate 575*7c478bd9Sstevel@tonic-gate #endif /* __i386 */ 576*7c478bd9Sstevel@tonic-gate 577*7c478bd9Sstevel@tonic-gate for (i = 0; i < mmu.num_level; ++i) { 578*7c478bd9Sstevel@tonic-gate mmu.level_size[i] = 1UL << mmu.level_shift[i]; 579*7c478bd9Sstevel@tonic-gate mmu.level_offset[i] = mmu.level_size[i] - 1; 580*7c478bd9Sstevel@tonic-gate mmu.level_mask[i] = ~mmu.level_offset[i]; 581*7c478bd9Sstevel@tonic-gate } 582*7c478bd9Sstevel@tonic-gate 583*7c478bd9Sstevel@tonic-gate mmu.pte_bits[0] = PT_VALID; 584*7c478bd9Sstevel@tonic-gate for (i = 1; i <= mmu.max_page_level; ++i) 585*7c478bd9Sstevel@tonic-gate mmu.pte_bits[i] = PT_VALID | PT_PAGESIZE; 586*7c478bd9Sstevel@tonic-gate 587*7c478bd9Sstevel@tonic-gate /* 588*7c478bd9Sstevel@tonic-gate * NOTE Legacy 32 bit PAE mode only has the P_VALID bit at top level. 589*7c478bd9Sstevel@tonic-gate */ 590*7c478bd9Sstevel@tonic-gate for (i = 1; i < mmu.num_level; ++i) 591*7c478bd9Sstevel@tonic-gate mmu.ptp_bits[i] = PT_PTPBITS; 592*7c478bd9Sstevel@tonic-gate #if defined(__i386) 593*7c478bd9Sstevel@tonic-gate mmu.ptp_bits[2] = PT_VALID; 594*7c478bd9Sstevel@tonic-gate #endif 595*7c478bd9Sstevel@tonic-gate 596*7c478bd9Sstevel@tonic-gate /* 597*7c478bd9Sstevel@tonic-gate * Compute how many hash table entries to have per process for htables. 598*7c478bd9Sstevel@tonic-gate * We start with 1 page's worth of entries. 599*7c478bd9Sstevel@tonic-gate * 600*7c478bd9Sstevel@tonic-gate * If physical memory is small, reduce the amount need to cover it. 601*7c478bd9Sstevel@tonic-gate */ 602*7c478bd9Sstevel@tonic-gate max_htables = physmax / mmu.ptes_per_table; 603*7c478bd9Sstevel@tonic-gate mmu.hash_cnt = MMU_PAGESIZE / sizeof (htable_t *); 604*7c478bd9Sstevel@tonic-gate while (mmu.hash_cnt > 16 && mmu.hash_cnt >= max_htables) 605*7c478bd9Sstevel@tonic-gate mmu.hash_cnt >>= 1; 606*7c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt = mmu.hash_cnt; 607*7c478bd9Sstevel@tonic-gate 608*7c478bd9Sstevel@tonic-gate #if defined(__amd64) 609*7c478bd9Sstevel@tonic-gate /* 610*7c478bd9Sstevel@tonic-gate * If running in 64 bits and physical memory is large, 611*7c478bd9Sstevel@tonic-gate * increase the size of the cache to cover all of memory for 612*7c478bd9Sstevel@tonic-gate * a 64 bit process. 613*7c478bd9Sstevel@tonic-gate */ 614*7c478bd9Sstevel@tonic-gate #define HASH_MAX_LENGTH 4 615*7c478bd9Sstevel@tonic-gate while (mmu.hash_cnt * HASH_MAX_LENGTH < max_htables) 616*7c478bd9Sstevel@tonic-gate mmu.hash_cnt <<= 1; 617*7c478bd9Sstevel@tonic-gate #endif 618*7c478bd9Sstevel@tonic-gate 619*7c478bd9Sstevel@tonic-gate /* 620*7c478bd9Sstevel@tonic-gate * This code knows that there are only 2 pagesizes. 621*7c478bd9Sstevel@tonic-gate * We ignore 4MB (non-PAE) for now. The value is only used 622*7c478bd9Sstevel@tonic-gate * for optimizing demaps across large ranges. 623*7c478bd9Sstevel@tonic-gate * These return zero if no information is known. 624*7c478bd9Sstevel@tonic-gate */ 625*7c478bd9Sstevel@tonic-gate mmu.tlb_entries[0] = cpuid_get_dtlb_nent(NULL, MMU_PAGESIZE); 626*7c478bd9Sstevel@tonic-gate mmu.tlb_entries[1] = cpuid_get_dtlb_nent(NULL, 2 * 1024 * 1024); 627*7c478bd9Sstevel@tonic-gate } 628*7c478bd9Sstevel@tonic-gate 629*7c478bd9Sstevel@tonic-gate 630*7c478bd9Sstevel@tonic-gate /* 631*7c478bd9Sstevel@tonic-gate * initialize hat data structures 632*7c478bd9Sstevel@tonic-gate */ 633*7c478bd9Sstevel@tonic-gate void 634*7c478bd9Sstevel@tonic-gate hat_init() 635*7c478bd9Sstevel@tonic-gate { 636*7c478bd9Sstevel@tonic-gate #if defined(__i386) 637*7c478bd9Sstevel@tonic-gate /* 638*7c478bd9Sstevel@tonic-gate * _userlimit must be aligned correctly 639*7c478bd9Sstevel@tonic-gate */ 640*7c478bd9Sstevel@tonic-gate if ((_userlimit & LEVEL_MASK(1)) != _userlimit) { 641*7c478bd9Sstevel@tonic-gate prom_printf("hat_init(): _userlimit=%p, not aligned at %p\n", 642*7c478bd9Sstevel@tonic-gate (void *)_userlimit, (void *)LEVEL_SIZE(1)); 643*7c478bd9Sstevel@tonic-gate halt("hat_init(): Unable to continue"); 644*7c478bd9Sstevel@tonic-gate } 645*7c478bd9Sstevel@tonic-gate #endif 646*7c478bd9Sstevel@tonic-gate 647*7c478bd9Sstevel@tonic-gate cv_init(&hat_list_cv, NULL, CV_DEFAULT, NULL); 648*7c478bd9Sstevel@tonic-gate 649*7c478bd9Sstevel@tonic-gate /* 650*7c478bd9Sstevel@tonic-gate * initialize kmem caches 651*7c478bd9Sstevel@tonic-gate */ 652*7c478bd9Sstevel@tonic-gate htable_init(); 653*7c478bd9Sstevel@tonic-gate hment_init(); 654*7c478bd9Sstevel@tonic-gate 655*7c478bd9Sstevel@tonic-gate hat_cache = kmem_cache_create("hat_t", 656*7c478bd9Sstevel@tonic-gate sizeof (hat_t), 0, hati_constructor, NULL, NULL, 657*7c478bd9Sstevel@tonic-gate NULL, 0, 0); 658*7c478bd9Sstevel@tonic-gate 659*7c478bd9Sstevel@tonic-gate hat_hash_cache = kmem_cache_create("HatHash", 660*7c478bd9Sstevel@tonic-gate mmu.hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 661*7c478bd9Sstevel@tonic-gate NULL, 0, 0); 662*7c478bd9Sstevel@tonic-gate 663*7c478bd9Sstevel@tonic-gate /* 664*7c478bd9Sstevel@tonic-gate * VLP hats can use a smaller hash table size on large memroy machines 665*7c478bd9Sstevel@tonic-gate */ 666*7c478bd9Sstevel@tonic-gate if (mmu.hash_cnt == mmu.vlp_hash_cnt) { 667*7c478bd9Sstevel@tonic-gate vlp_hash_cache = hat_hash_cache; 668*7c478bd9Sstevel@tonic-gate } else { 669*7c478bd9Sstevel@tonic-gate vlp_hash_cache = kmem_cache_create("HatVlpHash", 670*7c478bd9Sstevel@tonic-gate mmu.vlp_hash_cnt * sizeof (htable_t *), 0, NULL, NULL, NULL, 671*7c478bd9Sstevel@tonic-gate NULL, 0, 0); 672*7c478bd9Sstevel@tonic-gate } 673*7c478bd9Sstevel@tonic-gate 674*7c478bd9Sstevel@tonic-gate /* 675*7c478bd9Sstevel@tonic-gate * Set up the kernel's hat 676*7c478bd9Sstevel@tonic-gate */ 677*7c478bd9Sstevel@tonic-gate AS_LOCK_ENTER(&kas, &kas.a_lock, RW_WRITER); 678*7c478bd9Sstevel@tonic-gate kas.a_hat = kmem_cache_alloc(hat_cache, KM_NOSLEEP); 679*7c478bd9Sstevel@tonic-gate mutex_init(&kas.a_hat->hat_mutex, NULL, MUTEX_DEFAULT, NULL); 680*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_as = &kas; 681*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_flags = 0; 682*7c478bd9Sstevel@tonic-gate AS_LOCK_EXIT(&kas, &kas.a_lock); 683*7c478bd9Sstevel@tonic-gate 684*7c478bd9Sstevel@tonic-gate CPUSET_ZERO(khat_cpuset); 685*7c478bd9Sstevel@tonic-gate CPUSET_ADD(khat_cpuset, CPU->cpu_id); 686*7c478bd9Sstevel@tonic-gate 687*7c478bd9Sstevel@tonic-gate /* 688*7c478bd9Sstevel@tonic-gate * The kernel hat's next pointer serves as the head of the hat list . 689*7c478bd9Sstevel@tonic-gate */ 690*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_next = NULL; 691*7c478bd9Sstevel@tonic-gate 692*7c478bd9Sstevel@tonic-gate /* 693*7c478bd9Sstevel@tonic-gate * Allocate an htable hash bucket for the kernel 694*7c478bd9Sstevel@tonic-gate * XX64 - tune for 64 bit procs 695*7c478bd9Sstevel@tonic-gate */ 696*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_num_hash = mmu.hash_cnt; 697*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_hash = kmem_cache_alloc(hat_hash_cache, KM_NOSLEEP); 698*7c478bd9Sstevel@tonic-gate bzero(kas.a_hat->hat_ht_hash, mmu.hash_cnt * sizeof (htable_t *)); 699*7c478bd9Sstevel@tonic-gate 700*7c478bd9Sstevel@tonic-gate /* 701*7c478bd9Sstevel@tonic-gate * zero out the top level and cached htable pointers 702*7c478bd9Sstevel@tonic-gate */ 703*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_ht_cached = NULL; 704*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable = NULL; 705*7c478bd9Sstevel@tonic-gate } 706*7c478bd9Sstevel@tonic-gate 707*7c478bd9Sstevel@tonic-gate /* 708*7c478bd9Sstevel@tonic-gate * Prepare CPU specific pagetables for VLP processes on 64 bit kernels. 709*7c478bd9Sstevel@tonic-gate * 710*7c478bd9Sstevel@tonic-gate * Each CPU has a set of 2 pagetables that are reused for any 32 bit 711*7c478bd9Sstevel@tonic-gate * process it runs. They are the top level pagetable, hci_vlp_l3ptes, and 712*7c478bd9Sstevel@tonic-gate * the next to top level table for the bottom 512 Gig, hci_vlp_l2ptes. 713*7c478bd9Sstevel@tonic-gate */ 714*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 715*7c478bd9Sstevel@tonic-gate static void 716*7c478bd9Sstevel@tonic-gate hat_vlp_setup(struct cpu *cpu) 717*7c478bd9Sstevel@tonic-gate { 718*7c478bd9Sstevel@tonic-gate #if defined(__amd64) 719*7c478bd9Sstevel@tonic-gate struct hat_cpu_info *hci = cpu->cpu_hat_info; 720*7c478bd9Sstevel@tonic-gate pfn_t pfn; 721*7c478bd9Sstevel@tonic-gate 722*7c478bd9Sstevel@tonic-gate /* 723*7c478bd9Sstevel@tonic-gate * allocate the level==2 page table for the bottom most 724*7c478bd9Sstevel@tonic-gate * 512Gig of address space (this is where 32 bit apps live) 725*7c478bd9Sstevel@tonic-gate */ 726*7c478bd9Sstevel@tonic-gate ASSERT(hci != NULL); 727*7c478bd9Sstevel@tonic-gate hci->hci_vlp_l2ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 728*7c478bd9Sstevel@tonic-gate 729*7c478bd9Sstevel@tonic-gate /* 730*7c478bd9Sstevel@tonic-gate * Allocate a top level pagetable and copy the kernel's 731*7c478bd9Sstevel@tonic-gate * entries into it. Then link in hci_vlp_l2ptes in the 1st entry. 732*7c478bd9Sstevel@tonic-gate */ 733*7c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes = kmem_zalloc(MMU_PAGESIZE, KM_SLEEP); 734*7c478bd9Sstevel@tonic-gate hci->hci_vlp_pfn = 735*7c478bd9Sstevel@tonic-gate hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l3ptes); 736*7c478bd9Sstevel@tonic-gate ASSERT(hci->hci_vlp_pfn != PFN_INVALID); 737*7c478bd9Sstevel@tonic-gate bcopy(vlp_page + khat_start, hci->hci_vlp_l3ptes + khat_start, 738*7c478bd9Sstevel@tonic-gate khat_entries * sizeof (x86pte_t)); 739*7c478bd9Sstevel@tonic-gate 740*7c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, (caddr_t)hci->hci_vlp_l2ptes); 741*7c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 742*7c478bd9Sstevel@tonic-gate hci->hci_vlp_l3ptes[0] = MAKEPTP(pfn, 2); 743*7c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 744*7c478bd9Sstevel@tonic-gate } 745*7c478bd9Sstevel@tonic-gate 746*7c478bd9Sstevel@tonic-gate /* 747*7c478bd9Sstevel@tonic-gate * Finish filling in the kernel hat. 748*7c478bd9Sstevel@tonic-gate * Pre fill in all top level kernel page table entries for the kernel's 749*7c478bd9Sstevel@tonic-gate * part of the address range. From this point on we can't use any new 750*7c478bd9Sstevel@tonic-gate * kernel large pages if they need PTE's at max_level 751*7c478bd9Sstevel@tonic-gate */ 752*7c478bd9Sstevel@tonic-gate void 753*7c478bd9Sstevel@tonic-gate hat_init_finish(void) 754*7c478bd9Sstevel@tonic-gate { 755*7c478bd9Sstevel@tonic-gate htable_t *top = kas.a_hat->hat_htable; 756*7c478bd9Sstevel@tonic-gate htable_t *ht; 757*7c478bd9Sstevel@tonic-gate uint_t e; 758*7c478bd9Sstevel@tonic-gate x86pte_t pte; 759*7c478bd9Sstevel@tonic-gate uintptr_t va = kernelbase; 760*7c478bd9Sstevel@tonic-gate 761*7c478bd9Sstevel@tonic-gate 762*7c478bd9Sstevel@tonic-gate #if defined(__i386) 763*7c478bd9Sstevel@tonic-gate ASSERT((va & LEVEL_MASK(1)) == va); 764*7c478bd9Sstevel@tonic-gate 765*7c478bd9Sstevel@tonic-gate /* 766*7c478bd9Sstevel@tonic-gate * Deal with kernelbase not 1Gig aligned for 32 bit PAE hats. 767*7c478bd9Sstevel@tonic-gate */ 768*7c478bd9Sstevel@tonic-gate if (!mmu.pae_hat || (va & LEVEL_OFFSET(mmu.max_level)) == 0) { 769*7c478bd9Sstevel@tonic-gate khat_pae32_htable = NULL; 770*7c478bd9Sstevel@tonic-gate } else { 771*7c478bd9Sstevel@tonic-gate ASSERT(mmu.max_level == 2); 772*7c478bd9Sstevel@tonic-gate ASSERT((va & LEVEL_OFFSET(mmu.max_level - 1)) == 0); 773*7c478bd9Sstevel@tonic-gate khat_pae32_htable = 774*7c478bd9Sstevel@tonic-gate htable_create(kas.a_hat, va, mmu.max_level - 1, NULL); 775*7c478bd9Sstevel@tonic-gate khat_pae32_start = htable_va2entry(va, khat_pae32_htable); 776*7c478bd9Sstevel@tonic-gate khat_pae32_entries = mmu.ptes_per_table - khat_pae32_start; 777*7c478bd9Sstevel@tonic-gate for (e = khat_pae32_start; e < mmu.ptes_per_table; 778*7c478bd9Sstevel@tonic-gate ++e, va += LEVEL_SIZE(mmu.max_level - 1)) { 779*7c478bd9Sstevel@tonic-gate pte = x86pte_get(khat_pae32_htable, e); 780*7c478bd9Sstevel@tonic-gate if (PTE_ISVALID(pte)) 781*7c478bd9Sstevel@tonic-gate continue; 782*7c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, mmu.max_level - 2, 783*7c478bd9Sstevel@tonic-gate NULL); 784*7c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 785*7c478bd9Sstevel@tonic-gate } 786*7c478bd9Sstevel@tonic-gate } 787*7c478bd9Sstevel@tonic-gate #endif 788*7c478bd9Sstevel@tonic-gate 789*7c478bd9Sstevel@tonic-gate /* 790*7c478bd9Sstevel@tonic-gate * The kernel hat will need fixed values in the highest level 791*7c478bd9Sstevel@tonic-gate * ptable for copying to all other hat's. This implies 792*7c478bd9Sstevel@tonic-gate * alignment restrictions on _userlimit. 793*7c478bd9Sstevel@tonic-gate * 794*7c478bd9Sstevel@tonic-gate * Note we don't htable_release() these htables. This keeps them 795*7c478bd9Sstevel@tonic-gate * from ever being stolen or free'd. 796*7c478bd9Sstevel@tonic-gate * 797*7c478bd9Sstevel@tonic-gate * top_level_count is used instead of ptes_per_table, since 798*7c478bd9Sstevel@tonic-gate * on 32-bit PAE we only have 4 usable entries at the top level ptable. 799*7c478bd9Sstevel@tonic-gate */ 800*7c478bd9Sstevel@tonic-gate if (va == 0) 801*7c478bd9Sstevel@tonic-gate khat_start = mmu.top_level_count; 802*7c478bd9Sstevel@tonic-gate else 803*7c478bd9Sstevel@tonic-gate khat_start = htable_va2entry(va, kas.a_hat->hat_htable); 804*7c478bd9Sstevel@tonic-gate khat_entries = mmu.top_level_count - khat_start; 805*7c478bd9Sstevel@tonic-gate for (e = khat_start; e < mmu.top_level_count; 806*7c478bd9Sstevel@tonic-gate ++e, va += LEVEL_SIZE(mmu.max_level)) { 807*7c478bd9Sstevel@tonic-gate pte = x86pte_get(top, e); 808*7c478bd9Sstevel@tonic-gate if (PTE_ISVALID(pte)) 809*7c478bd9Sstevel@tonic-gate continue; 810*7c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, mmu.max_level - 1, NULL); 811*7c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 812*7c478bd9Sstevel@tonic-gate } 813*7c478bd9Sstevel@tonic-gate 814*7c478bd9Sstevel@tonic-gate /* 815*7c478bd9Sstevel@tonic-gate * We are now effectively running on the kernel hat. 816*7c478bd9Sstevel@tonic-gate * Clearing use_boot_reserve shuts off using the pre-allocated boot 817*7c478bd9Sstevel@tonic-gate * reserve for all HAT allocations. From here on, the reserves are 818*7c478bd9Sstevel@tonic-gate * only used when mapping in memory for the hat's own allocations. 819*7c478bd9Sstevel@tonic-gate */ 820*7c478bd9Sstevel@tonic-gate use_boot_reserve = 0; 821*7c478bd9Sstevel@tonic-gate htable_adjust_reserve(); 822*7c478bd9Sstevel@tonic-gate 823*7c478bd9Sstevel@tonic-gate /* 824*7c478bd9Sstevel@tonic-gate * 32 bit kernels use only 4 of the 512 entries in its top level 825*7c478bd9Sstevel@tonic-gate * pagetable. We'll use the remainder for the "per CPU" page tables 826*7c478bd9Sstevel@tonic-gate * for VLP processes. 827*7c478bd9Sstevel@tonic-gate * 828*7c478bd9Sstevel@tonic-gate * We map the top level kernel pagetable into the kernel's AS to make 829*7c478bd9Sstevel@tonic-gate * it easy to use bcopy for kernel entry PTEs. 830*7c478bd9Sstevel@tonic-gate * 831*7c478bd9Sstevel@tonic-gate * We were guaranteed to get a physical address < 4Gig, since the 32 bit 832*7c478bd9Sstevel@tonic-gate * boot loader uses non-PAE page tables. 833*7c478bd9Sstevel@tonic-gate */ 834*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 835*7c478bd9Sstevel@tonic-gate vlp_page = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 836*7c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, (caddr_t)vlp_page, MMU_PAGESIZE, 837*7c478bd9Sstevel@tonic-gate kas.a_hat->hat_htable->ht_pfn, 838*7c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | HAT_NOSYNC | HAT_UNORDERED_OK, 839*7c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 840*7c478bd9Sstevel@tonic-gate } 841*7c478bd9Sstevel@tonic-gate hat_vlp_setup(CPU); 842*7c478bd9Sstevel@tonic-gate } 843*7c478bd9Sstevel@tonic-gate 844*7c478bd9Sstevel@tonic-gate /* 845*7c478bd9Sstevel@tonic-gate * On 32 bit PAE mode, PTE's are 64 bits, but ordinary atomic memory references 846*7c478bd9Sstevel@tonic-gate * are 32 bit, so for safety we must use cas64() to install these. 847*7c478bd9Sstevel@tonic-gate */ 848*7c478bd9Sstevel@tonic-gate #ifdef __i386 849*7c478bd9Sstevel@tonic-gate static void 850*7c478bd9Sstevel@tonic-gate reload_pae32(hat_t *hat, cpu_t *cpu) 851*7c478bd9Sstevel@tonic-gate { 852*7c478bd9Sstevel@tonic-gate x86pte_t *src; 853*7c478bd9Sstevel@tonic-gate x86pte_t *dest; 854*7c478bd9Sstevel@tonic-gate x86pte_t pte; 855*7c478bd9Sstevel@tonic-gate int i; 856*7c478bd9Sstevel@tonic-gate 857*7c478bd9Sstevel@tonic-gate /* 858*7c478bd9Sstevel@tonic-gate * Load the 4 entries of the level 2 page table into this 859*7c478bd9Sstevel@tonic-gate * cpu's range of the vlp_page and point cr3 at them. 860*7c478bd9Sstevel@tonic-gate */ 861*7c478bd9Sstevel@tonic-gate ASSERT(mmu.pae_hat); 862*7c478bd9Sstevel@tonic-gate src = hat->hat_vlp_ptes; 863*7c478bd9Sstevel@tonic-gate dest = vlp_page + (cpu->cpu_id + 1) * VLP_NUM_PTES; 864*7c478bd9Sstevel@tonic-gate for (i = 0; i < VLP_NUM_PTES; ++i) { 865*7c478bd9Sstevel@tonic-gate for (;;) { 866*7c478bd9Sstevel@tonic-gate pte = dest[i]; 867*7c478bd9Sstevel@tonic-gate if (pte == src[i]) 868*7c478bd9Sstevel@tonic-gate break; 869*7c478bd9Sstevel@tonic-gate if (cas64(dest + i, pte, src[i]) != src[i]) 870*7c478bd9Sstevel@tonic-gate break; 871*7c478bd9Sstevel@tonic-gate } 872*7c478bd9Sstevel@tonic-gate } 873*7c478bd9Sstevel@tonic-gate } 874*7c478bd9Sstevel@tonic-gate #endif 875*7c478bd9Sstevel@tonic-gate 876*7c478bd9Sstevel@tonic-gate /* 877*7c478bd9Sstevel@tonic-gate * Switch to a new active hat, maintaining bit masks to track active CPUs. 878*7c478bd9Sstevel@tonic-gate */ 879*7c478bd9Sstevel@tonic-gate void 880*7c478bd9Sstevel@tonic-gate hat_switch(hat_t *hat) 881*7c478bd9Sstevel@tonic-gate { 882*7c478bd9Sstevel@tonic-gate uintptr_t newcr3; 883*7c478bd9Sstevel@tonic-gate cpu_t *cpu = CPU; 884*7c478bd9Sstevel@tonic-gate hat_t *old = cpu->cpu_current_hat; 885*7c478bd9Sstevel@tonic-gate 886*7c478bd9Sstevel@tonic-gate /* 887*7c478bd9Sstevel@tonic-gate * set up this information first, so we don't miss any cross calls 888*7c478bd9Sstevel@tonic-gate */ 889*7c478bd9Sstevel@tonic-gate if (old != NULL) { 890*7c478bd9Sstevel@tonic-gate if (old == hat) 891*7c478bd9Sstevel@tonic-gate return; 892*7c478bd9Sstevel@tonic-gate if (old != kas.a_hat) 893*7c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_DEL(old->hat_cpus, cpu->cpu_id); 894*7c478bd9Sstevel@tonic-gate } 895*7c478bd9Sstevel@tonic-gate 896*7c478bd9Sstevel@tonic-gate /* 897*7c478bd9Sstevel@tonic-gate * Wait for any in flight pagetable invalidates on this hat to finish. 898*7c478bd9Sstevel@tonic-gate * This is a spin lock at DISP_LEVEL 899*7c478bd9Sstevel@tonic-gate */ 900*7c478bd9Sstevel@tonic-gate if (hat != kas.a_hat) { 901*7c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_switch_mutex); 902*7c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(hat->hat_cpus, cpu->cpu_id); 903*7c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_switch_mutex); 904*7c478bd9Sstevel@tonic-gate } 905*7c478bd9Sstevel@tonic-gate cpu->cpu_current_hat = hat; 906*7c478bd9Sstevel@tonic-gate 907*7c478bd9Sstevel@tonic-gate /* 908*7c478bd9Sstevel@tonic-gate * now go ahead and load cr3 909*7c478bd9Sstevel@tonic-gate */ 910*7c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 911*7c478bd9Sstevel@tonic-gate #if defined(__amd64) 912*7c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = cpu->cpu_hat_info->hci_vlp_l2ptes; 913*7c478bd9Sstevel@tonic-gate 914*7c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 915*7c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(cpu->cpu_hat_info->hci_vlp_pfn); 916*7c478bd9Sstevel@tonic-gate #elif defined(__i386) 917*7c478bd9Sstevel@tonic-gate reload_pae32(hat, cpu); 918*7c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(kas.a_hat->hat_htable->ht_pfn) + 919*7c478bd9Sstevel@tonic-gate (cpu->cpu_id + 1) * VLP_SIZE; 920*7c478bd9Sstevel@tonic-gate #endif 921*7c478bd9Sstevel@tonic-gate } else { 922*7c478bd9Sstevel@tonic-gate newcr3 = MAKECR3(hat->hat_htable->ht_pfn); 923*7c478bd9Sstevel@tonic-gate } 924*7c478bd9Sstevel@tonic-gate setcr3(newcr3); 925*7c478bd9Sstevel@tonic-gate ASSERT(cpu == CPU); 926*7c478bd9Sstevel@tonic-gate } 927*7c478bd9Sstevel@tonic-gate 928*7c478bd9Sstevel@tonic-gate /* 929*7c478bd9Sstevel@tonic-gate * Utility to return a valid x86pte_t from protections, pfn, and level number 930*7c478bd9Sstevel@tonic-gate */ 931*7c478bd9Sstevel@tonic-gate static x86pte_t 932*7c478bd9Sstevel@tonic-gate hati_mkpte(pfn_t pfn, uint_t attr, level_t level, uint_t flags) 933*7c478bd9Sstevel@tonic-gate { 934*7c478bd9Sstevel@tonic-gate x86pte_t pte; 935*7c478bd9Sstevel@tonic-gate uint_t cache_attr = attr & HAT_ORDER_MASK; 936*7c478bd9Sstevel@tonic-gate 937*7c478bd9Sstevel@tonic-gate pte = MAKEPTE(pfn, level); 938*7c478bd9Sstevel@tonic-gate 939*7c478bd9Sstevel@tonic-gate if (attr & PROT_WRITE) 940*7c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITABLE); 941*7c478bd9Sstevel@tonic-gate 942*7c478bd9Sstevel@tonic-gate if (attr & PROT_USER) 943*7c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_USER); 944*7c478bd9Sstevel@tonic-gate 945*7c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC)) 946*7c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_nx); 947*7c478bd9Sstevel@tonic-gate 948*7c478bd9Sstevel@tonic-gate /* 949*7c478bd9Sstevel@tonic-gate * set the software bits used track ref/mod sync's and hments 950*7c478bd9Sstevel@tonic-gate */ 951*7c478bd9Sstevel@tonic-gate if (attr & HAT_NOSYNC) 952*7c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOSYNC); 953*7c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_NOCONSIST) 954*7c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCONSIST | PT_NOSYNC); 955*7c478bd9Sstevel@tonic-gate 956*7c478bd9Sstevel@tonic-gate /* 957*7c478bd9Sstevel@tonic-gate * Set the caching attributes in the PTE. The combination 958*7c478bd9Sstevel@tonic-gate * of attributes are poorly defined, so we pay attention 959*7c478bd9Sstevel@tonic-gate * to them in the given order. 960*7c478bd9Sstevel@tonic-gate * 961*7c478bd9Sstevel@tonic-gate * The test for HAT_STRICTORDER is different because it's defined 962*7c478bd9Sstevel@tonic-gate * as "0" - which was a stupid thing to do, but is too late to change! 963*7c478bd9Sstevel@tonic-gate */ 964*7c478bd9Sstevel@tonic-gate if (cache_attr == HAT_STRICTORDER) { 965*7c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 966*7c478bd9Sstevel@tonic-gate /*LINTED [Lint hates empty ifs, but it's the obvious way to do this] */ 967*7c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_UNORDERED_OK | HAT_STORECACHING_OK)) { 968*7c478bd9Sstevel@tonic-gate /* nothing to set */; 969*7c478bd9Sstevel@tonic-gate } else if (cache_attr & (HAT_MERGING_OK | HAT_LOADCACHING_OK)) { 970*7c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_NOCACHE); 971*7c478bd9Sstevel@tonic-gate if (x86_feature & X86_PAT) 972*7c478bd9Sstevel@tonic-gate PTE_SET(pte, (level == 0) ? PT_PAT_4K : PT_PAT_LARGE); 973*7c478bd9Sstevel@tonic-gate else 974*7c478bd9Sstevel@tonic-gate PTE_SET(pte, PT_WRITETHRU); 975*7c478bd9Sstevel@tonic-gate } else { 976*7c478bd9Sstevel@tonic-gate panic("hati_mkpte(): bad caching attributes: %x\n", cache_attr); 977*7c478bd9Sstevel@tonic-gate } 978*7c478bd9Sstevel@tonic-gate 979*7c478bd9Sstevel@tonic-gate return (pte); 980*7c478bd9Sstevel@tonic-gate } 981*7c478bd9Sstevel@tonic-gate 982*7c478bd9Sstevel@tonic-gate /* 983*7c478bd9Sstevel@tonic-gate * Duplicate address translations of the parent to the child. 984*7c478bd9Sstevel@tonic-gate * This function really isn't used anymore. 985*7c478bd9Sstevel@tonic-gate */ 986*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 987*7c478bd9Sstevel@tonic-gate int 988*7c478bd9Sstevel@tonic-gate hat_dup(hat_t *old, hat_t *new, caddr_t addr, size_t len, uint_t flag) 989*7c478bd9Sstevel@tonic-gate { 990*7c478bd9Sstevel@tonic-gate ASSERT((uintptr_t)addr < kernelbase); 991*7c478bd9Sstevel@tonic-gate ASSERT(new != kas.a_hat); 992*7c478bd9Sstevel@tonic-gate ASSERT(old != kas.a_hat); 993*7c478bd9Sstevel@tonic-gate return (0); 994*7c478bd9Sstevel@tonic-gate } 995*7c478bd9Sstevel@tonic-gate 996*7c478bd9Sstevel@tonic-gate /* 997*7c478bd9Sstevel@tonic-gate * Allocate any hat resources required for a process being swapped in. 998*7c478bd9Sstevel@tonic-gate */ 999*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 1000*7c478bd9Sstevel@tonic-gate void 1001*7c478bd9Sstevel@tonic-gate hat_swapin(hat_t *hat) 1002*7c478bd9Sstevel@tonic-gate { 1003*7c478bd9Sstevel@tonic-gate /* do nothing - we let everything fault back in */ 1004*7c478bd9Sstevel@tonic-gate } 1005*7c478bd9Sstevel@tonic-gate 1006*7c478bd9Sstevel@tonic-gate /* 1007*7c478bd9Sstevel@tonic-gate * Unload all translations associated with an address space of a process 1008*7c478bd9Sstevel@tonic-gate * that is being swapped out. 1009*7c478bd9Sstevel@tonic-gate */ 1010*7c478bd9Sstevel@tonic-gate void 1011*7c478bd9Sstevel@tonic-gate hat_swapout(hat_t *hat) 1012*7c478bd9Sstevel@tonic-gate { 1013*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)0; 1014*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = _userlimit; 1015*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 1016*7c478bd9Sstevel@tonic-gate level_t l; 1017*7c478bd9Sstevel@tonic-gate 1018*7c478bd9Sstevel@tonic-gate /* 1019*7c478bd9Sstevel@tonic-gate * We can't just call hat_unload(hat, 0, _userlimit...) here, because 1020*7c478bd9Sstevel@tonic-gate * seg_spt and shared pagetables can't be swapped out. 1021*7c478bd9Sstevel@tonic-gate * Take a look at segspt_shmswapout() - it's a big no-op. 1022*7c478bd9Sstevel@tonic-gate * 1023*7c478bd9Sstevel@tonic-gate * Instead we'll walk through all the address space and unload 1024*7c478bd9Sstevel@tonic-gate * any mappings which we are sure are not shared, not locked. 1025*7c478bd9Sstevel@tonic-gate */ 1026*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 1027*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 1028*7c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 1029*7c478bd9Sstevel@tonic-gate if ((uintptr_t)hat->hat_as->a_userlimit < eaddr) 1030*7c478bd9Sstevel@tonic-gate eaddr = (uintptr_t)hat->hat_as->a_userlimit; 1031*7c478bd9Sstevel@tonic-gate 1032*7c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 1033*7c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 1034*7c478bd9Sstevel@tonic-gate if (ht == NULL) 1035*7c478bd9Sstevel@tonic-gate break; 1036*7c478bd9Sstevel@tonic-gate 1037*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 1038*7c478bd9Sstevel@tonic-gate 1039*7c478bd9Sstevel@tonic-gate /* 1040*7c478bd9Sstevel@tonic-gate * If the page table is shared skip its entire range. 1041*7c478bd9Sstevel@tonic-gate * This code knows that only level 0 page tables are shared 1042*7c478bd9Sstevel@tonic-gate */ 1043*7c478bd9Sstevel@tonic-gate l = ht->ht_level; 1044*7c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 1045*7c478bd9Sstevel@tonic-gate ASSERT(l == 0); 1046*7c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 1047*7c478bd9Sstevel@tonic-gate htable_release(ht); 1048*7c478bd9Sstevel@tonic-gate ht = NULL; 1049*7c478bd9Sstevel@tonic-gate continue; 1050*7c478bd9Sstevel@tonic-gate } 1051*7c478bd9Sstevel@tonic-gate 1052*7c478bd9Sstevel@tonic-gate /* 1053*7c478bd9Sstevel@tonic-gate * If the page table has no locked entries, unload this one. 1054*7c478bd9Sstevel@tonic-gate */ 1055*7c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt == 0) 1056*7c478bd9Sstevel@tonic-gate hat_unload(hat, (caddr_t)vaddr, LEVEL_SIZE(l), 1057*7c478bd9Sstevel@tonic-gate HAT_UNLOAD_UNMAP); 1058*7c478bd9Sstevel@tonic-gate 1059*7c478bd9Sstevel@tonic-gate /* 1060*7c478bd9Sstevel@tonic-gate * If we have a level 0 page table with locked entries, 1061*7c478bd9Sstevel@tonic-gate * skip the entire page table, otherwise skip just one entry. 1062*7c478bd9Sstevel@tonic-gate */ 1063*7c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt > 0 && l == 0) 1064*7c478bd9Sstevel@tonic-gate vaddr = ht->ht_vaddr + LEVEL_SIZE(1); 1065*7c478bd9Sstevel@tonic-gate else 1066*7c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 1067*7c478bd9Sstevel@tonic-gate } 1068*7c478bd9Sstevel@tonic-gate if (ht) 1069*7c478bd9Sstevel@tonic-gate htable_release(ht); 1070*7c478bd9Sstevel@tonic-gate 1071*7c478bd9Sstevel@tonic-gate /* 1072*7c478bd9Sstevel@tonic-gate * We're in swapout because the system is low on memory, so 1073*7c478bd9Sstevel@tonic-gate * go back and flush all the htables off the cached list. 1074*7c478bd9Sstevel@tonic-gate */ 1075*7c478bd9Sstevel@tonic-gate htable_purge_hat(hat); 1076*7c478bd9Sstevel@tonic-gate } 1077*7c478bd9Sstevel@tonic-gate 1078*7c478bd9Sstevel@tonic-gate /* 1079*7c478bd9Sstevel@tonic-gate * returns number of bytes that have valid mappings in hat. 1080*7c478bd9Sstevel@tonic-gate */ 1081*7c478bd9Sstevel@tonic-gate size_t 1082*7c478bd9Sstevel@tonic-gate hat_get_mapped_size(hat_t *hat) 1083*7c478bd9Sstevel@tonic-gate { 1084*7c478bd9Sstevel@tonic-gate size_t total = 0; 1085*7c478bd9Sstevel@tonic-gate int l; 1086*7c478bd9Sstevel@tonic-gate 1087*7c478bd9Sstevel@tonic-gate for (l = 0; l <= mmu.max_page_level; l++) 1088*7c478bd9Sstevel@tonic-gate total += (hat->hat_pages_mapped[l] << LEVEL_SHIFT(l)); 1089*7c478bd9Sstevel@tonic-gate 1090*7c478bd9Sstevel@tonic-gate return (total); 1091*7c478bd9Sstevel@tonic-gate } 1092*7c478bd9Sstevel@tonic-gate 1093*7c478bd9Sstevel@tonic-gate /* 1094*7c478bd9Sstevel@tonic-gate * enable/disable collection of stats for hat. 1095*7c478bd9Sstevel@tonic-gate */ 1096*7c478bd9Sstevel@tonic-gate int 1097*7c478bd9Sstevel@tonic-gate hat_stats_enable(hat_t *hat) 1098*7c478bd9Sstevel@tonic-gate { 1099*7c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, 1); 1100*7c478bd9Sstevel@tonic-gate return (1); 1101*7c478bd9Sstevel@tonic-gate } 1102*7c478bd9Sstevel@tonic-gate 1103*7c478bd9Sstevel@tonic-gate void 1104*7c478bd9Sstevel@tonic-gate hat_stats_disable(hat_t *hat) 1105*7c478bd9Sstevel@tonic-gate { 1106*7c478bd9Sstevel@tonic-gate atomic_add_32(&hat->hat_stats, -1); 1107*7c478bd9Sstevel@tonic-gate } 1108*7c478bd9Sstevel@tonic-gate 1109*7c478bd9Sstevel@tonic-gate /* 1110*7c478bd9Sstevel@tonic-gate * Utility to sync the ref/mod bits from a page table entry to the page_t 1111*7c478bd9Sstevel@tonic-gate * We must be holding the mapping list lock when this is called. 1112*7c478bd9Sstevel@tonic-gate */ 1113*7c478bd9Sstevel@tonic-gate static void 1114*7c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(page_t *pp, x86pte_t pte, level_t level) 1115*7c478bd9Sstevel@tonic-gate { 1116*7c478bd9Sstevel@tonic-gate uint_t rm = 0; 1117*7c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 1118*7c478bd9Sstevel@tonic-gate 1119*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_NOSYNC)) 1120*7c478bd9Sstevel@tonic-gate return; 1121*7c478bd9Sstevel@tonic-gate 1122*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_REF)) 1123*7c478bd9Sstevel@tonic-gate rm |= P_REF; 1124*7c478bd9Sstevel@tonic-gate 1125*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_MOD)) 1126*7c478bd9Sstevel@tonic-gate rm |= P_MOD; 1127*7c478bd9Sstevel@tonic-gate 1128*7c478bd9Sstevel@tonic-gate if (rm == 0) 1129*7c478bd9Sstevel@tonic-gate return; 1130*7c478bd9Sstevel@tonic-gate 1131*7c478bd9Sstevel@tonic-gate /* 1132*7c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 1133*7c478bd9Sstevel@tonic-gate */ 1134*7c478bd9Sstevel@tonic-gate ASSERT(x86_hm_held(pp)); 1135*7c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(level); 1136*7c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 1137*7c478bd9Sstevel@tonic-gate for (; pgcnt > 0; --pgcnt) { 1138*7c478bd9Sstevel@tonic-gate /* 1139*7c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 1140*7c478bd9Sstevel@tonic-gate * pszc below this mapping size 1141*7c478bd9Sstevel@tonic-gate * since this large mapping existed after we 1142*7c478bd9Sstevel@tonic-gate * took mlist lock. 1143*7c478bd9Sstevel@tonic-gate */ 1144*7c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= level); 1145*7c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 1146*7c478bd9Sstevel@tonic-gate ++pp; 1147*7c478bd9Sstevel@tonic-gate } 1148*7c478bd9Sstevel@tonic-gate } 1149*7c478bd9Sstevel@tonic-gate 1150*7c478bd9Sstevel@tonic-gate /* 1151*7c478bd9Sstevel@tonic-gate * This the set of PTE bits for PFN, permissions and caching 1152*7c478bd9Sstevel@tonic-gate * that require a TLB flush (hat_demap) if changed on a HAT_LOAD_REMAP 1153*7c478bd9Sstevel@tonic-gate */ 1154*7c478bd9Sstevel@tonic-gate #define PT_REMAP_BITS \ 1155*7c478bd9Sstevel@tonic-gate (PT_PADDR | PT_NX | PT_WRITABLE | PT_WRITETHRU | \ 1156*7c478bd9Sstevel@tonic-gate PT_NOCACHE | PT_PAT_4K | PT_PAT_LARGE) 1157*7c478bd9Sstevel@tonic-gate 1158*7c478bd9Sstevel@tonic-gate /* 1159*7c478bd9Sstevel@tonic-gate * Do the low-level work to get a mapping entered into a HAT's pagetables 1160*7c478bd9Sstevel@tonic-gate * and in the mapping list of the associated page_t. 1161*7c478bd9Sstevel@tonic-gate */ 1162*7c478bd9Sstevel@tonic-gate static void 1163*7c478bd9Sstevel@tonic-gate hati_pte_map( 1164*7c478bd9Sstevel@tonic-gate htable_t *ht, 1165*7c478bd9Sstevel@tonic-gate uint_t entry, 1166*7c478bd9Sstevel@tonic-gate page_t *pp, 1167*7c478bd9Sstevel@tonic-gate x86pte_t pte, 1168*7c478bd9Sstevel@tonic-gate int flags, 1169*7c478bd9Sstevel@tonic-gate void *pte_ptr) 1170*7c478bd9Sstevel@tonic-gate { 1171*7c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 1172*7c478bd9Sstevel@tonic-gate x86pte_t old_pte; 1173*7c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 1174*7c478bd9Sstevel@tonic-gate hment_t *hm; 1175*7c478bd9Sstevel@tonic-gate uint_t is_consist; 1176*7c478bd9Sstevel@tonic-gate 1177*7c478bd9Sstevel@tonic-gate /* 1178*7c478bd9Sstevel@tonic-gate * Is this a consistant (ie. need mapping list lock) mapping? 1179*7c478bd9Sstevel@tonic-gate */ 1180*7c478bd9Sstevel@tonic-gate is_consist = (pp != NULL && (flags & HAT_LOAD_NOCONSIST) == 0); 1181*7c478bd9Sstevel@tonic-gate 1182*7c478bd9Sstevel@tonic-gate /* 1183*7c478bd9Sstevel@tonic-gate * Track locked mapping count in the htable. Do this first, 1184*7c478bd9Sstevel@tonic-gate * as we track locking even if there already is a mapping present. 1185*7c478bd9Sstevel@tonic-gate */ 1186*7c478bd9Sstevel@tonic-gate if ((flags & HAT_LOAD_LOCK) != 0 && hat != kas.a_hat) 1187*7c478bd9Sstevel@tonic-gate HTABLE_LOCK_INC(ht); 1188*7c478bd9Sstevel@tonic-gate 1189*7c478bd9Sstevel@tonic-gate /* 1190*7c478bd9Sstevel@tonic-gate * Acquire the page's mapping list lock and get an hment to use. 1191*7c478bd9Sstevel@tonic-gate * Note that hment_prepare() might return NULL. 1192*7c478bd9Sstevel@tonic-gate */ 1193*7c478bd9Sstevel@tonic-gate if (is_consist) { 1194*7c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 1195*7c478bd9Sstevel@tonic-gate hm = hment_prepare(ht, entry, pp); 1196*7c478bd9Sstevel@tonic-gate } 1197*7c478bd9Sstevel@tonic-gate 1198*7c478bd9Sstevel@tonic-gate /* 1199*7c478bd9Sstevel@tonic-gate * Set the new pte, retrieving the old one at the same time. 1200*7c478bd9Sstevel@tonic-gate */ 1201*7c478bd9Sstevel@tonic-gate old_pte = x86pte_set(ht, entry, pte, pte_ptr); 1202*7c478bd9Sstevel@tonic-gate 1203*7c478bd9Sstevel@tonic-gate /* 1204*7c478bd9Sstevel@tonic-gate * If the mapping didn't change there is nothing more to do. 1205*7c478bd9Sstevel@tonic-gate */ 1206*7c478bd9Sstevel@tonic-gate if (PTE_EQUIV(pte, old_pte)) { 1207*7c478bd9Sstevel@tonic-gate if (is_consist) { 1208*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 1209*7c478bd9Sstevel@tonic-gate if (hm != NULL) 1210*7c478bd9Sstevel@tonic-gate hment_free(hm); 1211*7c478bd9Sstevel@tonic-gate } 1212*7c478bd9Sstevel@tonic-gate return; 1213*7c478bd9Sstevel@tonic-gate } 1214*7c478bd9Sstevel@tonic-gate 1215*7c478bd9Sstevel@tonic-gate /* 1216*7c478bd9Sstevel@tonic-gate * Install a new mapping in the page's mapping list 1217*7c478bd9Sstevel@tonic-gate */ 1218*7c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 1219*7c478bd9Sstevel@tonic-gate if (is_consist) { 1220*7c478bd9Sstevel@tonic-gate hment_assign(ht, entry, pp, hm); 1221*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 1222*7c478bd9Sstevel@tonic-gate } else { 1223*7c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 1224*7c478bd9Sstevel@tonic-gate } 1225*7c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 1226*7c478bd9Sstevel@tonic-gate PGCNT_INC(hat, l); 1227*7c478bd9Sstevel@tonic-gate return; 1228*7c478bd9Sstevel@tonic-gate } 1229*7c478bd9Sstevel@tonic-gate 1230*7c478bd9Sstevel@tonic-gate /* 1231*7c478bd9Sstevel@tonic-gate * Remap's are more complicated: 1232*7c478bd9Sstevel@tonic-gate * - HAT_LOAD_REMAP must be specified if changing the pfn. 1233*7c478bd9Sstevel@tonic-gate * We also require that NOCONSIST be specified. 1234*7c478bd9Sstevel@tonic-gate * - Otherwise only permission or caching bits may change. 1235*7c478bd9Sstevel@tonic-gate */ 1236*7c478bd9Sstevel@tonic-gate if (!PTE_ISPAGE(old_pte, l)) 1237*7c478bd9Sstevel@tonic-gate panic("non-null/page mapping pte=" FMT_PTE, old_pte); 1238*7c478bd9Sstevel@tonic-gate 1239*7c478bd9Sstevel@tonic-gate if (PTE2PFN(old_pte, l) != PTE2PFN(pte, l)) { 1240*7c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_REMAP); 1241*7c478bd9Sstevel@tonic-gate ASSERT(flags & HAT_LOAD_NOCONSIST); 1242*7c478bd9Sstevel@tonic-gate ASSERT(PTE_GET(old_pte, PT_NOCONSIST)); 1243*7c478bd9Sstevel@tonic-gate ASSERT(pf_is_memory(PTE2PFN(old_pte, l)) == 1244*7c478bd9Sstevel@tonic-gate pf_is_memory(PTE2PFN(pte, l))); 1245*7c478bd9Sstevel@tonic-gate ASSERT(!is_consist); 1246*7c478bd9Sstevel@tonic-gate } 1247*7c478bd9Sstevel@tonic-gate 1248*7c478bd9Sstevel@tonic-gate /* 1249*7c478bd9Sstevel@tonic-gate * We only let remaps change the bits for PFNs, permissions 1250*7c478bd9Sstevel@tonic-gate * or caching type. 1251*7c478bd9Sstevel@tonic-gate */ 1252*7c478bd9Sstevel@tonic-gate ASSERT(PTE_GET(old_pte, ~(PT_REMAP_BITS | PT_REF | PT_MOD)) == 1253*7c478bd9Sstevel@tonic-gate PTE_GET(pte, ~PT_REMAP_BITS)); 1254*7c478bd9Sstevel@tonic-gate 1255*7c478bd9Sstevel@tonic-gate /* 1256*7c478bd9Sstevel@tonic-gate * A remap requires invalidating the TLBs, since remapping the 1257*7c478bd9Sstevel@tonic-gate * same PFN requires NOCONSIST, we don't have to sync R/M bits. 1258*7c478bd9Sstevel@tonic-gate */ 1259*7c478bd9Sstevel@tonic-gate hat_demap(hat, htable_e2va(ht, entry)); 1260*7c478bd9Sstevel@tonic-gate 1261*7c478bd9Sstevel@tonic-gate /* 1262*7c478bd9Sstevel@tonic-gate * We don't create any mapping list entries on a remap, so release 1263*7c478bd9Sstevel@tonic-gate * any allocated hment after we drop the mapping list lock. 1264*7c478bd9Sstevel@tonic-gate */ 1265*7c478bd9Sstevel@tonic-gate if (is_consist) { 1266*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 1267*7c478bd9Sstevel@tonic-gate if (hm != NULL) 1268*7c478bd9Sstevel@tonic-gate hment_free(hm); 1269*7c478bd9Sstevel@tonic-gate } 1270*7c478bd9Sstevel@tonic-gate } 1271*7c478bd9Sstevel@tonic-gate 1272*7c478bd9Sstevel@tonic-gate /* 1273*7c478bd9Sstevel@tonic-gate * The t_hatdepth field is an 8-bit counter. We use the lower seven bits 1274*7c478bd9Sstevel@tonic-gate * to track exactly how deep we are in the memload->kmem_alloc recursion. 1275*7c478bd9Sstevel@tonic-gate * If the depth is greater than 1, that indicates that we are performing a 1276*7c478bd9Sstevel@tonic-gate * hat operation to satisfy another hat operation. To prevent infinite 1277*7c478bd9Sstevel@tonic-gate * recursion, we switch over to using pre-allocated "reserves" of htables 1278*7c478bd9Sstevel@tonic-gate * and hments. 1279*7c478bd9Sstevel@tonic-gate * 1280*7c478bd9Sstevel@tonic-gate * The uppermost bit is used to indicate that we are transitioning away 1281*7c478bd9Sstevel@tonic-gate * from being the reserves thread. See hati_reserves_exit() for the 1282*7c478bd9Sstevel@tonic-gate * details. 1283*7c478bd9Sstevel@tonic-gate */ 1284*7c478bd9Sstevel@tonic-gate #define EXITING_FLAG (1 << 7) 1285*7c478bd9Sstevel@tonic-gate #define DEPTH_MASK (~EXITING_FLAG) 1286*7c478bd9Sstevel@tonic-gate #define HAT_DEPTH(t) ((t)->t_hatdepth & DEPTH_MASK) 1287*7c478bd9Sstevel@tonic-gate #define EXITING_RESERVES(t) ((t)->t_hatdepth & EXITING_FLAG) 1288*7c478bd9Sstevel@tonic-gate 1289*7c478bd9Sstevel@tonic-gate /* 1290*7c478bd9Sstevel@tonic-gate * Access to reserves for HAT_NO_KALLOC is single threaded. 1291*7c478bd9Sstevel@tonic-gate * If someone else is in the reserves, we'll politely wait for them 1292*7c478bd9Sstevel@tonic-gate * to finish. This keeps normal hat_memload()s from eating up 1293*7c478bd9Sstevel@tonic-gate * the mappings needed to replenish the reserve. 1294*7c478bd9Sstevel@tonic-gate */ 1295*7c478bd9Sstevel@tonic-gate static void 1296*7c478bd9Sstevel@tonic-gate hati_reserves_enter(uint_t kmem_for_hat) 1297*7c478bd9Sstevel@tonic-gate { 1298*7c478bd9Sstevel@tonic-gate /* 1299*7c478bd9Sstevel@tonic-gate * 64 is an arbitrary number to catch serious problems. I'm not 1300*7c478bd9Sstevel@tonic-gate * sure what the absolute maximum depth is, but it should be 1301*7c478bd9Sstevel@tonic-gate * substantially less than this. 1302*7c478bd9Sstevel@tonic-gate */ 1303*7c478bd9Sstevel@tonic-gate ASSERT(HAT_DEPTH(curthread) < 64); 1304*7c478bd9Sstevel@tonic-gate 1305*7c478bd9Sstevel@tonic-gate /* 1306*7c478bd9Sstevel@tonic-gate * If we are doing a memload to satisfy a kmem operation, we enter 1307*7c478bd9Sstevel@tonic-gate * the reserves immediately; we don't wait to recurse to a second 1308*7c478bd9Sstevel@tonic-gate * level of memload. 1309*7c478bd9Sstevel@tonic-gate */ 1310*7c478bd9Sstevel@tonic-gate ASSERT(kmem_for_hat < 2); 1311*7c478bd9Sstevel@tonic-gate curthread->t_hatdepth += (1 + kmem_for_hat); 1312*7c478bd9Sstevel@tonic-gate 1313*7c478bd9Sstevel@tonic-gate if (hat_reserves_thread == curthread || use_boot_reserve) 1314*7c478bd9Sstevel@tonic-gate return; 1315*7c478bd9Sstevel@tonic-gate 1316*7c478bd9Sstevel@tonic-gate if (HAT_DEPTH(curthread) > 1 || hat_reserves_thread != NULL) { 1317*7c478bd9Sstevel@tonic-gate mutex_enter(&hat_reserves_lock); 1318*7c478bd9Sstevel@tonic-gate while (hat_reserves_thread != NULL) 1319*7c478bd9Sstevel@tonic-gate cv_wait(&hat_reserves_cv, &hat_reserves_lock); 1320*7c478bd9Sstevel@tonic-gate 1321*7c478bd9Sstevel@tonic-gate if (HAT_DEPTH(curthread) > 1) 1322*7c478bd9Sstevel@tonic-gate hat_reserves_thread = curthread; 1323*7c478bd9Sstevel@tonic-gate 1324*7c478bd9Sstevel@tonic-gate mutex_exit(&hat_reserves_lock); 1325*7c478bd9Sstevel@tonic-gate } 1326*7c478bd9Sstevel@tonic-gate } 1327*7c478bd9Sstevel@tonic-gate 1328*7c478bd9Sstevel@tonic-gate /* 1329*7c478bd9Sstevel@tonic-gate * If we are the reserves_thread and we've finally finished with all our 1330*7c478bd9Sstevel@tonic-gate * memloads (ie. no longer doing hat slabs), we can release our use of the 1331*7c478bd9Sstevel@tonic-gate * reserve. 1332*7c478bd9Sstevel@tonic-gate */ 1333*7c478bd9Sstevel@tonic-gate static void 1334*7c478bd9Sstevel@tonic-gate hati_reserves_exit(uint_t kmem_for_hat) 1335*7c478bd9Sstevel@tonic-gate { 1336*7c478bd9Sstevel@tonic-gate ASSERT(kmem_for_hat < 2); 1337*7c478bd9Sstevel@tonic-gate curthread->t_hatdepth -= (1 + kmem_for_hat); 1338*7c478bd9Sstevel@tonic-gate 1339*7c478bd9Sstevel@tonic-gate /* 1340*7c478bd9Sstevel@tonic-gate * Simple case: either we are not the reserves thread, or we are 1341*7c478bd9Sstevel@tonic-gate * the reserves thread and we are nested deeply enough that we 1342*7c478bd9Sstevel@tonic-gate * should still be the reserves thread. 1343*7c478bd9Sstevel@tonic-gate * 1344*7c478bd9Sstevel@tonic-gate * Note: we may not become the reserves thread after we recursively 1345*7c478bd9Sstevel@tonic-gate * enter our second HAT routine, but we don't stop being the 1346*7c478bd9Sstevel@tonic-gate * reserves thread until we exit the toplevel HAT routine. This is 1347*7c478bd9Sstevel@tonic-gate * to work around vmem's inability to determine when an allocation 1348*7c478bd9Sstevel@tonic-gate * should be satisfied from the hat_memload arena, which can lead 1349*7c478bd9Sstevel@tonic-gate * to an infinite loop of memload->vmem_populate->memload->. 1350*7c478bd9Sstevel@tonic-gate */ 1351*7c478bd9Sstevel@tonic-gate if (curthread != hat_reserves_thread || HAT_DEPTH(curthread) > 0 || 1352*7c478bd9Sstevel@tonic-gate use_boot_reserve) 1353*7c478bd9Sstevel@tonic-gate return; 1354*7c478bd9Sstevel@tonic-gate 1355*7c478bd9Sstevel@tonic-gate mutex_enter(&hat_reserves_lock); 1356*7c478bd9Sstevel@tonic-gate ASSERT(hat_reserves_thread == curthread); 1357*7c478bd9Sstevel@tonic-gate hat_reserves_thread = NULL; 1358*7c478bd9Sstevel@tonic-gate cv_broadcast(&hat_reserves_cv); 1359*7c478bd9Sstevel@tonic-gate mutex_exit(&hat_reserves_lock); 1360*7c478bd9Sstevel@tonic-gate 1361*7c478bd9Sstevel@tonic-gate /* 1362*7c478bd9Sstevel@tonic-gate * As we leave the reserves, we want to be sure the reserve lists 1363*7c478bd9Sstevel@tonic-gate * aren't overstocked. Freeing excess reserves requires that we 1364*7c478bd9Sstevel@tonic-gate * call kmem_free(), which may require additional allocations, 1365*7c478bd9Sstevel@tonic-gate * causing us to re-enter the reserves. To avoid infinite 1366*7c478bd9Sstevel@tonic-gate * recursion, we only try to adjust reserves at the very top level. 1367*7c478bd9Sstevel@tonic-gate */ 1368*7c478bd9Sstevel@tonic-gate if (!kmem_for_hat && !EXITING_RESERVES(curthread)) { 1369*7c478bd9Sstevel@tonic-gate curthread->t_hatdepth |= EXITING_FLAG; 1370*7c478bd9Sstevel@tonic-gate htable_adjust_reserve(); 1371*7c478bd9Sstevel@tonic-gate hment_adjust_reserve(); 1372*7c478bd9Sstevel@tonic-gate curthread->t_hatdepth &= (~EXITING_FLAG); 1373*7c478bd9Sstevel@tonic-gate } 1374*7c478bd9Sstevel@tonic-gate 1375*7c478bd9Sstevel@tonic-gate /* 1376*7c478bd9Sstevel@tonic-gate * just in case something went wrong in doing adjust reserves 1377*7c478bd9Sstevel@tonic-gate */ 1378*7c478bd9Sstevel@tonic-gate ASSERT(hat_reserves_thread != curthread); 1379*7c478bd9Sstevel@tonic-gate } 1380*7c478bd9Sstevel@tonic-gate 1381*7c478bd9Sstevel@tonic-gate /* 1382*7c478bd9Sstevel@tonic-gate * Internal routine to load a single page table entry. 1383*7c478bd9Sstevel@tonic-gate */ 1384*7c478bd9Sstevel@tonic-gate static void 1385*7c478bd9Sstevel@tonic-gate hati_load_common( 1386*7c478bd9Sstevel@tonic-gate hat_t *hat, 1387*7c478bd9Sstevel@tonic-gate uintptr_t va, 1388*7c478bd9Sstevel@tonic-gate page_t *pp, 1389*7c478bd9Sstevel@tonic-gate uint_t attr, 1390*7c478bd9Sstevel@tonic-gate uint_t flags, 1391*7c478bd9Sstevel@tonic-gate level_t level, 1392*7c478bd9Sstevel@tonic-gate pfn_t pfn) 1393*7c478bd9Sstevel@tonic-gate { 1394*7c478bd9Sstevel@tonic-gate htable_t *ht; 1395*7c478bd9Sstevel@tonic-gate uint_t entry; 1396*7c478bd9Sstevel@tonic-gate x86pte_t pte; 1397*7c478bd9Sstevel@tonic-gate uint_t kmem_for_hat = (flags & HAT_NO_KALLOC) ? 1 : 0; 1398*7c478bd9Sstevel@tonic-gate 1399*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 1400*7c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 1401*7c478bd9Sstevel@tonic-gate 1402*7c478bd9Sstevel@tonic-gate if (flags & HAT_LOAD_SHARE) 1403*7c478bd9Sstevel@tonic-gate hat->hat_flags |= HAT_SHARED; 1404*7c478bd9Sstevel@tonic-gate 1405*7c478bd9Sstevel@tonic-gate /* 1406*7c478bd9Sstevel@tonic-gate * Find the page table that maps this page if it already exists. 1407*7c478bd9Sstevel@tonic-gate */ 1408*7c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, va, level); 1409*7c478bd9Sstevel@tonic-gate 1410*7c478bd9Sstevel@tonic-gate /* 1411*7c478bd9Sstevel@tonic-gate * All threads go through hati_reserves_enter() to at least wait 1412*7c478bd9Sstevel@tonic-gate * for any existing reserves user to finish. This helps reduce 1413*7c478bd9Sstevel@tonic-gate * pressure on the reserves. In addition, if this thread needs 1414*7c478bd9Sstevel@tonic-gate * to become the new reserve user it will. 1415*7c478bd9Sstevel@tonic-gate */ 1416*7c478bd9Sstevel@tonic-gate hati_reserves_enter(kmem_for_hat); 1417*7c478bd9Sstevel@tonic-gate 1418*7c478bd9Sstevel@tonic-gate ASSERT(HAT_DEPTH(curthread) == 1 || va >= kernelbase); 1419*7c478bd9Sstevel@tonic-gate 1420*7c478bd9Sstevel@tonic-gate /* 1421*7c478bd9Sstevel@tonic-gate * Kernel memloads for HAT data should never use hments! 1422*7c478bd9Sstevel@tonic-gate * If it did that would seriously complicate the reserves system, since 1423*7c478bd9Sstevel@tonic-gate * hment_alloc() would need to know about HAT_NO_KALLOC. 1424*7c478bd9Sstevel@tonic-gate * 1425*7c478bd9Sstevel@tonic-gate * We also must have HAT_LOAD_NOCONSIST if page_t is NULL. 1426*7c478bd9Sstevel@tonic-gate */ 1427*7c478bd9Sstevel@tonic-gate if (HAT_DEPTH(curthread) > 1 || pp == NULL) 1428*7c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 1429*7c478bd9Sstevel@tonic-gate 1430*7c478bd9Sstevel@tonic-gate if (ht == NULL) { 1431*7c478bd9Sstevel@tonic-gate ht = htable_create(hat, va, level, NULL); 1432*7c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 1433*7c478bd9Sstevel@tonic-gate } 1434*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1435*7c478bd9Sstevel@tonic-gate 1436*7c478bd9Sstevel@tonic-gate /* 1437*7c478bd9Sstevel@tonic-gate * a bunch of paranoid error checking 1438*7c478bd9Sstevel@tonic-gate */ 1439*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 1440*7c478bd9Sstevel@tonic-gate if (ht->ht_vaddr > va || va > HTABLE_LAST_PAGE(ht)) 1441*7c478bd9Sstevel@tonic-gate panic("hati_load_common: bad htable %p, va %p", ht, (void *)va); 1442*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == level); 1443*7c478bd9Sstevel@tonic-gate 1444*7c478bd9Sstevel@tonic-gate /* 1445*7c478bd9Sstevel@tonic-gate * construct the new PTE 1446*7c478bd9Sstevel@tonic-gate */ 1447*7c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 1448*7c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 1449*7c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, level, flags); 1450*7c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && va >= kernelbase) 1451*7c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 1452*7c478bd9Sstevel@tonic-gate 1453*7c478bd9Sstevel@tonic-gate /* 1454*7c478bd9Sstevel@tonic-gate * establish the mapping 1455*7c478bd9Sstevel@tonic-gate */ 1456*7c478bd9Sstevel@tonic-gate hati_pte_map(ht, entry, pp, pte, flags, NULL); 1457*7c478bd9Sstevel@tonic-gate 1458*7c478bd9Sstevel@tonic-gate /* 1459*7c478bd9Sstevel@tonic-gate * release the htable and any reserves 1460*7c478bd9Sstevel@tonic-gate */ 1461*7c478bd9Sstevel@tonic-gate htable_release(ht); 1462*7c478bd9Sstevel@tonic-gate hati_reserves_exit(kmem_for_hat); 1463*7c478bd9Sstevel@tonic-gate } 1464*7c478bd9Sstevel@tonic-gate 1465*7c478bd9Sstevel@tonic-gate /* 1466*7c478bd9Sstevel@tonic-gate * special case of hat_memload to deal with some kernel addrs for performance 1467*7c478bd9Sstevel@tonic-gate */ 1468*7c478bd9Sstevel@tonic-gate static void 1469*7c478bd9Sstevel@tonic-gate hat_kmap_load( 1470*7c478bd9Sstevel@tonic-gate caddr_t addr, 1471*7c478bd9Sstevel@tonic-gate page_t *pp, 1472*7c478bd9Sstevel@tonic-gate uint_t attr, 1473*7c478bd9Sstevel@tonic-gate uint_t flags) 1474*7c478bd9Sstevel@tonic-gate { 1475*7c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 1476*7c478bd9Sstevel@tonic-gate x86pte_t pte; 1477*7c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 1478*7c478bd9Sstevel@tonic-gate pgcnt_t pg_off = mmu_btop(va - mmu.kmap_addr); 1479*7c478bd9Sstevel@tonic-gate htable_t *ht; 1480*7c478bd9Sstevel@tonic-gate uint_t entry; 1481*7c478bd9Sstevel@tonic-gate void *pte_ptr; 1482*7c478bd9Sstevel@tonic-gate 1483*7c478bd9Sstevel@tonic-gate /* 1484*7c478bd9Sstevel@tonic-gate * construct the requested PTE 1485*7c478bd9Sstevel@tonic-gate */ 1486*7c478bd9Sstevel@tonic-gate attr &= ~PROT_USER; 1487*7c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1488*7c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 1489*7c478bd9Sstevel@tonic-gate PTE_SET(pte, mmu.pt_global); 1490*7c478bd9Sstevel@tonic-gate 1491*7c478bd9Sstevel@tonic-gate /* 1492*7c478bd9Sstevel@tonic-gate * Figure out the pte_ptr and htable and use common code to finish up 1493*7c478bd9Sstevel@tonic-gate */ 1494*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 1495*7c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 1496*7c478bd9Sstevel@tonic-gate else 1497*7c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 1498*7c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) >> 1499*7c478bd9Sstevel@tonic-gate LEVEL_SHIFT(1)]; 1500*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 1501*7c478bd9Sstevel@tonic-gate hati_pte_map(ht, entry, pp, pte, flags, pte_ptr); 1502*7c478bd9Sstevel@tonic-gate } 1503*7c478bd9Sstevel@tonic-gate 1504*7c478bd9Sstevel@tonic-gate /* 1505*7c478bd9Sstevel@tonic-gate * hat_memload() - load a translation to the given page struct 1506*7c478bd9Sstevel@tonic-gate * 1507*7c478bd9Sstevel@tonic-gate * Flags for hat_memload/hat_devload/hat_*attr. 1508*7c478bd9Sstevel@tonic-gate * 1509*7c478bd9Sstevel@tonic-gate * HAT_LOAD Default flags to load a translation to the page. 1510*7c478bd9Sstevel@tonic-gate * 1511*7c478bd9Sstevel@tonic-gate * HAT_LOAD_LOCK Lock down mapping resources; hat_map(), hat_memload(), 1512*7c478bd9Sstevel@tonic-gate * and hat_devload(). 1513*7c478bd9Sstevel@tonic-gate * 1514*7c478bd9Sstevel@tonic-gate * HAT_LOAD_NOCONSIST Do not add mapping to page_t mapping list. 1515*7c478bd9Sstevel@tonic-gate * sets PT_NOCONSIST (soft bit) 1516*7c478bd9Sstevel@tonic-gate * 1517*7c478bd9Sstevel@tonic-gate * HAT_LOAD_SHARE A flag to hat_memload() to indicate h/w page tables 1518*7c478bd9Sstevel@tonic-gate * that map some user pages (not kas) is shared by more 1519*7c478bd9Sstevel@tonic-gate * than one process (eg. ISM). 1520*7c478bd9Sstevel@tonic-gate * 1521*7c478bd9Sstevel@tonic-gate * HAT_LOAD_REMAP Reload a valid pte with a different page frame. 1522*7c478bd9Sstevel@tonic-gate * 1523*7c478bd9Sstevel@tonic-gate * HAT_NO_KALLOC Do not kmem_alloc while creating the mapping; at this 1524*7c478bd9Sstevel@tonic-gate * point, it's setting up mapping to allocate internal 1525*7c478bd9Sstevel@tonic-gate * hat layer data structures. This flag forces hat layer 1526*7c478bd9Sstevel@tonic-gate * to tap its reserves in order to prevent infinite 1527*7c478bd9Sstevel@tonic-gate * recursion. 1528*7c478bd9Sstevel@tonic-gate * 1529*7c478bd9Sstevel@tonic-gate * The following is a protection attribute (like PROT_READ, etc.) 1530*7c478bd9Sstevel@tonic-gate * 1531*7c478bd9Sstevel@tonic-gate * HAT_NOSYNC set PT_NOSYNC (soft bit) - this mapping's ref/mod bits 1532*7c478bd9Sstevel@tonic-gate * are never cleared. 1533*7c478bd9Sstevel@tonic-gate * 1534*7c478bd9Sstevel@tonic-gate * Installing new valid PTE's and creation of the mapping list 1535*7c478bd9Sstevel@tonic-gate * entry are controlled under the same lock. It's derived from the 1536*7c478bd9Sstevel@tonic-gate * page_t being mapped. 1537*7c478bd9Sstevel@tonic-gate */ 1538*7c478bd9Sstevel@tonic-gate static uint_t supported_memload_flags = 1539*7c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_LOCK | HAT_LOAD_ADV | HAT_LOAD_NOCONSIST | 1540*7c478bd9Sstevel@tonic-gate HAT_LOAD_SHARE | HAT_NO_KALLOC | HAT_LOAD_REMAP | HAT_LOAD_TEXT; 1541*7c478bd9Sstevel@tonic-gate 1542*7c478bd9Sstevel@tonic-gate void 1543*7c478bd9Sstevel@tonic-gate hat_memload( 1544*7c478bd9Sstevel@tonic-gate hat_t *hat, 1545*7c478bd9Sstevel@tonic-gate caddr_t addr, 1546*7c478bd9Sstevel@tonic-gate page_t *pp, 1547*7c478bd9Sstevel@tonic-gate uint_t attr, 1548*7c478bd9Sstevel@tonic-gate uint_t flags) 1549*7c478bd9Sstevel@tonic-gate { 1550*7c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 1551*7c478bd9Sstevel@tonic-gate level_t level = 0; 1552*7c478bd9Sstevel@tonic-gate pfn_t pfn = page_pptonum(pp); 1553*7c478bd9Sstevel@tonic-gate 1554*7c478bd9Sstevel@tonic-gate HATIN(hat_memload, hat, addr, (size_t)MMU_PAGESIZE); 1555*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1556*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || va <= kernelbase); 1557*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 1558*7c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 1559*7c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 1560*7c478bd9Sstevel@tonic-gate 1561*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1562*7c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 1563*7c478bd9Sstevel@tonic-gate 1564*7c478bd9Sstevel@tonic-gate /* 1565*7c478bd9Sstevel@tonic-gate * kernel address special case for performance. 1566*7c478bd9Sstevel@tonic-gate */ 1567*7c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 1568*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 1569*7c478bd9Sstevel@tonic-gate hat_kmap_load(addr, pp, attr, flags); 1570*7c478bd9Sstevel@tonic-gate return; 1571*7c478bd9Sstevel@tonic-gate } 1572*7c478bd9Sstevel@tonic-gate 1573*7c478bd9Sstevel@tonic-gate /* 1574*7c478bd9Sstevel@tonic-gate * This is used for memory with normal caching enabled, so 1575*7c478bd9Sstevel@tonic-gate * always set HAT_STORECACHING_OK. 1576*7c478bd9Sstevel@tonic-gate */ 1577*7c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1578*7c478bd9Sstevel@tonic-gate hati_load_common(hat, va, pp, attr, flags, level, pfn); 1579*7c478bd9Sstevel@tonic-gate HATOUT(hat_memload, hat, addr); 1580*7c478bd9Sstevel@tonic-gate } 1581*7c478bd9Sstevel@tonic-gate 1582*7c478bd9Sstevel@tonic-gate /* 1583*7c478bd9Sstevel@tonic-gate * Load the given array of page structs using large pages when possible 1584*7c478bd9Sstevel@tonic-gate */ 1585*7c478bd9Sstevel@tonic-gate void 1586*7c478bd9Sstevel@tonic-gate hat_memload_array( 1587*7c478bd9Sstevel@tonic-gate hat_t *hat, 1588*7c478bd9Sstevel@tonic-gate caddr_t addr, 1589*7c478bd9Sstevel@tonic-gate size_t len, 1590*7c478bd9Sstevel@tonic-gate page_t **pages, 1591*7c478bd9Sstevel@tonic-gate uint_t attr, 1592*7c478bd9Sstevel@tonic-gate uint_t flags) 1593*7c478bd9Sstevel@tonic-gate { 1594*7c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 1595*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = va + len; 1596*7c478bd9Sstevel@tonic-gate level_t level; 1597*7c478bd9Sstevel@tonic-gate size_t pgsize; 1598*7c478bd9Sstevel@tonic-gate pgcnt_t pgindx = 0; 1599*7c478bd9Sstevel@tonic-gate pfn_t pfn; 1600*7c478bd9Sstevel@tonic-gate pgcnt_t i; 1601*7c478bd9Sstevel@tonic-gate 1602*7c478bd9Sstevel@tonic-gate HATIN(hat_memload_array, hat, addr, len); 1603*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1604*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || va + len <= kernelbase); 1605*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 1606*7c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 1607*7c478bd9Sstevel@tonic-gate ASSERT((flags & supported_memload_flags) == flags); 1608*7c478bd9Sstevel@tonic-gate 1609*7c478bd9Sstevel@tonic-gate /* 1610*7c478bd9Sstevel@tonic-gate * memload is used for memory with full caching enabled, so 1611*7c478bd9Sstevel@tonic-gate * set HAT_STORECACHING_OK. 1612*7c478bd9Sstevel@tonic-gate */ 1613*7c478bd9Sstevel@tonic-gate attr |= HAT_STORECACHING_OK; 1614*7c478bd9Sstevel@tonic-gate 1615*7c478bd9Sstevel@tonic-gate /* 1616*7c478bd9Sstevel@tonic-gate * handle all pages using largest possible pagesize 1617*7c478bd9Sstevel@tonic-gate */ 1618*7c478bd9Sstevel@tonic-gate while (va < eaddr) { 1619*7c478bd9Sstevel@tonic-gate /* 1620*7c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 1621*7c478bd9Sstevel@tonic-gate */ 1622*7c478bd9Sstevel@tonic-gate pfn = page_pptonum(pages[pgindx]); 1623*7c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 1624*7c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 1625*7c478bd9Sstevel@tonic-gate if (level == 0) 1626*7c478bd9Sstevel@tonic-gate break; 1627*7c478bd9Sstevel@tonic-gate if (!IS_P2ALIGNED(va, pgsize) || 1628*7c478bd9Sstevel@tonic-gate (eaddr - va) < pgsize || 1629*7c478bd9Sstevel@tonic-gate !IS_P2ALIGNED(pfn << MMU_PAGESHIFT, pgsize)) 1630*7c478bd9Sstevel@tonic-gate continue; 1631*7c478bd9Sstevel@tonic-gate 1632*7c478bd9Sstevel@tonic-gate /* 1633*7c478bd9Sstevel@tonic-gate * To use a large mapping of this size, all the 1634*7c478bd9Sstevel@tonic-gate * pages we are passed must be sequential subpages 1635*7c478bd9Sstevel@tonic-gate * of the large page. 1636*7c478bd9Sstevel@tonic-gate * hat_page_demote() can't change p_szc because 1637*7c478bd9Sstevel@tonic-gate * all pages are locked. 1638*7c478bd9Sstevel@tonic-gate */ 1639*7c478bd9Sstevel@tonic-gate if (pages[pgindx]->p_szc >= level) { 1640*7c478bd9Sstevel@tonic-gate for (i = 0; i < mmu_btop(pgsize); ++i) { 1641*7c478bd9Sstevel@tonic-gate if (pfn + i != 1642*7c478bd9Sstevel@tonic-gate page_pptonum(pages[pgindx + i])) 1643*7c478bd9Sstevel@tonic-gate break; 1644*7c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx + i]->p_szc >= 1645*7c478bd9Sstevel@tonic-gate level); 1646*7c478bd9Sstevel@tonic-gate ASSERT(pages[pgindx] + i == 1647*7c478bd9Sstevel@tonic-gate pages[pgindx + i]); 1648*7c478bd9Sstevel@tonic-gate } 1649*7c478bd9Sstevel@tonic-gate if (i == mmu_btop(pgsize)) 1650*7c478bd9Sstevel@tonic-gate break; 1651*7c478bd9Sstevel@tonic-gate } 1652*7c478bd9Sstevel@tonic-gate } 1653*7c478bd9Sstevel@tonic-gate 1654*7c478bd9Sstevel@tonic-gate /* 1655*7c478bd9Sstevel@tonic-gate * Shared page tables for DISM might have a pre-existing 1656*7c478bd9Sstevel@tonic-gate * level 0 page table that wasn't unlinked from all the 1657*7c478bd9Sstevel@tonic-gate * sharing hats. If we hit this for a large page, back off 1658*7c478bd9Sstevel@tonic-gate * to using level 0 pages. 1659*7c478bd9Sstevel@tonic-gate * 1660*7c478bd9Sstevel@tonic-gate * This can't be made better (ie. use large pages) until we 1661*7c478bd9Sstevel@tonic-gate * track all the htable's sharing and rewrite hat_pageunload(). 1662*7c478bd9Sstevel@tonic-gate * Note that would cost a pointer in htable_t for a rare case. 1663*7c478bd9Sstevel@tonic-gate * 1664*7c478bd9Sstevel@tonic-gate * Since the 32 bit kernel caches empty page tables, check 1665*7c478bd9Sstevel@tonic-gate * the kernel too. 1666*7c478bd9Sstevel@tonic-gate */ 1667*7c478bd9Sstevel@tonic-gate if ((hat == kas.a_hat || (hat->hat_flags & HAT_SHARED)) && 1668*7c478bd9Sstevel@tonic-gate level > 0) { 1669*7c478bd9Sstevel@tonic-gate htable_t *lower; 1670*7c478bd9Sstevel@tonic-gate 1671*7c478bd9Sstevel@tonic-gate lower = htable_getpte(hat, va, NULL, NULL, level - 1); 1672*7c478bd9Sstevel@tonic-gate if (lower != NULL) { 1673*7c478bd9Sstevel@tonic-gate level = 0; 1674*7c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(0); 1675*7c478bd9Sstevel@tonic-gate htable_release(lower); 1676*7c478bd9Sstevel@tonic-gate } 1677*7c478bd9Sstevel@tonic-gate } 1678*7c478bd9Sstevel@tonic-gate 1679*7c478bd9Sstevel@tonic-gate /* 1680*7c478bd9Sstevel@tonic-gate * load this page mapping 1681*7c478bd9Sstevel@tonic-gate */ 1682*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1683*7c478bd9Sstevel@tonic-gate hati_load_common(hat, va, pages[pgindx], attr, flags, 1684*7c478bd9Sstevel@tonic-gate level, pfn); 1685*7c478bd9Sstevel@tonic-gate 1686*7c478bd9Sstevel@tonic-gate /* 1687*7c478bd9Sstevel@tonic-gate * move to next page 1688*7c478bd9Sstevel@tonic-gate */ 1689*7c478bd9Sstevel@tonic-gate va += pgsize; 1690*7c478bd9Sstevel@tonic-gate pgindx += mmu_btop(pgsize); 1691*7c478bd9Sstevel@tonic-gate } 1692*7c478bd9Sstevel@tonic-gate HATOUT(hat_memload_array, hat, addr); 1693*7c478bd9Sstevel@tonic-gate } 1694*7c478bd9Sstevel@tonic-gate 1695*7c478bd9Sstevel@tonic-gate /* 1696*7c478bd9Sstevel@tonic-gate * void hat_devload(hat, addr, len, pf, attr, flags) 1697*7c478bd9Sstevel@tonic-gate * load/lock the given page frame number 1698*7c478bd9Sstevel@tonic-gate * 1699*7c478bd9Sstevel@tonic-gate * Advisory ordering attributes. Apply only to device mappings. 1700*7c478bd9Sstevel@tonic-gate * 1701*7c478bd9Sstevel@tonic-gate * HAT_STRICTORDER: the CPU must issue the references in order, as the 1702*7c478bd9Sstevel@tonic-gate * programmer specified. This is the default. 1703*7c478bd9Sstevel@tonic-gate * HAT_UNORDERED_OK: the CPU may reorder the references (this is all kinds 1704*7c478bd9Sstevel@tonic-gate * of reordering; store or load with store or load). 1705*7c478bd9Sstevel@tonic-gate * HAT_MERGING_OK: merging and batching: the CPU may merge individual stores 1706*7c478bd9Sstevel@tonic-gate * to consecutive locations (for example, turn two consecutive byte 1707*7c478bd9Sstevel@tonic-gate * stores into one halfword store), and it may batch individual loads 1708*7c478bd9Sstevel@tonic-gate * (for example, turn two consecutive byte loads into one halfword load). 1709*7c478bd9Sstevel@tonic-gate * This also implies re-ordering. 1710*7c478bd9Sstevel@tonic-gate * HAT_LOADCACHING_OK: the CPU may cache the data it fetches and reuse it 1711*7c478bd9Sstevel@tonic-gate * until another store occurs. The default is to fetch new data 1712*7c478bd9Sstevel@tonic-gate * on every load. This also implies merging. 1713*7c478bd9Sstevel@tonic-gate * HAT_STORECACHING_OK: the CPU may keep the data in the cache and push it to 1714*7c478bd9Sstevel@tonic-gate * the device (perhaps with other data) at a later time. The default is 1715*7c478bd9Sstevel@tonic-gate * to push the data right away. This also implies load caching. 1716*7c478bd9Sstevel@tonic-gate * 1717*7c478bd9Sstevel@tonic-gate * Equivalent of hat_memload(), but can be used for device memory where 1718*7c478bd9Sstevel@tonic-gate * there are no page_t's and we support additional flags (write merging, etc). 1719*7c478bd9Sstevel@tonic-gate * Note that we can have large page mappings with this interface. 1720*7c478bd9Sstevel@tonic-gate */ 1721*7c478bd9Sstevel@tonic-gate int supported_devload_flags = HAT_LOAD | HAT_LOAD_LOCK | 1722*7c478bd9Sstevel@tonic-gate HAT_LOAD_NOCONSIST | HAT_STRICTORDER | HAT_UNORDERED_OK | 1723*7c478bd9Sstevel@tonic-gate HAT_MERGING_OK | HAT_LOADCACHING_OK | HAT_STORECACHING_OK; 1724*7c478bd9Sstevel@tonic-gate 1725*7c478bd9Sstevel@tonic-gate void 1726*7c478bd9Sstevel@tonic-gate hat_devload( 1727*7c478bd9Sstevel@tonic-gate hat_t *hat, 1728*7c478bd9Sstevel@tonic-gate caddr_t addr, 1729*7c478bd9Sstevel@tonic-gate size_t len, 1730*7c478bd9Sstevel@tonic-gate pfn_t pfn, 1731*7c478bd9Sstevel@tonic-gate uint_t attr, 1732*7c478bd9Sstevel@tonic-gate int flags) 1733*7c478bd9Sstevel@tonic-gate { 1734*7c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(addr); 1735*7c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 1736*7c478bd9Sstevel@tonic-gate level_t level; 1737*7c478bd9Sstevel@tonic-gate size_t pgsize; 1738*7c478bd9Sstevel@tonic-gate page_t *pp; 1739*7c478bd9Sstevel@tonic-gate int f; /* per PTE copy of flags - maybe modified */ 1740*7c478bd9Sstevel@tonic-gate uint_t a; /* per PTE copy of attr */ 1741*7c478bd9Sstevel@tonic-gate 1742*7c478bd9Sstevel@tonic-gate HATIN(hat_devload, hat, addr, len); 1743*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 1744*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || eva <= kernelbase); 1745*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 1746*7c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 1747*7c478bd9Sstevel@tonic-gate ASSERT((flags & supported_devload_flags) == flags); 1748*7c478bd9Sstevel@tonic-gate 1749*7c478bd9Sstevel@tonic-gate /* 1750*7c478bd9Sstevel@tonic-gate * handle all pages 1751*7c478bd9Sstevel@tonic-gate */ 1752*7c478bd9Sstevel@tonic-gate while (va < eva) { 1753*7c478bd9Sstevel@tonic-gate 1754*7c478bd9Sstevel@tonic-gate /* 1755*7c478bd9Sstevel@tonic-gate * decide what level mapping to use (ie. pagesize) 1756*7c478bd9Sstevel@tonic-gate */ 1757*7c478bd9Sstevel@tonic-gate for (level = mmu.max_page_level; ; --level) { 1758*7c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(level); 1759*7c478bd9Sstevel@tonic-gate if (level == 0) 1760*7c478bd9Sstevel@tonic-gate break; 1761*7c478bd9Sstevel@tonic-gate if (IS_P2ALIGNED(va, pgsize) && 1762*7c478bd9Sstevel@tonic-gate (eva - va) >= pgsize && 1763*7c478bd9Sstevel@tonic-gate IS_P2ALIGNED(pfn, mmu_btop(pgsize))) 1764*7c478bd9Sstevel@tonic-gate break; 1765*7c478bd9Sstevel@tonic-gate } 1766*7c478bd9Sstevel@tonic-gate 1767*7c478bd9Sstevel@tonic-gate /* 1768*7c478bd9Sstevel@tonic-gate * Some kernel addresses have permanently existing page tables, 1769*7c478bd9Sstevel@tonic-gate * so be sure to use a compatible pagesize. 1770*7c478bd9Sstevel@tonic-gate */ 1771*7c478bd9Sstevel@tonic-gate if (hat == kas.a_hat && level > 0) { 1772*7c478bd9Sstevel@tonic-gate htable_t *lower; 1773*7c478bd9Sstevel@tonic-gate 1774*7c478bd9Sstevel@tonic-gate lower = htable_getpte(hat, va, NULL, NULL, level - 1); 1775*7c478bd9Sstevel@tonic-gate if (lower != NULL) { 1776*7c478bd9Sstevel@tonic-gate level = 0; 1777*7c478bd9Sstevel@tonic-gate pgsize = LEVEL_SIZE(0); 1778*7c478bd9Sstevel@tonic-gate htable_release(lower); 1779*7c478bd9Sstevel@tonic-gate } 1780*7c478bd9Sstevel@tonic-gate } 1781*7c478bd9Sstevel@tonic-gate 1782*7c478bd9Sstevel@tonic-gate /* 1783*7c478bd9Sstevel@tonic-gate * If it is memory get page_t and allow caching (this happens 1784*7c478bd9Sstevel@tonic-gate * for the nucleus pages) - though HAT_PLAT_NOCACHE can be used 1785*7c478bd9Sstevel@tonic-gate * to override that. If we don't have a page_t, make sure 1786*7c478bd9Sstevel@tonic-gate * NOCONSIST is set. 1787*7c478bd9Sstevel@tonic-gate */ 1788*7c478bd9Sstevel@tonic-gate a = attr; 1789*7c478bd9Sstevel@tonic-gate f = flags; 1790*7c478bd9Sstevel@tonic-gate if (pf_is_memory(pfn)) { 1791*7c478bd9Sstevel@tonic-gate if (!(a & HAT_PLAT_NOCACHE)) 1792*7c478bd9Sstevel@tonic-gate a |= HAT_STORECACHING_OK; 1793*7c478bd9Sstevel@tonic-gate 1794*7c478bd9Sstevel@tonic-gate if (f & HAT_LOAD_NOCONSIST) 1795*7c478bd9Sstevel@tonic-gate pp = NULL; 1796*7c478bd9Sstevel@tonic-gate else 1797*7c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 1798*7c478bd9Sstevel@tonic-gate } else { 1799*7c478bd9Sstevel@tonic-gate pp = NULL; 1800*7c478bd9Sstevel@tonic-gate f |= HAT_LOAD_NOCONSIST; 1801*7c478bd9Sstevel@tonic-gate } 1802*7c478bd9Sstevel@tonic-gate 1803*7c478bd9Sstevel@tonic-gate /* 1804*7c478bd9Sstevel@tonic-gate * load this page mapping 1805*7c478bd9Sstevel@tonic-gate */ 1806*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 1807*7c478bd9Sstevel@tonic-gate hati_load_common(hat, va, pp, a, f, level, pfn); 1808*7c478bd9Sstevel@tonic-gate 1809*7c478bd9Sstevel@tonic-gate /* 1810*7c478bd9Sstevel@tonic-gate * move to next page 1811*7c478bd9Sstevel@tonic-gate */ 1812*7c478bd9Sstevel@tonic-gate va += pgsize; 1813*7c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 1814*7c478bd9Sstevel@tonic-gate } 1815*7c478bd9Sstevel@tonic-gate HATOUT(hat_devload, hat, addr); 1816*7c478bd9Sstevel@tonic-gate } 1817*7c478bd9Sstevel@tonic-gate 1818*7c478bd9Sstevel@tonic-gate /* 1819*7c478bd9Sstevel@tonic-gate * void hat_unlock(hat, addr, len) 1820*7c478bd9Sstevel@tonic-gate * unlock the mappings to a given range of addresses 1821*7c478bd9Sstevel@tonic-gate * 1822*7c478bd9Sstevel@tonic-gate * Locks are tracked by ht_lock_cnt in the htable. 1823*7c478bd9Sstevel@tonic-gate */ 1824*7c478bd9Sstevel@tonic-gate void 1825*7c478bd9Sstevel@tonic-gate hat_unlock(hat_t *hat, caddr_t addr, size_t len) 1826*7c478bd9Sstevel@tonic-gate { 1827*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 1828*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 1829*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 1830*7c478bd9Sstevel@tonic-gate 1831*7c478bd9Sstevel@tonic-gate /* 1832*7c478bd9Sstevel@tonic-gate * kernel entries are always locked, we don't track lock counts 1833*7c478bd9Sstevel@tonic-gate */ 1834*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || eaddr <= kernelbase); 1835*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 1836*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 1837*7c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) 1838*7c478bd9Sstevel@tonic-gate return; 1839*7c478bd9Sstevel@tonic-gate if (eaddr > _userlimit) 1840*7c478bd9Sstevel@tonic-gate panic("hat_unlock() address out of range - above _userlimit"); 1841*7c478bd9Sstevel@tonic-gate 1842*7c478bd9Sstevel@tonic-gate ASSERT(AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 1843*7c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 1844*7c478bd9Sstevel@tonic-gate (void) htable_walk(hat, &ht, &vaddr, eaddr); 1845*7c478bd9Sstevel@tonic-gate if (ht == NULL) 1846*7c478bd9Sstevel@tonic-gate break; 1847*7c478bd9Sstevel@tonic-gate 1848*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 1849*7c478bd9Sstevel@tonic-gate 1850*7c478bd9Sstevel@tonic-gate if (ht->ht_lock_cnt < 1) 1851*7c478bd9Sstevel@tonic-gate panic("hat_unlock(): lock_cnt < 1, " 1852*7c478bd9Sstevel@tonic-gate "htable=%p, vaddr=%p\n", ht, (caddr_t)vaddr); 1853*7c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 1854*7c478bd9Sstevel@tonic-gate 1855*7c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 1856*7c478bd9Sstevel@tonic-gate } 1857*7c478bd9Sstevel@tonic-gate if (ht) 1858*7c478bd9Sstevel@tonic-gate htable_release(ht); 1859*7c478bd9Sstevel@tonic-gate } 1860*7c478bd9Sstevel@tonic-gate 1861*7c478bd9Sstevel@tonic-gate /* 1862*7c478bd9Sstevel@tonic-gate * Cross call service routine to demap a virtual page on 1863*7c478bd9Sstevel@tonic-gate * the current CPU or flush all mappings in TLB. 1864*7c478bd9Sstevel@tonic-gate */ 1865*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 1866*7c478bd9Sstevel@tonic-gate static int 1867*7c478bd9Sstevel@tonic-gate hati_demap_func(xc_arg_t a1, xc_arg_t a2, xc_arg_t a3) 1868*7c478bd9Sstevel@tonic-gate { 1869*7c478bd9Sstevel@tonic-gate hat_t *hat = (hat_t *)a1; 1870*7c478bd9Sstevel@tonic-gate caddr_t addr = (caddr_t)a2; 1871*7c478bd9Sstevel@tonic-gate 1872*7c478bd9Sstevel@tonic-gate /* 1873*7c478bd9Sstevel@tonic-gate * If the target hat isn't the kernel and this CPU isn't operating 1874*7c478bd9Sstevel@tonic-gate * in the target hat, we can ignore the cross call. 1875*7c478bd9Sstevel@tonic-gate */ 1876*7c478bd9Sstevel@tonic-gate if (hat != kas.a_hat && hat != CPU->cpu_current_hat) 1877*7c478bd9Sstevel@tonic-gate return (0); 1878*7c478bd9Sstevel@tonic-gate 1879*7c478bd9Sstevel@tonic-gate /* 1880*7c478bd9Sstevel@tonic-gate * For a normal address, we just flush one page mapping 1881*7c478bd9Sstevel@tonic-gate */ 1882*7c478bd9Sstevel@tonic-gate if ((uintptr_t)addr != DEMAP_ALL_ADDR) { 1883*7c478bd9Sstevel@tonic-gate mmu_tlbflush_entry((caddr_t)addr); 1884*7c478bd9Sstevel@tonic-gate return (0); 1885*7c478bd9Sstevel@tonic-gate } 1886*7c478bd9Sstevel@tonic-gate 1887*7c478bd9Sstevel@tonic-gate /* 1888*7c478bd9Sstevel@tonic-gate * Otherwise we reload cr3 to effect a complete TLB flush. 1889*7c478bd9Sstevel@tonic-gate * 1890*7c478bd9Sstevel@tonic-gate * A reload of cr3 on a VLP process also means we must also recopy in 1891*7c478bd9Sstevel@tonic-gate * the pte values from the struct hat 1892*7c478bd9Sstevel@tonic-gate */ 1893*7c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_VLP) { 1894*7c478bd9Sstevel@tonic-gate #if defined(__amd64) 1895*7c478bd9Sstevel@tonic-gate x86pte_t *vlpptep = CPU->cpu_hat_info->hci_vlp_l2ptes; 1896*7c478bd9Sstevel@tonic-gate 1897*7c478bd9Sstevel@tonic-gate VLP_COPY(hat->hat_vlp_ptes, vlpptep); 1898*7c478bd9Sstevel@tonic-gate #elif defined(__i386) 1899*7c478bd9Sstevel@tonic-gate reload_pae32(hat, CPU); 1900*7c478bd9Sstevel@tonic-gate #endif 1901*7c478bd9Sstevel@tonic-gate } 1902*7c478bd9Sstevel@tonic-gate reload_cr3(); 1903*7c478bd9Sstevel@tonic-gate return (0); 1904*7c478bd9Sstevel@tonic-gate } 1905*7c478bd9Sstevel@tonic-gate 1906*7c478bd9Sstevel@tonic-gate /* 1907*7c478bd9Sstevel@tonic-gate * Internal routine to do cross calls to invalidate a range of pages on 1908*7c478bd9Sstevel@tonic-gate * all CPUs using a given hat. 1909*7c478bd9Sstevel@tonic-gate */ 1910*7c478bd9Sstevel@tonic-gate void 1911*7c478bd9Sstevel@tonic-gate hat_demap(hat_t *hat, uintptr_t va) 1912*7c478bd9Sstevel@tonic-gate { 1913*7c478bd9Sstevel@tonic-gate extern int flushes_require_xcalls; /* from mp_startup.c */ 1914*7c478bd9Sstevel@tonic-gate cpuset_t justme; 1915*7c478bd9Sstevel@tonic-gate 1916*7c478bd9Sstevel@tonic-gate /* 1917*7c478bd9Sstevel@tonic-gate * If the hat is being destroyed, there are no more users, so 1918*7c478bd9Sstevel@tonic-gate * demap need not do anything. 1919*7c478bd9Sstevel@tonic-gate */ 1920*7c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_FREEING) 1921*7c478bd9Sstevel@tonic-gate return; 1922*7c478bd9Sstevel@tonic-gate 1923*7c478bd9Sstevel@tonic-gate /* 1924*7c478bd9Sstevel@tonic-gate * If demapping from a shared pagetable, we best demap the 1925*7c478bd9Sstevel@tonic-gate * entire set of user TLBs, since we don't know what addresses 1926*7c478bd9Sstevel@tonic-gate * these were shared at. 1927*7c478bd9Sstevel@tonic-gate */ 1928*7c478bd9Sstevel@tonic-gate if (hat->hat_flags & HAT_SHARED) { 1929*7c478bd9Sstevel@tonic-gate hat = kas.a_hat; 1930*7c478bd9Sstevel@tonic-gate va = DEMAP_ALL_ADDR; 1931*7c478bd9Sstevel@tonic-gate } 1932*7c478bd9Sstevel@tonic-gate 1933*7c478bd9Sstevel@tonic-gate /* 1934*7c478bd9Sstevel@tonic-gate * if not running with multiple CPUs, don't use cross calls 1935*7c478bd9Sstevel@tonic-gate */ 1936*7c478bd9Sstevel@tonic-gate if (panicstr || !flushes_require_xcalls) { 1937*7c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 1938*7c478bd9Sstevel@tonic-gate return; 1939*7c478bd9Sstevel@tonic-gate } 1940*7c478bd9Sstevel@tonic-gate 1941*7c478bd9Sstevel@tonic-gate 1942*7c478bd9Sstevel@tonic-gate /* 1943*7c478bd9Sstevel@tonic-gate * All CPUs must see kernel hat changes. 1944*7c478bd9Sstevel@tonic-gate */ 1945*7c478bd9Sstevel@tonic-gate if (hat == kas.a_hat) { 1946*7c478bd9Sstevel@tonic-gate kpreempt_disable(); 1947*7c478bd9Sstevel@tonic-gate xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, 1948*7c478bd9Sstevel@tonic-gate X_CALL_HIPRI, khat_cpuset, hati_demap_func); 1949*7c478bd9Sstevel@tonic-gate kpreempt_enable(); 1950*7c478bd9Sstevel@tonic-gate return; 1951*7c478bd9Sstevel@tonic-gate } 1952*7c478bd9Sstevel@tonic-gate 1953*7c478bd9Sstevel@tonic-gate /* 1954*7c478bd9Sstevel@tonic-gate * Otherwise we notify CPUs currently running in this HAT 1955*7c478bd9Sstevel@tonic-gate */ 1956*7c478bd9Sstevel@tonic-gate hat_enter(hat); 1957*7c478bd9Sstevel@tonic-gate kpreempt_disable(); 1958*7c478bd9Sstevel@tonic-gate CPUSET_ONLY(justme, CPU->cpu_id); 1959*7c478bd9Sstevel@tonic-gate if (CPUSET_ISEQUAL(hat->hat_cpus, justme)) 1960*7c478bd9Sstevel@tonic-gate (void) hati_demap_func((xc_arg_t)hat, (xc_arg_t)va, NULL); 1961*7c478bd9Sstevel@tonic-gate else 1962*7c478bd9Sstevel@tonic-gate xc_call((xc_arg_t)hat, (xc_arg_t)va, NULL, 1963*7c478bd9Sstevel@tonic-gate X_CALL_HIPRI, hat->hat_cpus, hati_demap_func); 1964*7c478bd9Sstevel@tonic-gate kpreempt_enable(); 1965*7c478bd9Sstevel@tonic-gate hat_exit(hat); 1966*7c478bd9Sstevel@tonic-gate } 1967*7c478bd9Sstevel@tonic-gate 1968*7c478bd9Sstevel@tonic-gate /* 1969*7c478bd9Sstevel@tonic-gate * Interior routine for HAT_UNLOADs from hat_unload_callback(), 1970*7c478bd9Sstevel@tonic-gate * hat_kmap_unload() OR from hat_steal() code. This routine doesn't 1971*7c478bd9Sstevel@tonic-gate * handle releasing of the htables. 1972*7c478bd9Sstevel@tonic-gate */ 1973*7c478bd9Sstevel@tonic-gate void 1974*7c478bd9Sstevel@tonic-gate hat_pte_unmap( 1975*7c478bd9Sstevel@tonic-gate htable_t *ht, 1976*7c478bd9Sstevel@tonic-gate uint_t entry, 1977*7c478bd9Sstevel@tonic-gate uint_t flags, 1978*7c478bd9Sstevel@tonic-gate x86pte_t old_pte, 1979*7c478bd9Sstevel@tonic-gate void *pte_ptr) 1980*7c478bd9Sstevel@tonic-gate { 1981*7c478bd9Sstevel@tonic-gate hat_t *hat = ht->ht_hat; 1982*7c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 1983*7c478bd9Sstevel@tonic-gate page_t *pp = NULL; 1984*7c478bd9Sstevel@tonic-gate level_t l = ht->ht_level; 1985*7c478bd9Sstevel@tonic-gate pfn_t pfn; 1986*7c478bd9Sstevel@tonic-gate 1987*7c478bd9Sstevel@tonic-gate /* 1988*7c478bd9Sstevel@tonic-gate * We always track the locking counts, even if nothing is unmapped 1989*7c478bd9Sstevel@tonic-gate */ 1990*7c478bd9Sstevel@tonic-gate if ((flags & HAT_UNLOAD_UNLOCK) != 0 && hat != kas.a_hat) { 1991*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_lock_cnt > 0); 1992*7c478bd9Sstevel@tonic-gate HTABLE_LOCK_DEC(ht); 1993*7c478bd9Sstevel@tonic-gate } 1994*7c478bd9Sstevel@tonic-gate 1995*7c478bd9Sstevel@tonic-gate /* 1996*7c478bd9Sstevel@tonic-gate * Figure out which page's mapping list lock to acquire using the PFN 1997*7c478bd9Sstevel@tonic-gate * passed in "old" PTE. We then attempt to invalidate the PTE. 1998*7c478bd9Sstevel@tonic-gate * If another thread, probably a hat_pageunload, has asynchronously 1999*7c478bd9Sstevel@tonic-gate * unmapped/remapped this address we'll loop here. 2000*7c478bd9Sstevel@tonic-gate */ 2001*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_busy > 0); 2002*7c478bd9Sstevel@tonic-gate while (PTE_ISVALID(old_pte)) { 2003*7c478bd9Sstevel@tonic-gate pfn = PTE2PFN(old_pte, l); 2004*7c478bd9Sstevel@tonic-gate if (PTE_GET(old_pte, PT_NOCONSIST)) { 2005*7c478bd9Sstevel@tonic-gate pp = NULL; 2006*7c478bd9Sstevel@tonic-gate } else { 2007*7c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 2008*7c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 2009*7c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 2010*7c478bd9Sstevel@tonic-gate } 2011*7c478bd9Sstevel@tonic-gate old_pte = x86pte_invalidate_pfn(ht, entry, pfn, pte_ptr); 2012*7c478bd9Sstevel@tonic-gate 2013*7c478bd9Sstevel@tonic-gate /* 2014*7c478bd9Sstevel@tonic-gate * If the page hadn't changed we've unmapped it and can proceed 2015*7c478bd9Sstevel@tonic-gate */ 2016*7c478bd9Sstevel@tonic-gate if (PTE_ISVALID(old_pte) && PTE2PFN(old_pte, l) == pfn) 2017*7c478bd9Sstevel@tonic-gate break; 2018*7c478bd9Sstevel@tonic-gate 2019*7c478bd9Sstevel@tonic-gate /* 2020*7c478bd9Sstevel@tonic-gate * Otherwise, we'll have to retry with the current old_pte. 2021*7c478bd9Sstevel@tonic-gate * Drop the hment lock, since the pfn may have changed. 2022*7c478bd9Sstevel@tonic-gate */ 2023*7c478bd9Sstevel@tonic-gate if (pp != NULL) { 2024*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2025*7c478bd9Sstevel@tonic-gate pp = NULL; 2026*7c478bd9Sstevel@tonic-gate } else { 2027*7c478bd9Sstevel@tonic-gate ASSERT(PTE_GET(old_pte, PT_NOCONSIST)); 2028*7c478bd9Sstevel@tonic-gate } 2029*7c478bd9Sstevel@tonic-gate } 2030*7c478bd9Sstevel@tonic-gate 2031*7c478bd9Sstevel@tonic-gate /* 2032*7c478bd9Sstevel@tonic-gate * If the old mapping wasn't valid, there's nothing more to do 2033*7c478bd9Sstevel@tonic-gate */ 2034*7c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(old_pte)) { 2035*7c478bd9Sstevel@tonic-gate if (pp != NULL) 2036*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2037*7c478bd9Sstevel@tonic-gate return; 2038*7c478bd9Sstevel@tonic-gate } 2039*7c478bd9Sstevel@tonic-gate 2040*7c478bd9Sstevel@tonic-gate /* 2041*7c478bd9Sstevel@tonic-gate * Take care of syncing any MOD/REF bits and removing the hment. 2042*7c478bd9Sstevel@tonic-gate */ 2043*7c478bd9Sstevel@tonic-gate if (pp != NULL) { 2044*7c478bd9Sstevel@tonic-gate if (!(flags & HAT_UNLOAD_NOSYNC)) 2045*7c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, l); 2046*7c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 2047*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2048*7c478bd9Sstevel@tonic-gate if (hm != NULL) 2049*7c478bd9Sstevel@tonic-gate hment_free(hm); 2050*7c478bd9Sstevel@tonic-gate } 2051*7c478bd9Sstevel@tonic-gate 2052*7c478bd9Sstevel@tonic-gate /* 2053*7c478bd9Sstevel@tonic-gate * Handle book keeping in the htable and hat 2054*7c478bd9Sstevel@tonic-gate */ 2055*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 2056*7c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 2057*7c478bd9Sstevel@tonic-gate PGCNT_DEC(hat, l); 2058*7c478bd9Sstevel@tonic-gate } 2059*7c478bd9Sstevel@tonic-gate 2060*7c478bd9Sstevel@tonic-gate /* 2061*7c478bd9Sstevel@tonic-gate * very cheap unload implementation to special case some kernel addresses 2062*7c478bd9Sstevel@tonic-gate */ 2063*7c478bd9Sstevel@tonic-gate static void 2064*7c478bd9Sstevel@tonic-gate hat_kmap_unload(caddr_t addr, size_t len, uint_t flags) 2065*7c478bd9Sstevel@tonic-gate { 2066*7c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2067*7c478bd9Sstevel@tonic-gate uintptr_t eva = va + len; 2068*7c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 2069*7c478bd9Sstevel@tonic-gate htable_t *ht; 2070*7c478bd9Sstevel@tonic-gate uint_t entry; 2071*7c478bd9Sstevel@tonic-gate void *pte_ptr; 2072*7c478bd9Sstevel@tonic-gate x86pte_t old_pte; 2073*7c478bd9Sstevel@tonic-gate 2074*7c478bd9Sstevel@tonic-gate for (; va < eva; va += MMU_PAGESIZE) { 2075*7c478bd9Sstevel@tonic-gate /* 2076*7c478bd9Sstevel@tonic-gate * Get the PTE 2077*7c478bd9Sstevel@tonic-gate */ 2078*7c478bd9Sstevel@tonic-gate pg_off = mmu_btop(va - mmu.kmap_addr); 2079*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 2080*7c478bd9Sstevel@tonic-gate pte_ptr = mmu.kmap_ptes + pg_off; 2081*7c478bd9Sstevel@tonic-gate old_pte = *(x86pte_t *)pte_ptr; 2082*7c478bd9Sstevel@tonic-gate } else { 2083*7c478bd9Sstevel@tonic-gate pte_ptr = (x86pte32_t *)mmu.kmap_ptes + pg_off; 2084*7c478bd9Sstevel@tonic-gate old_pte = *(x86pte32_t *)pte_ptr; 2085*7c478bd9Sstevel@tonic-gate } 2086*7c478bd9Sstevel@tonic-gate 2087*7c478bd9Sstevel@tonic-gate /* 2088*7c478bd9Sstevel@tonic-gate * get the htable / entry 2089*7c478bd9Sstevel@tonic-gate */ 2090*7c478bd9Sstevel@tonic-gate ht = mmu.kmap_htables[(va - mmu.kmap_htables[0]->ht_vaddr) 2091*7c478bd9Sstevel@tonic-gate >> LEVEL_SHIFT(1)]; 2092*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 2093*7c478bd9Sstevel@tonic-gate 2094*7c478bd9Sstevel@tonic-gate /* 2095*7c478bd9Sstevel@tonic-gate * use mostly common code to unmap it. 2096*7c478bd9Sstevel@tonic-gate */ 2097*7c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, pte_ptr); 2098*7c478bd9Sstevel@tonic-gate } 2099*7c478bd9Sstevel@tonic-gate } 2100*7c478bd9Sstevel@tonic-gate 2101*7c478bd9Sstevel@tonic-gate 2102*7c478bd9Sstevel@tonic-gate /* 2103*7c478bd9Sstevel@tonic-gate * unload a range of virtual address space (no callback) 2104*7c478bd9Sstevel@tonic-gate */ 2105*7c478bd9Sstevel@tonic-gate void 2106*7c478bd9Sstevel@tonic-gate hat_unload(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 2107*7c478bd9Sstevel@tonic-gate { 2108*7c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 2109*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || va + len <= kernelbase); 2110*7c478bd9Sstevel@tonic-gate 2111*7c478bd9Sstevel@tonic-gate /* 2112*7c478bd9Sstevel@tonic-gate * special case for performance. 2113*7c478bd9Sstevel@tonic-gate */ 2114*7c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= va && va < mmu.kmap_eaddr) { 2115*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat); 2116*7c478bd9Sstevel@tonic-gate hat_kmap_unload(addr, len, flags); 2117*7c478bd9Sstevel@tonic-gate return; 2118*7c478bd9Sstevel@tonic-gate } 2119*7c478bd9Sstevel@tonic-gate hat_unload_callback(hat, addr, len, flags, NULL); 2120*7c478bd9Sstevel@tonic-gate } 2121*7c478bd9Sstevel@tonic-gate 2122*7c478bd9Sstevel@tonic-gate /* 2123*7c478bd9Sstevel@tonic-gate * Do the callbacks for ranges being unloaded. 2124*7c478bd9Sstevel@tonic-gate */ 2125*7c478bd9Sstevel@tonic-gate typedef struct range_info { 2126*7c478bd9Sstevel@tonic-gate uintptr_t rng_va; 2127*7c478bd9Sstevel@tonic-gate ulong_t rng_cnt; 2128*7c478bd9Sstevel@tonic-gate level_t rng_level; 2129*7c478bd9Sstevel@tonic-gate } range_info_t; 2130*7c478bd9Sstevel@tonic-gate 2131*7c478bd9Sstevel@tonic-gate static void 2132*7c478bd9Sstevel@tonic-gate handle_ranges(hat_callback_t *cb, uint_t cnt, range_info_t *range) 2133*7c478bd9Sstevel@tonic-gate { 2134*7c478bd9Sstevel@tonic-gate /* 2135*7c478bd9Sstevel@tonic-gate * do callbacks to upper level VM system 2136*7c478bd9Sstevel@tonic-gate */ 2137*7c478bd9Sstevel@tonic-gate while (cb != NULL && cnt > 0) { 2138*7c478bd9Sstevel@tonic-gate --cnt; 2139*7c478bd9Sstevel@tonic-gate cb->hcb_start_addr = (caddr_t)range[cnt].rng_va; 2140*7c478bd9Sstevel@tonic-gate cb->hcb_end_addr = cb->hcb_start_addr; 2141*7c478bd9Sstevel@tonic-gate cb->hcb_end_addr += 2142*7c478bd9Sstevel@tonic-gate range[cnt].rng_cnt << LEVEL_SIZE(range[cnt].rng_level); 2143*7c478bd9Sstevel@tonic-gate cb->hcb_function(cb); 2144*7c478bd9Sstevel@tonic-gate } 2145*7c478bd9Sstevel@tonic-gate } 2146*7c478bd9Sstevel@tonic-gate 2147*7c478bd9Sstevel@tonic-gate /* 2148*7c478bd9Sstevel@tonic-gate * Unload a given range of addresses (has optional callback) 2149*7c478bd9Sstevel@tonic-gate * 2150*7c478bd9Sstevel@tonic-gate * Flags: 2151*7c478bd9Sstevel@tonic-gate * define HAT_UNLOAD 0x00 2152*7c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_NOSYNC 0x02 2153*7c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNLOCK 0x04 2154*7c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_OTHER 0x08 - not used 2155*7c478bd9Sstevel@tonic-gate * define HAT_UNLOAD_UNMAP 0x10 - same as HAT_UNLOAD 2156*7c478bd9Sstevel@tonic-gate */ 2157*7c478bd9Sstevel@tonic-gate #define MAX_UNLOAD_CNT (8) 2158*7c478bd9Sstevel@tonic-gate void 2159*7c478bd9Sstevel@tonic-gate hat_unload_callback( 2160*7c478bd9Sstevel@tonic-gate hat_t *hat, 2161*7c478bd9Sstevel@tonic-gate caddr_t addr, 2162*7c478bd9Sstevel@tonic-gate size_t len, 2163*7c478bd9Sstevel@tonic-gate uint_t flags, 2164*7c478bd9Sstevel@tonic-gate hat_callback_t *cb) 2165*7c478bd9Sstevel@tonic-gate { 2166*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 2167*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 2168*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 2169*7c478bd9Sstevel@tonic-gate uint_t entry; 2170*7c478bd9Sstevel@tonic-gate uintptr_t last_va = (uintptr_t)-1L; 2171*7c478bd9Sstevel@tonic-gate range_info_t r[MAX_UNLOAD_CNT]; 2172*7c478bd9Sstevel@tonic-gate uint_t r_cnt = 0; 2173*7c478bd9Sstevel@tonic-gate x86pte_t old_pte; 2174*7c478bd9Sstevel@tonic-gate 2175*7c478bd9Sstevel@tonic-gate HATIN(hat_unload_callback, hat, addr, len); 2176*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || eaddr <= kernelbase); 2177*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 2178*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2179*7c478bd9Sstevel@tonic-gate 2180*7c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 2181*7c478bd9Sstevel@tonic-gate old_pte = htable_walk(hat, &ht, &vaddr, eaddr); 2182*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2183*7c478bd9Sstevel@tonic-gate break; 2184*7c478bd9Sstevel@tonic-gate 2185*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 2186*7c478bd9Sstevel@tonic-gate 2187*7c478bd9Sstevel@tonic-gate if (vaddr < (uintptr_t)addr) 2188*7c478bd9Sstevel@tonic-gate panic("hat_unload_callback(): unmap inside large page"); 2189*7c478bd9Sstevel@tonic-gate 2190*7c478bd9Sstevel@tonic-gate /* 2191*7c478bd9Sstevel@tonic-gate * We'll do the call backs for contiguous ranges 2192*7c478bd9Sstevel@tonic-gate */ 2193*7c478bd9Sstevel@tonic-gate if (vaddr != last_va || 2194*7c478bd9Sstevel@tonic-gate (r_cnt > 0 && r[r_cnt - 1].rng_level != ht->ht_level)) { 2195*7c478bd9Sstevel@tonic-gate if (r_cnt == MAX_UNLOAD_CNT) { 2196*7c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 2197*7c478bd9Sstevel@tonic-gate r_cnt = 0; 2198*7c478bd9Sstevel@tonic-gate } 2199*7c478bd9Sstevel@tonic-gate r[r_cnt].rng_va = vaddr; 2200*7c478bd9Sstevel@tonic-gate r[r_cnt].rng_cnt = 0; 2201*7c478bd9Sstevel@tonic-gate r[r_cnt].rng_level = ht->ht_level; 2202*7c478bd9Sstevel@tonic-gate ++r_cnt; 2203*7c478bd9Sstevel@tonic-gate } 2204*7c478bd9Sstevel@tonic-gate 2205*7c478bd9Sstevel@tonic-gate /* 2206*7c478bd9Sstevel@tonic-gate * Unload one mapping from the page tables. 2207*7c478bd9Sstevel@tonic-gate */ 2208*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 2209*7c478bd9Sstevel@tonic-gate hat_pte_unmap(ht, entry, flags, old_pte, NULL); 2210*7c478bd9Sstevel@tonic-gate 2211*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level <= mmu.max_page_level); 2212*7c478bd9Sstevel@tonic-gate last_va = vaddr; 2213*7c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 2214*7c478bd9Sstevel@tonic-gate ++r[r_cnt - 1].rng_cnt; 2215*7c478bd9Sstevel@tonic-gate } 2216*7c478bd9Sstevel@tonic-gate if (ht) 2217*7c478bd9Sstevel@tonic-gate htable_release(ht); 2218*7c478bd9Sstevel@tonic-gate 2219*7c478bd9Sstevel@tonic-gate /* 2220*7c478bd9Sstevel@tonic-gate * handle last range for callbacks 2221*7c478bd9Sstevel@tonic-gate */ 2222*7c478bd9Sstevel@tonic-gate if (r_cnt > 0) 2223*7c478bd9Sstevel@tonic-gate handle_ranges(cb, r_cnt, r); 2224*7c478bd9Sstevel@tonic-gate 2225*7c478bd9Sstevel@tonic-gate HATOUT(hat_unload_callback, hat, addr); 2226*7c478bd9Sstevel@tonic-gate } 2227*7c478bd9Sstevel@tonic-gate 2228*7c478bd9Sstevel@tonic-gate /* 2229*7c478bd9Sstevel@tonic-gate * synchronize mapping with software data structures 2230*7c478bd9Sstevel@tonic-gate * 2231*7c478bd9Sstevel@tonic-gate * This interface is currently only used by the working set monitor 2232*7c478bd9Sstevel@tonic-gate * driver. 2233*7c478bd9Sstevel@tonic-gate */ 2234*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2235*7c478bd9Sstevel@tonic-gate void 2236*7c478bd9Sstevel@tonic-gate hat_sync(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 2237*7c478bd9Sstevel@tonic-gate { 2238*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 2239*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 2240*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 2241*7c478bd9Sstevel@tonic-gate uint_t entry; 2242*7c478bd9Sstevel@tonic-gate x86pte_t pte; 2243*7c478bd9Sstevel@tonic-gate x86pte_t save_pte; 2244*7c478bd9Sstevel@tonic-gate x86pte_t new; 2245*7c478bd9Sstevel@tonic-gate page_t *pp; 2246*7c478bd9Sstevel@tonic-gate 2247*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 2248*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 2249*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2250*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || eaddr <= kernelbase); 2251*7c478bd9Sstevel@tonic-gate 2252*7c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 2253*7c478bd9Sstevel@tonic-gate try_again: 2254*7c478bd9Sstevel@tonic-gate pte = htable_walk(hat, &ht, &vaddr, eaddr); 2255*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2256*7c478bd9Sstevel@tonic-gate break; 2257*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 2258*7c478bd9Sstevel@tonic-gate 2259*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_NOSYNC) || 2260*7c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) 2261*7c478bd9Sstevel@tonic-gate continue; 2262*7c478bd9Sstevel@tonic-gate 2263*7c478bd9Sstevel@tonic-gate /* 2264*7c478bd9Sstevel@tonic-gate * We need to acquire the mapping list lock to protect 2265*7c478bd9Sstevel@tonic-gate * against hat_pageunload(), hat_unload(), etc. 2266*7c478bd9Sstevel@tonic-gate */ 2267*7c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(pte, ht->ht_level)); 2268*7c478bd9Sstevel@tonic-gate if (pp == NULL) 2269*7c478bd9Sstevel@tonic-gate break; 2270*7c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 2271*7c478bd9Sstevel@tonic-gate save_pte = pte; 2272*7c478bd9Sstevel@tonic-gate pte = x86pte_get(ht, entry); 2273*7c478bd9Sstevel@tonic-gate if (pte != save_pte) { 2274*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2275*7c478bd9Sstevel@tonic-gate goto try_again; 2276*7c478bd9Sstevel@tonic-gate } 2277*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_NOSYNC) || 2278*7c478bd9Sstevel@tonic-gate PTE_GET(pte, PT_REF | PT_MOD) == 0) { 2279*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2280*7c478bd9Sstevel@tonic-gate continue; 2281*7c478bd9Sstevel@tonic-gate } 2282*7c478bd9Sstevel@tonic-gate 2283*7c478bd9Sstevel@tonic-gate /* 2284*7c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. We may compete with 2285*7c478bd9Sstevel@tonic-gate * hardware updating the R/M bits and have to try again. 2286*7c478bd9Sstevel@tonic-gate */ 2287*7c478bd9Sstevel@tonic-gate if (flags == HAT_SYNC_ZERORM) { 2288*7c478bd9Sstevel@tonic-gate new = pte; 2289*7c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 2290*7c478bd9Sstevel@tonic-gate pte = hati_update_pte(ht, entry, pte, new); 2291*7c478bd9Sstevel@tonic-gate if (pte != 0) { 2292*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2293*7c478bd9Sstevel@tonic-gate goto try_again; 2294*7c478bd9Sstevel@tonic-gate } 2295*7c478bd9Sstevel@tonic-gate } else { 2296*7c478bd9Sstevel@tonic-gate /* 2297*7c478bd9Sstevel@tonic-gate * sync the PTE to the page_t 2298*7c478bd9Sstevel@tonic-gate */ 2299*7c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, save_pte, ht->ht_level); 2300*7c478bd9Sstevel@tonic-gate } 2301*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2302*7c478bd9Sstevel@tonic-gate } 2303*7c478bd9Sstevel@tonic-gate if (ht) 2304*7c478bd9Sstevel@tonic-gate htable_release(ht); 2305*7c478bd9Sstevel@tonic-gate } 2306*7c478bd9Sstevel@tonic-gate 2307*7c478bd9Sstevel@tonic-gate /* 2308*7c478bd9Sstevel@tonic-gate * void hat_map(hat, addr, len, flags) 2309*7c478bd9Sstevel@tonic-gate */ 2310*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2311*7c478bd9Sstevel@tonic-gate void 2312*7c478bd9Sstevel@tonic-gate hat_map(hat_t *hat, caddr_t addr, size_t len, uint_t flags) 2313*7c478bd9Sstevel@tonic-gate { 2314*7c478bd9Sstevel@tonic-gate /* does nothing */ 2315*7c478bd9Sstevel@tonic-gate } 2316*7c478bd9Sstevel@tonic-gate 2317*7c478bd9Sstevel@tonic-gate /* 2318*7c478bd9Sstevel@tonic-gate * uint_t hat_getattr(hat, addr, *attr) 2319*7c478bd9Sstevel@tonic-gate * returns attr for <hat,addr> in *attr. returns 0 if there was a 2320*7c478bd9Sstevel@tonic-gate * mapping and *attr is valid, nonzero if there was no mapping and 2321*7c478bd9Sstevel@tonic-gate * *attr is not valid. 2322*7c478bd9Sstevel@tonic-gate */ 2323*7c478bd9Sstevel@tonic-gate uint_t 2324*7c478bd9Sstevel@tonic-gate hat_getattr(hat_t *hat, caddr_t addr, uint_t *attr) 2325*7c478bd9Sstevel@tonic-gate { 2326*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 2327*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 2328*7c478bd9Sstevel@tonic-gate x86pte_t pte; 2329*7c478bd9Sstevel@tonic-gate 2330*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || vaddr < kernelbase); 2331*7c478bd9Sstevel@tonic-gate 2332*7c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 2333*7c478bd9Sstevel@tonic-gate return ((uint_t)-1); 2334*7c478bd9Sstevel@tonic-gate 2335*7c478bd9Sstevel@tonic-gate ht = htable_getpte(hat, vaddr, NULL, &pte, MAX_PAGE_LEVEL); 2336*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2337*7c478bd9Sstevel@tonic-gate return ((uint_t)-1); 2338*7c478bd9Sstevel@tonic-gate 2339*7c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte) || !PTE_ISPAGE(pte, ht->ht_level)) { 2340*7c478bd9Sstevel@tonic-gate htable_release(ht); 2341*7c478bd9Sstevel@tonic-gate return ((uint_t)-1); 2342*7c478bd9Sstevel@tonic-gate } 2343*7c478bd9Sstevel@tonic-gate 2344*7c478bd9Sstevel@tonic-gate *attr = PROT_READ; 2345*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 2346*7c478bd9Sstevel@tonic-gate *attr |= PROT_WRITE; 2347*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_USER)) 2348*7c478bd9Sstevel@tonic-gate *attr |= PROT_USER; 2349*7c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, mmu.pt_nx)) 2350*7c478bd9Sstevel@tonic-gate *attr |= PROT_EXEC; 2351*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_NOSYNC)) 2352*7c478bd9Sstevel@tonic-gate *attr |= HAT_NOSYNC; 2353*7c478bd9Sstevel@tonic-gate htable_release(ht); 2354*7c478bd9Sstevel@tonic-gate return (0); 2355*7c478bd9Sstevel@tonic-gate } 2356*7c478bd9Sstevel@tonic-gate 2357*7c478bd9Sstevel@tonic-gate /* 2358*7c478bd9Sstevel@tonic-gate * hat_updateattr() applies the given attribute change to an existing mapping 2359*7c478bd9Sstevel@tonic-gate */ 2360*7c478bd9Sstevel@tonic-gate #define HAT_LOAD_ATTR 1 2361*7c478bd9Sstevel@tonic-gate #define HAT_SET_ATTR 2 2362*7c478bd9Sstevel@tonic-gate #define HAT_CLR_ATTR 3 2363*7c478bd9Sstevel@tonic-gate 2364*7c478bd9Sstevel@tonic-gate static void 2365*7c478bd9Sstevel@tonic-gate hat_updateattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr, int what) 2366*7c478bd9Sstevel@tonic-gate { 2367*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 2368*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = (uintptr_t)addr + len; 2369*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 2370*7c478bd9Sstevel@tonic-gate uint_t entry; 2371*7c478bd9Sstevel@tonic-gate x86pte_t oldpte, newpte; 2372*7c478bd9Sstevel@tonic-gate page_t *pp; 2373*7c478bd9Sstevel@tonic-gate 2374*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 2375*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2376*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 2377*7c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 2378*7c478bd9Sstevel@tonic-gate for (; vaddr < eaddr; vaddr += LEVEL_SIZE(ht->ht_level)) { 2379*7c478bd9Sstevel@tonic-gate try_again: 2380*7c478bd9Sstevel@tonic-gate oldpte = htable_walk(hat, &ht, &vaddr, eaddr); 2381*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2382*7c478bd9Sstevel@tonic-gate break; 2383*7c478bd9Sstevel@tonic-gate if (PTE_GET(oldpte, PT_NOCONSIST)) 2384*7c478bd9Sstevel@tonic-gate continue; 2385*7c478bd9Sstevel@tonic-gate 2386*7c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(oldpte, ht->ht_level)); 2387*7c478bd9Sstevel@tonic-gate if (pp == NULL) 2388*7c478bd9Sstevel@tonic-gate continue; 2389*7c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 2390*7c478bd9Sstevel@tonic-gate 2391*7c478bd9Sstevel@tonic-gate newpte = oldpte; 2392*7c478bd9Sstevel@tonic-gate /* 2393*7c478bd9Sstevel@tonic-gate * We found a page table entry in the desired range, 2394*7c478bd9Sstevel@tonic-gate * figure out the new attributes. 2395*7c478bd9Sstevel@tonic-gate */ 2396*7c478bd9Sstevel@tonic-gate if (what == HAT_SET_ATTR || what == HAT_LOAD_ATTR) { 2397*7c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && 2398*7c478bd9Sstevel@tonic-gate !PTE_GET(oldpte, PT_WRITABLE)) 2399*7c478bd9Sstevel@tonic-gate newpte |= PT_WRITABLE; 2400*7c478bd9Sstevel@tonic-gate 2401*7c478bd9Sstevel@tonic-gate if ((attr & HAT_NOSYNC) && !PTE_GET(oldpte, PT_NOSYNC)) 2402*7c478bd9Sstevel@tonic-gate newpte |= PT_NOSYNC; 2403*7c478bd9Sstevel@tonic-gate 2404*7c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && PTE_GET(oldpte, mmu.pt_nx)) 2405*7c478bd9Sstevel@tonic-gate newpte &= ~mmu.pt_nx; 2406*7c478bd9Sstevel@tonic-gate } 2407*7c478bd9Sstevel@tonic-gate 2408*7c478bd9Sstevel@tonic-gate if (what == HAT_LOAD_ATTR) { 2409*7c478bd9Sstevel@tonic-gate if (!(attr & PROT_WRITE) && 2410*7c478bd9Sstevel@tonic-gate PTE_GET(oldpte, PT_WRITABLE)) 2411*7c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 2412*7c478bd9Sstevel@tonic-gate 2413*7c478bd9Sstevel@tonic-gate if (!(attr & HAT_NOSYNC) && PTE_GET(oldpte, PT_NOSYNC)) 2414*7c478bd9Sstevel@tonic-gate newpte &= ~PT_NOSYNC; 2415*7c478bd9Sstevel@tonic-gate 2416*7c478bd9Sstevel@tonic-gate if (!(attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 2417*7c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 2418*7c478bd9Sstevel@tonic-gate } 2419*7c478bd9Sstevel@tonic-gate 2420*7c478bd9Sstevel@tonic-gate if (what == HAT_CLR_ATTR) { 2421*7c478bd9Sstevel@tonic-gate if ((attr & PROT_WRITE) && PTE_GET(oldpte, PT_WRITABLE)) 2422*7c478bd9Sstevel@tonic-gate newpte &= ~PT_WRITABLE; 2423*7c478bd9Sstevel@tonic-gate 2424*7c478bd9Sstevel@tonic-gate if ((attr & HAT_NOSYNC) && PTE_GET(oldpte, PT_NOSYNC)) 2425*7c478bd9Sstevel@tonic-gate newpte &= ~PT_NOSYNC; 2426*7c478bd9Sstevel@tonic-gate 2427*7c478bd9Sstevel@tonic-gate if ((attr & PROT_EXEC) && !PTE_GET(oldpte, mmu.pt_nx)) 2428*7c478bd9Sstevel@tonic-gate newpte |= mmu.pt_nx; 2429*7c478bd9Sstevel@tonic-gate } 2430*7c478bd9Sstevel@tonic-gate 2431*7c478bd9Sstevel@tonic-gate /* 2432*7c478bd9Sstevel@tonic-gate * what about PROT_READ or others? this code only handles: 2433*7c478bd9Sstevel@tonic-gate * EXEC, WRITE, NOSYNC 2434*7c478bd9Sstevel@tonic-gate */ 2435*7c478bd9Sstevel@tonic-gate 2436*7c478bd9Sstevel@tonic-gate /* 2437*7c478bd9Sstevel@tonic-gate * If new PTE really changed, update the table. 2438*7c478bd9Sstevel@tonic-gate */ 2439*7c478bd9Sstevel@tonic-gate if (newpte != oldpte) { 2440*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 2441*7c478bd9Sstevel@tonic-gate oldpte = hati_update_pte(ht, entry, oldpte, newpte); 2442*7c478bd9Sstevel@tonic-gate if (oldpte != 0) { 2443*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2444*7c478bd9Sstevel@tonic-gate goto try_again; 2445*7c478bd9Sstevel@tonic-gate } 2446*7c478bd9Sstevel@tonic-gate } 2447*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2448*7c478bd9Sstevel@tonic-gate } 2449*7c478bd9Sstevel@tonic-gate if (ht) 2450*7c478bd9Sstevel@tonic-gate htable_release(ht); 2451*7c478bd9Sstevel@tonic-gate } 2452*7c478bd9Sstevel@tonic-gate 2453*7c478bd9Sstevel@tonic-gate /* 2454*7c478bd9Sstevel@tonic-gate * Various wrappers for hat_updateattr() 2455*7c478bd9Sstevel@tonic-gate */ 2456*7c478bd9Sstevel@tonic-gate void 2457*7c478bd9Sstevel@tonic-gate hat_setattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 2458*7c478bd9Sstevel@tonic-gate { 2459*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= kernelbase); 2460*7c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_SET_ATTR); 2461*7c478bd9Sstevel@tonic-gate } 2462*7c478bd9Sstevel@tonic-gate 2463*7c478bd9Sstevel@tonic-gate void 2464*7c478bd9Sstevel@tonic-gate hat_clrattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 2465*7c478bd9Sstevel@tonic-gate { 2466*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= kernelbase); 2467*7c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_CLR_ATTR); 2468*7c478bd9Sstevel@tonic-gate } 2469*7c478bd9Sstevel@tonic-gate 2470*7c478bd9Sstevel@tonic-gate void 2471*7c478bd9Sstevel@tonic-gate hat_chgattr(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 2472*7c478bd9Sstevel@tonic-gate { 2473*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= kernelbase); 2474*7c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, attr, HAT_LOAD_ATTR); 2475*7c478bd9Sstevel@tonic-gate } 2476*7c478bd9Sstevel@tonic-gate 2477*7c478bd9Sstevel@tonic-gate void 2478*7c478bd9Sstevel@tonic-gate hat_chgprot(hat_t *hat, caddr_t addr, size_t len, uint_t vprot) 2479*7c478bd9Sstevel@tonic-gate { 2480*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || (uintptr_t)addr + len <= kernelbase); 2481*7c478bd9Sstevel@tonic-gate hat_updateattr(hat, addr, len, vprot & HAT_PROT_MASK, HAT_LOAD_ATTR); 2482*7c478bd9Sstevel@tonic-gate } 2483*7c478bd9Sstevel@tonic-gate 2484*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2485*7c478bd9Sstevel@tonic-gate void 2486*7c478bd9Sstevel@tonic-gate hat_chgattr_pagedir(hat_t *hat, caddr_t addr, size_t len, uint_t attr) 2487*7c478bd9Sstevel@tonic-gate { 2488*7c478bd9Sstevel@tonic-gate panic("hat_chgattr_pgdir() not supported - used by 80387 emulation"); 2489*7c478bd9Sstevel@tonic-gate } 2490*7c478bd9Sstevel@tonic-gate 2491*7c478bd9Sstevel@tonic-gate /* 2492*7c478bd9Sstevel@tonic-gate * size_t hat_getpagesize(hat, addr) 2493*7c478bd9Sstevel@tonic-gate * returns pagesize in bytes for <hat, addr>. returns -1 of there is 2494*7c478bd9Sstevel@tonic-gate * no mapping. This is an advisory call. 2495*7c478bd9Sstevel@tonic-gate */ 2496*7c478bd9Sstevel@tonic-gate ssize_t 2497*7c478bd9Sstevel@tonic-gate hat_getpagesize(hat_t *hat, caddr_t addr) 2498*7c478bd9Sstevel@tonic-gate { 2499*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 2500*7c478bd9Sstevel@tonic-gate htable_t *ht; 2501*7c478bd9Sstevel@tonic-gate size_t pagesize; 2502*7c478bd9Sstevel@tonic-gate 2503*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || vaddr < kernelbase); 2504*7c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 2505*7c478bd9Sstevel@tonic-gate return (-1); 2506*7c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, NULL); 2507*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2508*7c478bd9Sstevel@tonic-gate return (-1); 2509*7c478bd9Sstevel@tonic-gate pagesize = LEVEL_SIZE(ht->ht_level); 2510*7c478bd9Sstevel@tonic-gate htable_release(ht); 2511*7c478bd9Sstevel@tonic-gate return (pagesize); 2512*7c478bd9Sstevel@tonic-gate } 2513*7c478bd9Sstevel@tonic-gate 2514*7c478bd9Sstevel@tonic-gate 2515*7c478bd9Sstevel@tonic-gate 2516*7c478bd9Sstevel@tonic-gate /* 2517*7c478bd9Sstevel@tonic-gate * pfn_t hat_getpfnum(hat, addr) 2518*7c478bd9Sstevel@tonic-gate * returns pfn for <hat, addr> or PFN_INVALID if mapping is invalid. 2519*7c478bd9Sstevel@tonic-gate */ 2520*7c478bd9Sstevel@tonic-gate pfn_t 2521*7c478bd9Sstevel@tonic-gate hat_getpfnum(hat_t *hat, caddr_t addr) 2522*7c478bd9Sstevel@tonic-gate { 2523*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 2524*7c478bd9Sstevel@tonic-gate htable_t *ht; 2525*7c478bd9Sstevel@tonic-gate uint_t entry; 2526*7c478bd9Sstevel@tonic-gate pfn_t pfn = PFN_INVALID; 2527*7c478bd9Sstevel@tonic-gate 2528*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || vaddr < kernelbase); 2529*7c478bd9Sstevel@tonic-gate if (khat_running == 0) 2530*7c478bd9Sstevel@tonic-gate panic("hat_getpfnum(): called too early\n"); 2531*7c478bd9Sstevel@tonic-gate 2532*7c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 2533*7c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2534*7c478bd9Sstevel@tonic-gate 2535*7c478bd9Sstevel@tonic-gate /* 2536*7c478bd9Sstevel@tonic-gate * A very common use of hat_getpfnum() is from the DDI for kernel pages. 2537*7c478bd9Sstevel@tonic-gate * Use the kmap_ptes (which also covers the 32 bit heap) to speed 2538*7c478bd9Sstevel@tonic-gate * this up. 2539*7c478bd9Sstevel@tonic-gate */ 2540*7c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 2541*7c478bd9Sstevel@tonic-gate x86pte_t pte; 2542*7c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 2543*7c478bd9Sstevel@tonic-gate 2544*7c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 2545*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) { 2546*7c478bd9Sstevel@tonic-gate #ifdef __i386 2547*7c478bd9Sstevel@tonic-gate volatile x86pte_t *p = mmu.kmap_ptes + pg_off; 2548*7c478bd9Sstevel@tonic-gate 2549*7c478bd9Sstevel@tonic-gate do { 2550*7c478bd9Sstevel@tonic-gate pte = *p; 2551*7c478bd9Sstevel@tonic-gate } while (pte != *p); 2552*7c478bd9Sstevel@tonic-gate #else 2553*7c478bd9Sstevel@tonic-gate pte = mmu.kmap_ptes[pg_off]; 2554*7c478bd9Sstevel@tonic-gate #endif 2555*7c478bd9Sstevel@tonic-gate } else { 2556*7c478bd9Sstevel@tonic-gate pte = ((x86pte32_t *)mmu.kmap_ptes)[pg_off]; 2557*7c478bd9Sstevel@tonic-gate } 2558*7c478bd9Sstevel@tonic-gate if (!PTE_ISVALID(pte)) 2559*7c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2560*7c478bd9Sstevel@tonic-gate /*LINTED [use of constant 0 causes a silly lint warning] */ 2561*7c478bd9Sstevel@tonic-gate return (PTE2PFN(pte, 0)); 2562*7c478bd9Sstevel@tonic-gate } 2563*7c478bd9Sstevel@tonic-gate 2564*7c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 2565*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2566*7c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2567*7c478bd9Sstevel@tonic-gate ASSERT(vaddr >= ht->ht_vaddr); 2568*7c478bd9Sstevel@tonic-gate ASSERT(vaddr <= HTABLE_LAST_PAGE(ht)); 2569*7c478bd9Sstevel@tonic-gate pfn = PTE2PFN(x86pte_get(ht, entry), ht->ht_level); 2570*7c478bd9Sstevel@tonic-gate if (ht->ht_level > 0) 2571*7c478bd9Sstevel@tonic-gate pfn += mmu_btop(vaddr & LEVEL_OFFSET(ht->ht_level)); 2572*7c478bd9Sstevel@tonic-gate htable_release(ht); 2573*7c478bd9Sstevel@tonic-gate return (pfn); 2574*7c478bd9Sstevel@tonic-gate } 2575*7c478bd9Sstevel@tonic-gate 2576*7c478bd9Sstevel@tonic-gate /* 2577*7c478bd9Sstevel@tonic-gate * hat_getkpfnum() is an obsolete DDI routine, and its use is discouraged. 2578*7c478bd9Sstevel@tonic-gate * Use hat_getpfnum(kas.a_hat, ...) instead. 2579*7c478bd9Sstevel@tonic-gate * 2580*7c478bd9Sstevel@tonic-gate * We'd like to return PFN_INVALID if the mappings have underlying page_t's 2581*7c478bd9Sstevel@tonic-gate * but can't right now due to the fact that some software has grown to use 2582*7c478bd9Sstevel@tonic-gate * this interface incorrectly. So for now when the interface is misused, 2583*7c478bd9Sstevel@tonic-gate * return a warning to the user that in the future it won't work in the 2584*7c478bd9Sstevel@tonic-gate * way they're abusing it, and carry on. 2585*7c478bd9Sstevel@tonic-gate * 2586*7c478bd9Sstevel@tonic-gate * Note that hat_getkpfnum() is never supported on amd64. 2587*7c478bd9Sstevel@tonic-gate */ 2588*7c478bd9Sstevel@tonic-gate #if !defined(__amd64) 2589*7c478bd9Sstevel@tonic-gate pfn_t 2590*7c478bd9Sstevel@tonic-gate hat_getkpfnum(caddr_t addr) 2591*7c478bd9Sstevel@tonic-gate { 2592*7c478bd9Sstevel@tonic-gate pfn_t pfn; 2593*7c478bd9Sstevel@tonic-gate int badcaller = 0; 2594*7c478bd9Sstevel@tonic-gate 2595*7c478bd9Sstevel@tonic-gate 2596*7c478bd9Sstevel@tonic-gate if (khat_running == 0) 2597*7c478bd9Sstevel@tonic-gate panic("hat_getkpfnum(): called too early\n"); 2598*7c478bd9Sstevel@tonic-gate if ((uintptr_t)addr < kernelbase) 2599*7c478bd9Sstevel@tonic-gate return (PFN_INVALID); 2600*7c478bd9Sstevel@tonic-gate 2601*7c478bd9Sstevel@tonic-gate 2602*7c478bd9Sstevel@tonic-gate if (segkpm && IS_KPM_ADDR(addr)) { 2603*7c478bd9Sstevel@tonic-gate badcaller = 1; 2604*7c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(addr); 2605*7c478bd9Sstevel@tonic-gate } else { 2606*7c478bd9Sstevel@tonic-gate pfn = hat_getpfnum(kas.a_hat, addr); 2607*7c478bd9Sstevel@tonic-gate badcaller = pf_is_memory(pfn); 2608*7c478bd9Sstevel@tonic-gate } 2609*7c478bd9Sstevel@tonic-gate 2610*7c478bd9Sstevel@tonic-gate if (badcaller) 2611*7c478bd9Sstevel@tonic-gate hat_getkpfnum_badcall(caller()); 2612*7c478bd9Sstevel@tonic-gate return (pfn); 2613*7c478bd9Sstevel@tonic-gate } 2614*7c478bd9Sstevel@tonic-gate #endif /* __amd64 */ 2615*7c478bd9Sstevel@tonic-gate 2616*7c478bd9Sstevel@tonic-gate /* 2617*7c478bd9Sstevel@tonic-gate * int hat_probe(hat, addr) 2618*7c478bd9Sstevel@tonic-gate * return 0 if no valid mapping is present. Faster version 2619*7c478bd9Sstevel@tonic-gate * of hat_getattr in certain architectures. 2620*7c478bd9Sstevel@tonic-gate */ 2621*7c478bd9Sstevel@tonic-gate int 2622*7c478bd9Sstevel@tonic-gate hat_probe(hat_t *hat, caddr_t addr) 2623*7c478bd9Sstevel@tonic-gate { 2624*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = ALIGN2PAGE(addr); 2625*7c478bd9Sstevel@tonic-gate uint_t entry; 2626*7c478bd9Sstevel@tonic-gate htable_t *ht; 2627*7c478bd9Sstevel@tonic-gate pgcnt_t pg_off; 2628*7c478bd9Sstevel@tonic-gate 2629*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || vaddr < kernelbase); 2630*7c478bd9Sstevel@tonic-gate ASSERT(hat == kas.a_hat || 2631*7c478bd9Sstevel@tonic-gate AS_LOCK_HELD(hat->hat_as, &hat->hat_as->a_lock)); 2632*7c478bd9Sstevel@tonic-gate if (IN_VA_HOLE(vaddr)) 2633*7c478bd9Sstevel@tonic-gate return (0); 2634*7c478bd9Sstevel@tonic-gate 2635*7c478bd9Sstevel@tonic-gate /* 2636*7c478bd9Sstevel@tonic-gate * Most common use of hat_probe is from segmap. We special case it 2637*7c478bd9Sstevel@tonic-gate * for performance. 2638*7c478bd9Sstevel@tonic-gate */ 2639*7c478bd9Sstevel@tonic-gate if (mmu.kmap_addr <= vaddr && vaddr < mmu.kmap_eaddr) { 2640*7c478bd9Sstevel@tonic-gate pg_off = mmu_btop(vaddr - mmu.kmap_addr); 2641*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 2642*7c478bd9Sstevel@tonic-gate return (PTE_ISVALID(mmu.kmap_ptes[pg_off])); 2643*7c478bd9Sstevel@tonic-gate else 2644*7c478bd9Sstevel@tonic-gate return (PTE_ISVALID( 2645*7c478bd9Sstevel@tonic-gate ((x86pte32_t *)mmu.kmap_ptes)[pg_off])); 2646*7c478bd9Sstevel@tonic-gate } 2647*7c478bd9Sstevel@tonic-gate 2648*7c478bd9Sstevel@tonic-gate ht = htable_getpage(hat, vaddr, &entry); 2649*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2650*7c478bd9Sstevel@tonic-gate return (0); 2651*7c478bd9Sstevel@tonic-gate htable_release(ht); 2652*7c478bd9Sstevel@tonic-gate return (1); 2653*7c478bd9Sstevel@tonic-gate } 2654*7c478bd9Sstevel@tonic-gate 2655*7c478bd9Sstevel@tonic-gate /* 2656*7c478bd9Sstevel@tonic-gate * Simple implementation of ISM. hat_share() is just like hat_memload_array(), 2657*7c478bd9Sstevel@tonic-gate * except that we use the ism_hat's existing mappings to determine the pages 2658*7c478bd9Sstevel@tonic-gate * and protections to use for this hat. In case we find a properly aligned 2659*7c478bd9Sstevel@tonic-gate * and sized pagetable of 4K mappings, we will attempt to share the pagetable 2660*7c478bd9Sstevel@tonic-gate * itself. 2661*7c478bd9Sstevel@tonic-gate */ 2662*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2663*7c478bd9Sstevel@tonic-gate int 2664*7c478bd9Sstevel@tonic-gate hat_share( 2665*7c478bd9Sstevel@tonic-gate hat_t *hat, 2666*7c478bd9Sstevel@tonic-gate caddr_t addr, 2667*7c478bd9Sstevel@tonic-gate hat_t *ism_hat, 2668*7c478bd9Sstevel@tonic-gate caddr_t src_addr, 2669*7c478bd9Sstevel@tonic-gate size_t len, /* almost useless value, see below.. */ 2670*7c478bd9Sstevel@tonic-gate uint_t ismszc) 2671*7c478bd9Sstevel@tonic-gate { 2672*7c478bd9Sstevel@tonic-gate uintptr_t vaddr_start = (uintptr_t)addr; 2673*7c478bd9Sstevel@tonic-gate uintptr_t vaddr; 2674*7c478bd9Sstevel@tonic-gate uintptr_t pt_vaddr; 2675*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr_start + len; 2676*7c478bd9Sstevel@tonic-gate uintptr_t ism_addr_start = (uintptr_t)src_addr; 2677*7c478bd9Sstevel@tonic-gate uintptr_t ism_addr = ism_addr_start; 2678*7c478bd9Sstevel@tonic-gate uintptr_t e_ism_addr = ism_addr + len; 2679*7c478bd9Sstevel@tonic-gate htable_t *ism_ht = NULL; 2680*7c478bd9Sstevel@tonic-gate htable_t *ht; 2681*7c478bd9Sstevel@tonic-gate x86pte_t pte; 2682*7c478bd9Sstevel@tonic-gate page_t *pp; 2683*7c478bd9Sstevel@tonic-gate pfn_t pfn; 2684*7c478bd9Sstevel@tonic-gate level_t l; 2685*7c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 2686*7c478bd9Sstevel@tonic-gate uint_t prot; 2687*7c478bd9Sstevel@tonic-gate uint_t valid_cnt; 2688*7c478bd9Sstevel@tonic-gate 2689*7c478bd9Sstevel@tonic-gate /* 2690*7c478bd9Sstevel@tonic-gate * We might be asked to share an empty DISM hat by as_dup() 2691*7c478bd9Sstevel@tonic-gate */ 2692*7c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2693*7c478bd9Sstevel@tonic-gate ASSERT(eaddr <= kernelbase); 2694*7c478bd9Sstevel@tonic-gate if (!(ism_hat->hat_flags & HAT_SHARED)) { 2695*7c478bd9Sstevel@tonic-gate ASSERT(hat_get_mapped_size(ism_hat) == 0); 2696*7c478bd9Sstevel@tonic-gate return (0); 2697*7c478bd9Sstevel@tonic-gate } 2698*7c478bd9Sstevel@tonic-gate 2699*7c478bd9Sstevel@tonic-gate /* 2700*7c478bd9Sstevel@tonic-gate * The SPT segment driver often passes us a size larger than there are 2701*7c478bd9Sstevel@tonic-gate * valid mappings. That's because it rounds the segment size up to a 2702*7c478bd9Sstevel@tonic-gate * large pagesize, even if the actual memory mapped by ism_hat is less. 2703*7c478bd9Sstevel@tonic-gate */ 2704*7c478bd9Sstevel@tonic-gate HATIN(hat_share, hat, addr, len); 2705*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr_start)); 2706*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(ism_addr_start)); 2707*7c478bd9Sstevel@tonic-gate ASSERT(ism_hat->hat_flags & HAT_SHARED); 2708*7c478bd9Sstevel@tonic-gate while (ism_addr < e_ism_addr) { 2709*7c478bd9Sstevel@tonic-gate /* 2710*7c478bd9Sstevel@tonic-gate * use htable_walk to get the next valid ISM mapping 2711*7c478bd9Sstevel@tonic-gate */ 2712*7c478bd9Sstevel@tonic-gate pte = htable_walk(ism_hat, &ism_ht, &ism_addr, e_ism_addr); 2713*7c478bd9Sstevel@tonic-gate if (ism_ht == NULL) 2714*7c478bd9Sstevel@tonic-gate break; 2715*7c478bd9Sstevel@tonic-gate 2716*7c478bd9Sstevel@tonic-gate /* 2717*7c478bd9Sstevel@tonic-gate * Find the largest page size we can use, based on the 2718*7c478bd9Sstevel@tonic-gate * ISM mapping size, our address alignment and the remaining 2719*7c478bd9Sstevel@tonic-gate * map length. 2720*7c478bd9Sstevel@tonic-gate */ 2721*7c478bd9Sstevel@tonic-gate vaddr = vaddr_start + (ism_addr - ism_addr_start); 2722*7c478bd9Sstevel@tonic-gate for (l = ism_ht->ht_level; l > 0; --l) { 2723*7c478bd9Sstevel@tonic-gate if (LEVEL_SIZE(l) <= eaddr - vaddr && 2724*7c478bd9Sstevel@tonic-gate (vaddr & LEVEL_OFFSET(l)) == 0) 2725*7c478bd9Sstevel@tonic-gate break; 2726*7c478bd9Sstevel@tonic-gate } 2727*7c478bd9Sstevel@tonic-gate 2728*7c478bd9Sstevel@tonic-gate /* 2729*7c478bd9Sstevel@tonic-gate * attempt to share the pagetable 2730*7c478bd9Sstevel@tonic-gate * 2731*7c478bd9Sstevel@tonic-gate * - only 4K pagetables are shared (ie. level == 0) 2732*7c478bd9Sstevel@tonic-gate * - the hat_share() length must cover the whole pagetable 2733*7c478bd9Sstevel@tonic-gate * - the shared address must align at level 1 2734*7c478bd9Sstevel@tonic-gate * - a shared PTE for this address already exists OR 2735*7c478bd9Sstevel@tonic-gate * - no page table for this address exists yet 2736*7c478bd9Sstevel@tonic-gate */ 2737*7c478bd9Sstevel@tonic-gate pt_vaddr = 2738*7c478bd9Sstevel@tonic-gate vaddr_start + (ism_ht->ht_vaddr - ism_addr_start); 2739*7c478bd9Sstevel@tonic-gate if (ism_ht->ht_level == 0 && 2740*7c478bd9Sstevel@tonic-gate ism_ht->ht_vaddr + LEVEL_SIZE(1) <= e_ism_addr && 2741*7c478bd9Sstevel@tonic-gate (pt_vaddr & LEVEL_OFFSET(1)) == 0) { 2742*7c478bd9Sstevel@tonic-gate 2743*7c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, pt_vaddr, 0); 2744*7c478bd9Sstevel@tonic-gate if (ht == NULL) 2745*7c478bd9Sstevel@tonic-gate ht = htable_create(hat, pt_vaddr, 0, ism_ht); 2746*7c478bd9Sstevel@tonic-gate 2747*7c478bd9Sstevel@tonic-gate if (ht->ht_level > 0 || 2748*7c478bd9Sstevel@tonic-gate !(ht->ht_flags & HTABLE_SHARED_PFN)) { 2749*7c478bd9Sstevel@tonic-gate 2750*7c478bd9Sstevel@tonic-gate htable_release(ht); 2751*7c478bd9Sstevel@tonic-gate 2752*7c478bd9Sstevel@tonic-gate } else { 2753*7c478bd9Sstevel@tonic-gate 2754*7c478bd9Sstevel@tonic-gate /* 2755*7c478bd9Sstevel@tonic-gate * share the page table 2756*7c478bd9Sstevel@tonic-gate */ 2757*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 2758*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_shares == ism_ht); 2759*7c478bd9Sstevel@tonic-gate valid_cnt = ism_ht->ht_valid_cnt; 2760*7c478bd9Sstevel@tonic-gate atomic_add_long(&hat->hat_pages_mapped[0], 2761*7c478bd9Sstevel@tonic-gate valid_cnt - ht->ht_valid_cnt); 2762*7c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = valid_cnt; 2763*7c478bd9Sstevel@tonic-gate htable_release(ht); 2764*7c478bd9Sstevel@tonic-gate ism_addr = ism_ht->ht_vaddr + LEVEL_SIZE(1); 2765*7c478bd9Sstevel@tonic-gate htable_release(ism_ht); 2766*7c478bd9Sstevel@tonic-gate ism_ht = NULL; 2767*7c478bd9Sstevel@tonic-gate continue; 2768*7c478bd9Sstevel@tonic-gate } 2769*7c478bd9Sstevel@tonic-gate } 2770*7c478bd9Sstevel@tonic-gate 2771*7c478bd9Sstevel@tonic-gate /* 2772*7c478bd9Sstevel@tonic-gate * Unable to share the page table. Instead we will 2773*7c478bd9Sstevel@tonic-gate * create new mappings from the values in the ISM mappings. 2774*7c478bd9Sstevel@tonic-gate * 2775*7c478bd9Sstevel@tonic-gate * The ISM mapping might be larger than the share area, 2776*7c478bd9Sstevel@tonic-gate * be careful to trunctate it if needed. 2777*7c478bd9Sstevel@tonic-gate */ 2778*7c478bd9Sstevel@tonic-gate if (eaddr - vaddr >= LEVEL_SIZE(ism_ht->ht_level)) { 2779*7c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(LEVEL_SIZE(ism_ht->ht_level)); 2780*7c478bd9Sstevel@tonic-gate } else { 2781*7c478bd9Sstevel@tonic-gate pgcnt = mmu_btop(eaddr - vaddr); 2782*7c478bd9Sstevel@tonic-gate l = 0; 2783*7c478bd9Sstevel@tonic-gate } 2784*7c478bd9Sstevel@tonic-gate 2785*7c478bd9Sstevel@tonic-gate pfn = PTE2PFN(pte, ism_ht->ht_level); 2786*7c478bd9Sstevel@tonic-gate ASSERT(pfn != PFN_INVALID); 2787*7c478bd9Sstevel@tonic-gate while (pgcnt > 0) { 2788*7c478bd9Sstevel@tonic-gate /* 2789*7c478bd9Sstevel@tonic-gate * Make a new pte for the PFN for this level. 2790*7c478bd9Sstevel@tonic-gate * Copy protections for the pte from the ISM pte. 2791*7c478bd9Sstevel@tonic-gate */ 2792*7c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 2793*7c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 2794*7c478bd9Sstevel@tonic-gate 2795*7c478bd9Sstevel@tonic-gate prot = PROT_USER | PROT_READ | HAT_UNORDERED_OK; 2796*7c478bd9Sstevel@tonic-gate if (PTE_GET(pte, PT_WRITABLE)) 2797*7c478bd9Sstevel@tonic-gate prot |= PROT_WRITE; 2798*7c478bd9Sstevel@tonic-gate if (!PTE_GET(pte, PT_NX)) 2799*7c478bd9Sstevel@tonic-gate prot |= PROT_EXEC; 2800*7c478bd9Sstevel@tonic-gate 2801*7c478bd9Sstevel@tonic-gate /* 2802*7c478bd9Sstevel@tonic-gate * XX64 -- can shm ever be written to swap? 2803*7c478bd9Sstevel@tonic-gate * if not we could use HAT_NOSYNC here. 2804*7c478bd9Sstevel@tonic-gate */ 2805*7c478bd9Sstevel@tonic-gate hati_load_common(hat, vaddr, pp, prot, 2806*7c478bd9Sstevel@tonic-gate HAT_LOAD, l, pfn); 2807*7c478bd9Sstevel@tonic-gate 2808*7c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(l); 2809*7c478bd9Sstevel@tonic-gate ism_addr += LEVEL_SIZE(l); 2810*7c478bd9Sstevel@tonic-gate pfn += mmu_btop(LEVEL_SIZE(l)); 2811*7c478bd9Sstevel@tonic-gate pgcnt -= mmu_btop(LEVEL_SIZE(l)); 2812*7c478bd9Sstevel@tonic-gate } 2813*7c478bd9Sstevel@tonic-gate } 2814*7c478bd9Sstevel@tonic-gate if (ism_ht != NULL) 2815*7c478bd9Sstevel@tonic-gate htable_release(ism_ht); 2816*7c478bd9Sstevel@tonic-gate 2817*7c478bd9Sstevel@tonic-gate HATOUT(hat_share, hat, addr); 2818*7c478bd9Sstevel@tonic-gate return (0); 2819*7c478bd9Sstevel@tonic-gate } 2820*7c478bd9Sstevel@tonic-gate 2821*7c478bd9Sstevel@tonic-gate 2822*7c478bd9Sstevel@tonic-gate /* 2823*7c478bd9Sstevel@tonic-gate * hat_unshare() is similar to hat_unload_callback(), but 2824*7c478bd9Sstevel@tonic-gate * we have to look for empty shared pagetables. Note that 2825*7c478bd9Sstevel@tonic-gate * hat_unshare() is always invoked against an entire segment. 2826*7c478bd9Sstevel@tonic-gate */ 2827*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2828*7c478bd9Sstevel@tonic-gate void 2829*7c478bd9Sstevel@tonic-gate hat_unshare(hat_t *hat, caddr_t addr, size_t len, uint_t ismszc) 2830*7c478bd9Sstevel@tonic-gate { 2831*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = (uintptr_t)addr; 2832*7c478bd9Sstevel@tonic-gate uintptr_t eaddr = vaddr + len; 2833*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 2834*7c478bd9Sstevel@tonic-gate uint_t need_demaps = 0; 2835*7c478bd9Sstevel@tonic-gate 2836*7c478bd9Sstevel@tonic-gate ASSERT(hat != kas.a_hat); 2837*7c478bd9Sstevel@tonic-gate ASSERT(eaddr <= kernelbase); 2838*7c478bd9Sstevel@tonic-gate HATIN(hat_unshare, hat, addr, len); 2839*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(vaddr)); 2840*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(eaddr)); 2841*7c478bd9Sstevel@tonic-gate 2842*7c478bd9Sstevel@tonic-gate /* 2843*7c478bd9Sstevel@tonic-gate * First go through and remove any shared pagetables. 2844*7c478bd9Sstevel@tonic-gate * 2845*7c478bd9Sstevel@tonic-gate * Note that it's ok to delay the demap until the entire range is 2846*7c478bd9Sstevel@tonic-gate * finished, because if hat_pageunload() were to unload a shared 2847*7c478bd9Sstevel@tonic-gate * pagetable page, its hat_demap() will do a global user TLB invalidate. 2848*7c478bd9Sstevel@tonic-gate */ 2849*7c478bd9Sstevel@tonic-gate while (vaddr < eaddr) { 2850*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(vaddr)); 2851*7c478bd9Sstevel@tonic-gate /* 2852*7c478bd9Sstevel@tonic-gate * find the pagetable that would map the current address 2853*7c478bd9Sstevel@tonic-gate */ 2854*7c478bd9Sstevel@tonic-gate ht = htable_lookup(hat, vaddr, 0); 2855*7c478bd9Sstevel@tonic-gate if (ht != NULL) { 2856*7c478bd9Sstevel@tonic-gate if (ht->ht_flags & HTABLE_SHARED_PFN) { 2857*7c478bd9Sstevel@tonic-gate /* 2858*7c478bd9Sstevel@tonic-gate * clear mapped pages count, set valid_cnt to 0 2859*7c478bd9Sstevel@tonic-gate * and let htable_release() finish the job 2860*7c478bd9Sstevel@tonic-gate */ 2861*7c478bd9Sstevel@tonic-gate atomic_add_long(&hat->hat_pages_mapped[0], 2862*7c478bd9Sstevel@tonic-gate -ht->ht_valid_cnt); 2863*7c478bd9Sstevel@tonic-gate ht->ht_valid_cnt = 0; 2864*7c478bd9Sstevel@tonic-gate need_demaps = 1; 2865*7c478bd9Sstevel@tonic-gate } 2866*7c478bd9Sstevel@tonic-gate htable_release(ht); 2867*7c478bd9Sstevel@tonic-gate } 2868*7c478bd9Sstevel@tonic-gate vaddr = (vaddr & LEVEL_MASK(1)) + LEVEL_SIZE(1); 2869*7c478bd9Sstevel@tonic-gate } 2870*7c478bd9Sstevel@tonic-gate 2871*7c478bd9Sstevel@tonic-gate /* 2872*7c478bd9Sstevel@tonic-gate * flush the TLBs - since we're probably dealing with MANY mappings 2873*7c478bd9Sstevel@tonic-gate * we do just one CR3 reload. 2874*7c478bd9Sstevel@tonic-gate */ 2875*7c478bd9Sstevel@tonic-gate if (!(hat->hat_flags & HAT_FREEING) && need_demaps) 2876*7c478bd9Sstevel@tonic-gate hat_demap(hat, DEMAP_ALL_ADDR); 2877*7c478bd9Sstevel@tonic-gate 2878*7c478bd9Sstevel@tonic-gate /* 2879*7c478bd9Sstevel@tonic-gate * Now go back and clean up any unaligned mappings that 2880*7c478bd9Sstevel@tonic-gate * couldn't share pagetables. 2881*7c478bd9Sstevel@tonic-gate */ 2882*7c478bd9Sstevel@tonic-gate hat_unload(hat, addr, len, HAT_UNLOAD_UNMAP); 2883*7c478bd9Sstevel@tonic-gate 2884*7c478bd9Sstevel@tonic-gate HATOUT(hat_unshare, hat, addr); 2885*7c478bd9Sstevel@tonic-gate } 2886*7c478bd9Sstevel@tonic-gate 2887*7c478bd9Sstevel@tonic-gate 2888*7c478bd9Sstevel@tonic-gate /* 2889*7c478bd9Sstevel@tonic-gate * hat_reserve() does nothing 2890*7c478bd9Sstevel@tonic-gate */ 2891*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 2892*7c478bd9Sstevel@tonic-gate void 2893*7c478bd9Sstevel@tonic-gate hat_reserve(struct as *as, caddr_t addr, size_t len) 2894*7c478bd9Sstevel@tonic-gate { 2895*7c478bd9Sstevel@tonic-gate } 2896*7c478bd9Sstevel@tonic-gate 2897*7c478bd9Sstevel@tonic-gate 2898*7c478bd9Sstevel@tonic-gate /* 2899*7c478bd9Sstevel@tonic-gate * Called when all mappings to a page should have write permission removed. 2900*7c478bd9Sstevel@tonic-gate * Mostly stolem from hat_pagesync() 2901*7c478bd9Sstevel@tonic-gate */ 2902*7c478bd9Sstevel@tonic-gate static void 2903*7c478bd9Sstevel@tonic-gate hati_page_clrwrt(struct page *pp) 2904*7c478bd9Sstevel@tonic-gate { 2905*7c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 2906*7c478bd9Sstevel@tonic-gate htable_t *ht; 2907*7c478bd9Sstevel@tonic-gate uint_t entry; 2908*7c478bd9Sstevel@tonic-gate x86pte_t old; 2909*7c478bd9Sstevel@tonic-gate x86pte_t new; 2910*7c478bd9Sstevel@tonic-gate uint_t pszc = 0; 2911*7c478bd9Sstevel@tonic-gate 2912*7c478bd9Sstevel@tonic-gate next_size: 2913*7c478bd9Sstevel@tonic-gate /* 2914*7c478bd9Sstevel@tonic-gate * walk thru the mapping list clearing write permission 2915*7c478bd9Sstevel@tonic-gate */ 2916*7c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 2917*7c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 2918*7c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 2919*7c478bd9Sstevel@tonic-gate continue; 2920*7c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 2921*7c478bd9Sstevel@tonic-gate 2922*7c478bd9Sstevel@tonic-gate for (;;) { 2923*7c478bd9Sstevel@tonic-gate /* 2924*7c478bd9Sstevel@tonic-gate * Is this mapping of interest? 2925*7c478bd9Sstevel@tonic-gate */ 2926*7c478bd9Sstevel@tonic-gate if (PTE2PFN(old, ht->ht_level) != pp->p_pagenum || 2927*7c478bd9Sstevel@tonic-gate PTE_GET(old, PT_WRITABLE) == 0) 2928*7c478bd9Sstevel@tonic-gate break; 2929*7c478bd9Sstevel@tonic-gate 2930*7c478bd9Sstevel@tonic-gate /* 2931*7c478bd9Sstevel@tonic-gate * Clear ref/mod writable bits. This requires cross 2932*7c478bd9Sstevel@tonic-gate * calls to ensure any executing TLBs see cleared bits. 2933*7c478bd9Sstevel@tonic-gate */ 2934*7c478bd9Sstevel@tonic-gate new = old; 2935*7c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD | PT_WRITABLE); 2936*7c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 2937*7c478bd9Sstevel@tonic-gate if (old != 0) 2938*7c478bd9Sstevel@tonic-gate continue; 2939*7c478bd9Sstevel@tonic-gate 2940*7c478bd9Sstevel@tonic-gate break; 2941*7c478bd9Sstevel@tonic-gate } 2942*7c478bd9Sstevel@tonic-gate } 2943*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 2944*7c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 2945*7c478bd9Sstevel@tonic-gate page_t *tpp; 2946*7c478bd9Sstevel@tonic-gate pszc++; 2947*7c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 2948*7c478bd9Sstevel@tonic-gate if (pp != tpp) { 2949*7c478bd9Sstevel@tonic-gate pp = tpp; 2950*7c478bd9Sstevel@tonic-gate goto next_size; 2951*7c478bd9Sstevel@tonic-gate } 2952*7c478bd9Sstevel@tonic-gate } 2953*7c478bd9Sstevel@tonic-gate } 2954*7c478bd9Sstevel@tonic-gate 2955*7c478bd9Sstevel@tonic-gate /* 2956*7c478bd9Sstevel@tonic-gate * void hat_page_setattr(pp, flag) 2957*7c478bd9Sstevel@tonic-gate * void hat_page_clrattr(pp, flag) 2958*7c478bd9Sstevel@tonic-gate * used to set/clr ref/mod bits. 2959*7c478bd9Sstevel@tonic-gate */ 2960*7c478bd9Sstevel@tonic-gate void 2961*7c478bd9Sstevel@tonic-gate hat_page_setattr(struct page *pp, uint_t flag) 2962*7c478bd9Sstevel@tonic-gate { 2963*7c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 2964*7c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 2965*7c478bd9Sstevel@tonic-gate page_t **listp; 2966*7c478bd9Sstevel@tonic-gate 2967*7c478bd9Sstevel@tonic-gate if (PP_GETRM(pp, flag) == flag) 2968*7c478bd9Sstevel@tonic-gate return; 2969*7c478bd9Sstevel@tonic-gate 2970*7c478bd9Sstevel@tonic-gate if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 2971*7c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 2972*7c478bd9Sstevel@tonic-gate mutex_enter(vphm); 2973*7c478bd9Sstevel@tonic-gate } 2974*7c478bd9Sstevel@tonic-gate 2975*7c478bd9Sstevel@tonic-gate PP_SETRM(pp, flag); 2976*7c478bd9Sstevel@tonic-gate 2977*7c478bd9Sstevel@tonic-gate if (vphm != NULL) { 2978*7c478bd9Sstevel@tonic-gate 2979*7c478bd9Sstevel@tonic-gate /* 2980*7c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 2981*7c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 2982*7c478bd9Sstevel@tonic-gate * pp is the only page on the list. 2983*7c478bd9Sstevel@tonic-gate */ 2984*7c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 2985*7c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 2986*7c478bd9Sstevel@tonic-gate if (vp->v_pages != NULL) 2987*7c478bd9Sstevel@tonic-gate listp = &vp->v_pages->p_vpprev->p_vpnext; 2988*7c478bd9Sstevel@tonic-gate else 2989*7c478bd9Sstevel@tonic-gate listp = &vp->v_pages; 2990*7c478bd9Sstevel@tonic-gate page_vpadd(listp, pp); 2991*7c478bd9Sstevel@tonic-gate } 2992*7c478bd9Sstevel@tonic-gate mutex_exit(vphm); 2993*7c478bd9Sstevel@tonic-gate } 2994*7c478bd9Sstevel@tonic-gate } 2995*7c478bd9Sstevel@tonic-gate 2996*7c478bd9Sstevel@tonic-gate void 2997*7c478bd9Sstevel@tonic-gate hat_page_clrattr(struct page *pp, uint_t flag) 2998*7c478bd9Sstevel@tonic-gate { 2999*7c478bd9Sstevel@tonic-gate vnode_t *vp = pp->p_vnode; 3000*7c478bd9Sstevel@tonic-gate kmutex_t *vphm = NULL; 3001*7c478bd9Sstevel@tonic-gate ASSERT(!(flag & ~(P_MOD | P_REF | P_RO))); 3002*7c478bd9Sstevel@tonic-gate 3003*7c478bd9Sstevel@tonic-gate /* 3004*7c478bd9Sstevel@tonic-gate * for vnode with a sorted v_pages list, we need to change 3005*7c478bd9Sstevel@tonic-gate * the attributes and the v_pages list together under page_vnode_mutex. 3006*7c478bd9Sstevel@tonic-gate */ 3007*7c478bd9Sstevel@tonic-gate if ((flag & P_MOD) != 0 && vp != NULL && IS_VMODSORT(vp)) { 3008*7c478bd9Sstevel@tonic-gate vphm = page_vnode_mutex(vp); 3009*7c478bd9Sstevel@tonic-gate mutex_enter(vphm); 3010*7c478bd9Sstevel@tonic-gate } 3011*7c478bd9Sstevel@tonic-gate 3012*7c478bd9Sstevel@tonic-gate PP_CLRRM(pp, flag); 3013*7c478bd9Sstevel@tonic-gate 3014*7c478bd9Sstevel@tonic-gate if (vphm != NULL) { 3015*7c478bd9Sstevel@tonic-gate 3016*7c478bd9Sstevel@tonic-gate /* 3017*7c478bd9Sstevel@tonic-gate * Some File Systems examine v_pages for NULL w/o 3018*7c478bd9Sstevel@tonic-gate * grabbing the vphm mutex. Must not let it become NULL when 3019*7c478bd9Sstevel@tonic-gate * pp is the only page on the list. 3020*7c478bd9Sstevel@tonic-gate */ 3021*7c478bd9Sstevel@tonic-gate if (pp->p_vpnext != pp) { 3022*7c478bd9Sstevel@tonic-gate page_vpsub(&vp->v_pages, pp); 3023*7c478bd9Sstevel@tonic-gate page_vpadd(&vp->v_pages, pp); 3024*7c478bd9Sstevel@tonic-gate } 3025*7c478bd9Sstevel@tonic-gate mutex_exit(vphm); 3026*7c478bd9Sstevel@tonic-gate 3027*7c478bd9Sstevel@tonic-gate /* 3028*7c478bd9Sstevel@tonic-gate * VMODSORT works by removing write permissions and getting 3029*7c478bd9Sstevel@tonic-gate * a fault when a page is made dirty. At this point 3030*7c478bd9Sstevel@tonic-gate * we need to remove write permission from all mappings 3031*7c478bd9Sstevel@tonic-gate * to this page. 3032*7c478bd9Sstevel@tonic-gate */ 3033*7c478bd9Sstevel@tonic-gate hati_page_clrwrt(pp); 3034*7c478bd9Sstevel@tonic-gate } 3035*7c478bd9Sstevel@tonic-gate } 3036*7c478bd9Sstevel@tonic-gate 3037*7c478bd9Sstevel@tonic-gate /* 3038*7c478bd9Sstevel@tonic-gate * If flag is specified, returns 0 if attribute is disabled 3039*7c478bd9Sstevel@tonic-gate * and non zero if enabled. If flag specifes multiple attributs 3040*7c478bd9Sstevel@tonic-gate * then returns 0 if ALL atriibutes are disabled. This is an advisory 3041*7c478bd9Sstevel@tonic-gate * call. 3042*7c478bd9Sstevel@tonic-gate */ 3043*7c478bd9Sstevel@tonic-gate uint_t 3044*7c478bd9Sstevel@tonic-gate hat_page_getattr(struct page *pp, uint_t flag) 3045*7c478bd9Sstevel@tonic-gate { 3046*7c478bd9Sstevel@tonic-gate return (PP_GETRM(pp, flag)); 3047*7c478bd9Sstevel@tonic-gate } 3048*7c478bd9Sstevel@tonic-gate 3049*7c478bd9Sstevel@tonic-gate 3050*7c478bd9Sstevel@tonic-gate /* 3051*7c478bd9Sstevel@tonic-gate * common code used by hat_pageunload() and hment_steal() 3052*7c478bd9Sstevel@tonic-gate */ 3053*7c478bd9Sstevel@tonic-gate hment_t * 3054*7c478bd9Sstevel@tonic-gate hati_page_unmap(page_t *pp, htable_t *ht, uint_t entry) 3055*7c478bd9Sstevel@tonic-gate { 3056*7c478bd9Sstevel@tonic-gate x86pte_t old_pte; 3057*7c478bd9Sstevel@tonic-gate pfn_t pfn = pp->p_pagenum; 3058*7c478bd9Sstevel@tonic-gate hment_t *hm; 3059*7c478bd9Sstevel@tonic-gate 3060*7c478bd9Sstevel@tonic-gate /* 3061*7c478bd9Sstevel@tonic-gate * We need to acquire a hold on the htable in order to 3062*7c478bd9Sstevel@tonic-gate * do the invalidate. We know the htable must exist, since 3063*7c478bd9Sstevel@tonic-gate * unmap's don't release the htable until after removing any 3064*7c478bd9Sstevel@tonic-gate * hment. Having x86_hm_enter() keeps that from proceeding. 3065*7c478bd9Sstevel@tonic-gate */ 3066*7c478bd9Sstevel@tonic-gate htable_acquire(ht); 3067*7c478bd9Sstevel@tonic-gate 3068*7c478bd9Sstevel@tonic-gate /* 3069*7c478bd9Sstevel@tonic-gate * Invalidate the PTE and remove the hment. 3070*7c478bd9Sstevel@tonic-gate */ 3071*7c478bd9Sstevel@tonic-gate old_pte = x86pte_invalidate_pfn(ht, entry, pfn, NULL); 3072*7c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old_pte, ht->ht_level) == pfn); 3073*7c478bd9Sstevel@tonic-gate 3074*7c478bd9Sstevel@tonic-gate /* 3075*7c478bd9Sstevel@tonic-gate * Clean up all the htable information for this mapping 3076*7c478bd9Sstevel@tonic-gate */ 3077*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3078*7c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 3079*7c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 3080*7c478bd9Sstevel@tonic-gate 3081*7c478bd9Sstevel@tonic-gate /* 3082*7c478bd9Sstevel@tonic-gate * sync ref/mod bits to the page_t 3083*7c478bd9Sstevel@tonic-gate */ 3084*7c478bd9Sstevel@tonic-gate if (PTE_GET(old_pte, PT_NOSYNC) == 0) 3085*7c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old_pte, ht->ht_level); 3086*7c478bd9Sstevel@tonic-gate 3087*7c478bd9Sstevel@tonic-gate /* 3088*7c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 3089*7c478bd9Sstevel@tonic-gate */ 3090*7c478bd9Sstevel@tonic-gate hm = hment_remove(pp, ht, entry); 3091*7c478bd9Sstevel@tonic-gate 3092*7c478bd9Sstevel@tonic-gate /* 3093*7c478bd9Sstevel@tonic-gate * drop the mapping list lock so that we might free the 3094*7c478bd9Sstevel@tonic-gate * hment and htable. 3095*7c478bd9Sstevel@tonic-gate */ 3096*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3097*7c478bd9Sstevel@tonic-gate htable_release(ht); 3098*7c478bd9Sstevel@tonic-gate return (hm); 3099*7c478bd9Sstevel@tonic-gate } 3100*7c478bd9Sstevel@tonic-gate 3101*7c478bd9Sstevel@tonic-gate /* 3102*7c478bd9Sstevel@tonic-gate * Unload all translations to a page. If the page is a subpage of a large 3103*7c478bd9Sstevel@tonic-gate * page, the large page mappings are also removed. 3104*7c478bd9Sstevel@tonic-gate * 3105*7c478bd9Sstevel@tonic-gate * The forceflags are unused. 3106*7c478bd9Sstevel@tonic-gate */ 3107*7c478bd9Sstevel@tonic-gate 3108*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3109*7c478bd9Sstevel@tonic-gate static int 3110*7c478bd9Sstevel@tonic-gate hati_pageunload(struct page *pp, uint_t pg_szcd, uint_t forceflag) 3111*7c478bd9Sstevel@tonic-gate { 3112*7c478bd9Sstevel@tonic-gate page_t *cur_pp = pp; 3113*7c478bd9Sstevel@tonic-gate hment_t *hm; 3114*7c478bd9Sstevel@tonic-gate hment_t *prev; 3115*7c478bd9Sstevel@tonic-gate htable_t *ht; 3116*7c478bd9Sstevel@tonic-gate uint_t entry; 3117*7c478bd9Sstevel@tonic-gate level_t level; 3118*7c478bd9Sstevel@tonic-gate 3119*7c478bd9Sstevel@tonic-gate /* 3120*7c478bd9Sstevel@tonic-gate * The loop with next_size handles pages with multiple pagesize mappings 3121*7c478bd9Sstevel@tonic-gate */ 3122*7c478bd9Sstevel@tonic-gate next_size: 3123*7c478bd9Sstevel@tonic-gate for (;;) { 3124*7c478bd9Sstevel@tonic-gate 3125*7c478bd9Sstevel@tonic-gate /* 3126*7c478bd9Sstevel@tonic-gate * Get a mapping list entry 3127*7c478bd9Sstevel@tonic-gate */ 3128*7c478bd9Sstevel@tonic-gate x86_hm_enter(cur_pp); 3129*7c478bd9Sstevel@tonic-gate for (prev = NULL; ; prev = hm) { 3130*7c478bd9Sstevel@tonic-gate hm = hment_walk(cur_pp, &ht, &entry, prev); 3131*7c478bd9Sstevel@tonic-gate if (hm == NULL) { 3132*7c478bd9Sstevel@tonic-gate x86_hm_exit(cur_pp); 3133*7c478bd9Sstevel@tonic-gate 3134*7c478bd9Sstevel@tonic-gate /* 3135*7c478bd9Sstevel@tonic-gate * If not part of a larger page, we're done. 3136*7c478bd9Sstevel@tonic-gate */ 3137*7c478bd9Sstevel@tonic-gate if (cur_pp->p_szc <= pg_szcd) 3138*7c478bd9Sstevel@tonic-gate return (0); 3139*7c478bd9Sstevel@tonic-gate 3140*7c478bd9Sstevel@tonic-gate /* 3141*7c478bd9Sstevel@tonic-gate * Else check the next larger page size. 3142*7c478bd9Sstevel@tonic-gate * hat_page_demote() may decrease p_szc 3143*7c478bd9Sstevel@tonic-gate * but that's ok we'll just take an extra 3144*7c478bd9Sstevel@tonic-gate * trip discover there're no larger mappings 3145*7c478bd9Sstevel@tonic-gate * and return. 3146*7c478bd9Sstevel@tonic-gate */ 3147*7c478bd9Sstevel@tonic-gate ++pg_szcd; 3148*7c478bd9Sstevel@tonic-gate cur_pp = PP_GROUPLEADER(cur_pp, pg_szcd); 3149*7c478bd9Sstevel@tonic-gate goto next_size; 3150*7c478bd9Sstevel@tonic-gate } 3151*7c478bd9Sstevel@tonic-gate 3152*7c478bd9Sstevel@tonic-gate /* 3153*7c478bd9Sstevel@tonic-gate * If this mapping size matches, remove it. 3154*7c478bd9Sstevel@tonic-gate */ 3155*7c478bd9Sstevel@tonic-gate level = ht->ht_level; 3156*7c478bd9Sstevel@tonic-gate if (level == pg_szcd) 3157*7c478bd9Sstevel@tonic-gate break; 3158*7c478bd9Sstevel@tonic-gate } 3159*7c478bd9Sstevel@tonic-gate 3160*7c478bd9Sstevel@tonic-gate /* 3161*7c478bd9Sstevel@tonic-gate * Remove the mapping list entry for this page. 3162*7c478bd9Sstevel@tonic-gate * Note this does the x86_hm_exit() for us. 3163*7c478bd9Sstevel@tonic-gate */ 3164*7c478bd9Sstevel@tonic-gate hm = hati_page_unmap(cur_pp, ht, entry); 3165*7c478bd9Sstevel@tonic-gate if (hm != NULL) 3166*7c478bd9Sstevel@tonic-gate hment_free(hm); 3167*7c478bd9Sstevel@tonic-gate } 3168*7c478bd9Sstevel@tonic-gate } 3169*7c478bd9Sstevel@tonic-gate 3170*7c478bd9Sstevel@tonic-gate int 3171*7c478bd9Sstevel@tonic-gate hat_pageunload(struct page *pp, uint_t forceflag) 3172*7c478bd9Sstevel@tonic-gate { 3173*7c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 3174*7c478bd9Sstevel@tonic-gate return (hati_pageunload(pp, 0, forceflag)); 3175*7c478bd9Sstevel@tonic-gate } 3176*7c478bd9Sstevel@tonic-gate 3177*7c478bd9Sstevel@tonic-gate /* 3178*7c478bd9Sstevel@tonic-gate * Unload all large mappings to pp and reduce by 1 p_szc field of every large 3179*7c478bd9Sstevel@tonic-gate * page level that included pp. 3180*7c478bd9Sstevel@tonic-gate * 3181*7c478bd9Sstevel@tonic-gate * pp must be locked EXCL. Even though no other constituent pages are locked 3182*7c478bd9Sstevel@tonic-gate * it's legal to unload large mappings to pp because all constituent pages of 3183*7c478bd9Sstevel@tonic-gate * large locked mappings have to be locked SHARED. therefore if we have EXCL 3184*7c478bd9Sstevel@tonic-gate * lock on one of constituent pages none of the large mappings to pp are 3185*7c478bd9Sstevel@tonic-gate * locked. 3186*7c478bd9Sstevel@tonic-gate * 3187*7c478bd9Sstevel@tonic-gate * Change (always decrease) p_szc field starting from the last constituent 3188*7c478bd9Sstevel@tonic-gate * page and ending with root constituent page so that root's pszc always shows 3189*7c478bd9Sstevel@tonic-gate * the area where hat_page_demote() may be active. 3190*7c478bd9Sstevel@tonic-gate * 3191*7c478bd9Sstevel@tonic-gate * This mechanism is only used for file system pages where it's not always 3192*7c478bd9Sstevel@tonic-gate * possible to get EXCL locks on all constituent pages to demote the size code 3193*7c478bd9Sstevel@tonic-gate * (as is done for anonymous or kernel large pages). 3194*7c478bd9Sstevel@tonic-gate */ 3195*7c478bd9Sstevel@tonic-gate void 3196*7c478bd9Sstevel@tonic-gate hat_page_demote(page_t *pp) 3197*7c478bd9Sstevel@tonic-gate { 3198*7c478bd9Sstevel@tonic-gate uint_t pszc; 3199*7c478bd9Sstevel@tonic-gate uint_t rszc; 3200*7c478bd9Sstevel@tonic-gate uint_t szc; 3201*7c478bd9Sstevel@tonic-gate page_t *rootpp; 3202*7c478bd9Sstevel@tonic-gate page_t *firstpp; 3203*7c478bd9Sstevel@tonic-gate page_t *lastpp; 3204*7c478bd9Sstevel@tonic-gate pgcnt_t pgcnt; 3205*7c478bd9Sstevel@tonic-gate 3206*7c478bd9Sstevel@tonic-gate ASSERT(PAGE_EXCL(pp)); 3207*7c478bd9Sstevel@tonic-gate ASSERT(!PP_ISFREE(pp)); 3208*7c478bd9Sstevel@tonic-gate ASSERT(page_szc_lock_assert(pp)); 3209*7c478bd9Sstevel@tonic-gate 3210*7c478bd9Sstevel@tonic-gate if (pp->p_szc == 0) 3211*7c478bd9Sstevel@tonic-gate return; 3212*7c478bd9Sstevel@tonic-gate 3213*7c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, 1); 3214*7c478bd9Sstevel@tonic-gate (void) hati_pageunload(rootpp, 1, HAT_FORCE_PGUNLOAD); 3215*7c478bd9Sstevel@tonic-gate 3216*7c478bd9Sstevel@tonic-gate /* 3217*7c478bd9Sstevel@tonic-gate * all large mappings to pp are gone 3218*7c478bd9Sstevel@tonic-gate * and no new can be setup since pp is locked exclusively. 3219*7c478bd9Sstevel@tonic-gate * 3220*7c478bd9Sstevel@tonic-gate * Lock the root to make sure there's only one hat_page_demote() 3221*7c478bd9Sstevel@tonic-gate * outstanding within the area of this root's pszc. 3222*7c478bd9Sstevel@tonic-gate * 3223*7c478bd9Sstevel@tonic-gate * Second potential hat_page_demote() is already eliminated by upper 3224*7c478bd9Sstevel@tonic-gate * VM layer via page_szc_lock() but we don't rely on it and use our 3225*7c478bd9Sstevel@tonic-gate * own locking (so that upper layer locking can be changed without 3226*7c478bd9Sstevel@tonic-gate * assumptions that hat depends on upper layer VM to prevent multiple 3227*7c478bd9Sstevel@tonic-gate * hat_page_demote() to be issued simultaneously to the same large 3228*7c478bd9Sstevel@tonic-gate * page). 3229*7c478bd9Sstevel@tonic-gate */ 3230*7c478bd9Sstevel@tonic-gate again: 3231*7c478bd9Sstevel@tonic-gate pszc = pp->p_szc; 3232*7c478bd9Sstevel@tonic-gate if (pszc == 0) 3233*7c478bd9Sstevel@tonic-gate return; 3234*7c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(pp, pszc); 3235*7c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 3236*7c478bd9Sstevel@tonic-gate /* 3237*7c478bd9Sstevel@tonic-gate * If root's p_szc is different from pszc we raced with another 3238*7c478bd9Sstevel@tonic-gate * hat_page_demote(). Drop the lock and try to find the root again. 3239*7c478bd9Sstevel@tonic-gate * If root's p_szc is greater than pszc previous hat_page_demote() is 3240*7c478bd9Sstevel@tonic-gate * not done yet. Take and release mlist lock of root's root to wait 3241*7c478bd9Sstevel@tonic-gate * for previous hat_page_demote() to complete. 3242*7c478bd9Sstevel@tonic-gate */ 3243*7c478bd9Sstevel@tonic-gate if ((rszc = rootpp->p_szc) != pszc) { 3244*7c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 3245*7c478bd9Sstevel@tonic-gate if (rszc > pszc) { 3246*7c478bd9Sstevel@tonic-gate /* p_szc of a locked non free page can't increase */ 3247*7c478bd9Sstevel@tonic-gate ASSERT(pp != rootpp); 3248*7c478bd9Sstevel@tonic-gate 3249*7c478bd9Sstevel@tonic-gate rootpp = PP_GROUPLEADER(rootpp, rszc); 3250*7c478bd9Sstevel@tonic-gate x86_hm_enter(rootpp); 3251*7c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 3252*7c478bd9Sstevel@tonic-gate } 3253*7c478bd9Sstevel@tonic-gate goto again; 3254*7c478bd9Sstevel@tonic-gate } 3255*7c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc == pszc); 3256*7c478bd9Sstevel@tonic-gate 3257*7c478bd9Sstevel@tonic-gate /* 3258*7c478bd9Sstevel@tonic-gate * Decrement by 1 p_szc of every constituent page of a region that 3259*7c478bd9Sstevel@tonic-gate * covered pp. For example if original szc is 3 it gets changed to 2 3260*7c478bd9Sstevel@tonic-gate * everywhere except in region 2 that covered pp. Region 2 that 3261*7c478bd9Sstevel@tonic-gate * covered pp gets demoted to 1 everywhere except in region 1 that 3262*7c478bd9Sstevel@tonic-gate * covered pp. The region 1 that covered pp is demoted to region 3263*7c478bd9Sstevel@tonic-gate * 0. It's done this way because from region 3 we removed level 3 3264*7c478bd9Sstevel@tonic-gate * mappings, from region 2 that covered pp we removed level 2 mappings 3265*7c478bd9Sstevel@tonic-gate * and from region 1 that covered pp we removed level 1 mappings. All 3266*7c478bd9Sstevel@tonic-gate * changes are done from from high pfn's to low pfn's so that roots 3267*7c478bd9Sstevel@tonic-gate * are changed last allowing one to know the largest region where 3268*7c478bd9Sstevel@tonic-gate * hat_page_demote() is stil active by only looking at the root page. 3269*7c478bd9Sstevel@tonic-gate * 3270*7c478bd9Sstevel@tonic-gate * This algorithm is implemented in 2 while loops. First loop changes 3271*7c478bd9Sstevel@tonic-gate * p_szc of pages to the right of pp's level 1 region and second 3272*7c478bd9Sstevel@tonic-gate * loop changes p_szc of pages of level 1 region that covers pp 3273*7c478bd9Sstevel@tonic-gate * and all pages to the left of level 1 region that covers pp. 3274*7c478bd9Sstevel@tonic-gate * In the first loop p_szc keeps dropping with every iteration 3275*7c478bd9Sstevel@tonic-gate * and in the second loop it keeps increasing with every iteration. 3276*7c478bd9Sstevel@tonic-gate * 3277*7c478bd9Sstevel@tonic-gate * First loop description: Demote pages to the right of pp outside of 3278*7c478bd9Sstevel@tonic-gate * level 1 region that covers pp. In every iteration of the while 3279*7c478bd9Sstevel@tonic-gate * loop below find the last page of szc region and the first page of 3280*7c478bd9Sstevel@tonic-gate * (szc - 1) region that is immediately to the right of (szc - 1) 3281*7c478bd9Sstevel@tonic-gate * region that covers pp. From last such page to first such page 3282*7c478bd9Sstevel@tonic-gate * change every page's szc to szc - 1. Decrement szc and continue 3283*7c478bd9Sstevel@tonic-gate * looping until szc is 1. If pp belongs to the last (szc - 1) region 3284*7c478bd9Sstevel@tonic-gate * of szc region skip to the next iteration. 3285*7c478bd9Sstevel@tonic-gate */ 3286*7c478bd9Sstevel@tonic-gate szc = pszc; 3287*7c478bd9Sstevel@tonic-gate while (szc > 1) { 3288*7c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 3289*7c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 3290*7c478bd9Sstevel@tonic-gate lastpp += pgcnt - 1; 3291*7c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc - 1)); 3292*7c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc - 1); 3293*7c478bd9Sstevel@tonic-gate if (lastpp - firstpp < pgcnt) { 3294*7c478bd9Sstevel@tonic-gate szc--; 3295*7c478bd9Sstevel@tonic-gate continue; 3296*7c478bd9Sstevel@tonic-gate } 3297*7c478bd9Sstevel@tonic-gate firstpp += pgcnt; 3298*7c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 3299*7c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 3300*7c478bd9Sstevel@tonic-gate lastpp->p_szc = szc - 1; 3301*7c478bd9Sstevel@tonic-gate lastpp--; 3302*7c478bd9Sstevel@tonic-gate } 3303*7c478bd9Sstevel@tonic-gate firstpp->p_szc = szc - 1; 3304*7c478bd9Sstevel@tonic-gate szc--; 3305*7c478bd9Sstevel@tonic-gate } 3306*7c478bd9Sstevel@tonic-gate 3307*7c478bd9Sstevel@tonic-gate /* 3308*7c478bd9Sstevel@tonic-gate * Second loop description: 3309*7c478bd9Sstevel@tonic-gate * First iteration changes p_szc to 0 of every 3310*7c478bd9Sstevel@tonic-gate * page of level 1 region that covers pp. 3311*7c478bd9Sstevel@tonic-gate * Subsequent iterations find last page of szc region 3312*7c478bd9Sstevel@tonic-gate * immediately to the left of szc region that covered pp 3313*7c478bd9Sstevel@tonic-gate * and first page of (szc + 1) region that covers pp. 3314*7c478bd9Sstevel@tonic-gate * From last to first page change p_szc of every page to szc. 3315*7c478bd9Sstevel@tonic-gate * Increment szc and continue looping until szc is pszc. 3316*7c478bd9Sstevel@tonic-gate * If pp belongs to the fist szc region of (szc + 1) region 3317*7c478bd9Sstevel@tonic-gate * skip to the next iteration. 3318*7c478bd9Sstevel@tonic-gate * 3319*7c478bd9Sstevel@tonic-gate */ 3320*7c478bd9Sstevel@tonic-gate szc = 0; 3321*7c478bd9Sstevel@tonic-gate while (szc < pszc) { 3322*7c478bd9Sstevel@tonic-gate firstpp = PP_GROUPLEADER(pp, (szc + 1)); 3323*7c478bd9Sstevel@tonic-gate if (szc == 0) { 3324*7c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(1); 3325*7c478bd9Sstevel@tonic-gate lastpp = firstpp + (pgcnt - 1); 3326*7c478bd9Sstevel@tonic-gate } else { 3327*7c478bd9Sstevel@tonic-gate lastpp = PP_GROUPLEADER(pp, szc); 3328*7c478bd9Sstevel@tonic-gate if (firstpp == lastpp) { 3329*7c478bd9Sstevel@tonic-gate szc++; 3330*7c478bd9Sstevel@tonic-gate continue; 3331*7c478bd9Sstevel@tonic-gate } 3332*7c478bd9Sstevel@tonic-gate lastpp--; 3333*7c478bd9Sstevel@tonic-gate pgcnt = page_get_pagecnt(szc); 3334*7c478bd9Sstevel@tonic-gate } 3335*7c478bd9Sstevel@tonic-gate while (lastpp != firstpp) { 3336*7c478bd9Sstevel@tonic-gate ASSERT(lastpp->p_szc == pszc); 3337*7c478bd9Sstevel@tonic-gate lastpp->p_szc = szc; 3338*7c478bd9Sstevel@tonic-gate lastpp--; 3339*7c478bd9Sstevel@tonic-gate } 3340*7c478bd9Sstevel@tonic-gate firstpp->p_szc = szc; 3341*7c478bd9Sstevel@tonic-gate if (firstpp == rootpp) 3342*7c478bd9Sstevel@tonic-gate break; 3343*7c478bd9Sstevel@tonic-gate szc++; 3344*7c478bd9Sstevel@tonic-gate } 3345*7c478bd9Sstevel@tonic-gate x86_hm_exit(rootpp); 3346*7c478bd9Sstevel@tonic-gate } 3347*7c478bd9Sstevel@tonic-gate 3348*7c478bd9Sstevel@tonic-gate /* 3349*7c478bd9Sstevel@tonic-gate * get hw stats from hardware into page struct and reset hw stats 3350*7c478bd9Sstevel@tonic-gate * returns attributes of page 3351*7c478bd9Sstevel@tonic-gate * Flags for hat_pagesync, hat_getstat, hat_sync 3352*7c478bd9Sstevel@tonic-gate * 3353*7c478bd9Sstevel@tonic-gate * define HAT_SYNC_ZERORM 0x01 3354*7c478bd9Sstevel@tonic-gate * 3355*7c478bd9Sstevel@tonic-gate * Additional flags for hat_pagesync 3356*7c478bd9Sstevel@tonic-gate * 3357*7c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_REF 0x02 3358*7c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_MOD 0x04 3359*7c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_RM 0x06 3360*7c478bd9Sstevel@tonic-gate * define HAT_SYNC_STOPON_SHARED 0x08 3361*7c478bd9Sstevel@tonic-gate */ 3362*7c478bd9Sstevel@tonic-gate uint_t 3363*7c478bd9Sstevel@tonic-gate hat_pagesync(struct page *pp, uint_t flags) 3364*7c478bd9Sstevel@tonic-gate { 3365*7c478bd9Sstevel@tonic-gate hment_t *hm = NULL; 3366*7c478bd9Sstevel@tonic-gate htable_t *ht; 3367*7c478bd9Sstevel@tonic-gate uint_t entry; 3368*7c478bd9Sstevel@tonic-gate x86pte_t old, save_old; 3369*7c478bd9Sstevel@tonic-gate x86pte_t new; 3370*7c478bd9Sstevel@tonic-gate uchar_t nrmbits = P_REF|P_MOD|P_RO; 3371*7c478bd9Sstevel@tonic-gate extern ulong_t po_share; 3372*7c478bd9Sstevel@tonic-gate page_t *save_pp = pp; 3373*7c478bd9Sstevel@tonic-gate uint_t pszc = 0; 3374*7c478bd9Sstevel@tonic-gate 3375*7c478bd9Sstevel@tonic-gate ASSERT(PAGE_LOCKED(pp) || panicstr); 3376*7c478bd9Sstevel@tonic-gate 3377*7c478bd9Sstevel@tonic-gate if (PP_ISRO(pp) && (flags & HAT_SYNC_STOPON_MOD)) 3378*7c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 3379*7c478bd9Sstevel@tonic-gate 3380*7c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) == 0) { 3381*7c478bd9Sstevel@tonic-gate 3382*7c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_REF) != 0 && PP_ISREF(pp)) 3383*7c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 3384*7c478bd9Sstevel@tonic-gate 3385*7c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) != 0 && PP_ISMOD(pp)) 3386*7c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 3387*7c478bd9Sstevel@tonic-gate 3388*7c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_SHARED) != 0 && 3389*7c478bd9Sstevel@tonic-gate hat_page_getshare(pp) > po_share) { 3390*7c478bd9Sstevel@tonic-gate if (PP_ISRO(pp)) 3391*7c478bd9Sstevel@tonic-gate PP_SETREF(pp); 3392*7c478bd9Sstevel@tonic-gate return (pp->p_nrm & nrmbits); 3393*7c478bd9Sstevel@tonic-gate } 3394*7c478bd9Sstevel@tonic-gate } 3395*7c478bd9Sstevel@tonic-gate 3396*7c478bd9Sstevel@tonic-gate next_size: 3397*7c478bd9Sstevel@tonic-gate /* 3398*7c478bd9Sstevel@tonic-gate * walk thru the mapping list syncing (and clearing) ref/mod bits. 3399*7c478bd9Sstevel@tonic-gate */ 3400*7c478bd9Sstevel@tonic-gate x86_hm_enter(pp); 3401*7c478bd9Sstevel@tonic-gate while ((hm = hment_walk(pp, &ht, &entry, hm)) != NULL) { 3402*7c478bd9Sstevel@tonic-gate if (ht->ht_level < pszc) 3403*7c478bd9Sstevel@tonic-gate continue; 3404*7c478bd9Sstevel@tonic-gate old = x86pte_get(ht, entry); 3405*7c478bd9Sstevel@tonic-gate try_again: 3406*7c478bd9Sstevel@tonic-gate 3407*7c478bd9Sstevel@tonic-gate ASSERT(PTE2PFN(old, ht->ht_level) == pp->p_pagenum); 3408*7c478bd9Sstevel@tonic-gate 3409*7c478bd9Sstevel@tonic-gate if (PTE_GET(old, PT_REF | PT_MOD) == 0) 3410*7c478bd9Sstevel@tonic-gate continue; 3411*7c478bd9Sstevel@tonic-gate 3412*7c478bd9Sstevel@tonic-gate save_old = old; 3413*7c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_ZERORM) != 0) { 3414*7c478bd9Sstevel@tonic-gate 3415*7c478bd9Sstevel@tonic-gate /* 3416*7c478bd9Sstevel@tonic-gate * Need to clear ref or mod bits. Need to demap 3417*7c478bd9Sstevel@tonic-gate * to make sure any executing TLBs see cleared bits. 3418*7c478bd9Sstevel@tonic-gate */ 3419*7c478bd9Sstevel@tonic-gate new = old; 3420*7c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_REF | PT_MOD); 3421*7c478bd9Sstevel@tonic-gate old = hati_update_pte(ht, entry, old, new); 3422*7c478bd9Sstevel@tonic-gate if (old != 0) 3423*7c478bd9Sstevel@tonic-gate goto try_again; 3424*7c478bd9Sstevel@tonic-gate 3425*7c478bd9Sstevel@tonic-gate old = save_old; 3426*7c478bd9Sstevel@tonic-gate } 3427*7c478bd9Sstevel@tonic-gate 3428*7c478bd9Sstevel@tonic-gate /* 3429*7c478bd9Sstevel@tonic-gate * Sync the PTE 3430*7c478bd9Sstevel@tonic-gate */ 3431*7c478bd9Sstevel@tonic-gate if (!(flags & HAT_SYNC_ZERORM) && PTE_GET(old, PT_NOSYNC) == 0) 3432*7c478bd9Sstevel@tonic-gate hati_sync_pte_to_page(pp, old, ht->ht_level); 3433*7c478bd9Sstevel@tonic-gate 3434*7c478bd9Sstevel@tonic-gate /* 3435*7c478bd9Sstevel@tonic-gate * can stop short if we found a ref'd or mod'd page 3436*7c478bd9Sstevel@tonic-gate */ 3437*7c478bd9Sstevel@tonic-gate if ((flags & HAT_SYNC_STOPON_MOD) && PP_ISMOD(save_pp) || 3438*7c478bd9Sstevel@tonic-gate (flags & HAT_SYNC_STOPON_REF) && PP_ISREF(save_pp)) { 3439*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3440*7c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 3441*7c478bd9Sstevel@tonic-gate } 3442*7c478bd9Sstevel@tonic-gate } 3443*7c478bd9Sstevel@tonic-gate x86_hm_exit(pp); 3444*7c478bd9Sstevel@tonic-gate while (pszc < pp->p_szc) { 3445*7c478bd9Sstevel@tonic-gate page_t *tpp; 3446*7c478bd9Sstevel@tonic-gate pszc++; 3447*7c478bd9Sstevel@tonic-gate tpp = PP_GROUPLEADER(pp, pszc); 3448*7c478bd9Sstevel@tonic-gate if (pp != tpp) { 3449*7c478bd9Sstevel@tonic-gate pp = tpp; 3450*7c478bd9Sstevel@tonic-gate goto next_size; 3451*7c478bd9Sstevel@tonic-gate } 3452*7c478bd9Sstevel@tonic-gate } 3453*7c478bd9Sstevel@tonic-gate return (save_pp->p_nrm & nrmbits); 3454*7c478bd9Sstevel@tonic-gate } 3455*7c478bd9Sstevel@tonic-gate 3456*7c478bd9Sstevel@tonic-gate /* 3457*7c478bd9Sstevel@tonic-gate * returns approx number of mappings to this pp. A return of 0 implies 3458*7c478bd9Sstevel@tonic-gate * there are no mappings to the page. 3459*7c478bd9Sstevel@tonic-gate */ 3460*7c478bd9Sstevel@tonic-gate ulong_t 3461*7c478bd9Sstevel@tonic-gate hat_page_getshare(page_t *pp) 3462*7c478bd9Sstevel@tonic-gate { 3463*7c478bd9Sstevel@tonic-gate uint_t cnt; 3464*7c478bd9Sstevel@tonic-gate cnt = hment_mapcnt(pp); 3465*7c478bd9Sstevel@tonic-gate return (cnt); 3466*7c478bd9Sstevel@tonic-gate } 3467*7c478bd9Sstevel@tonic-gate 3468*7c478bd9Sstevel@tonic-gate /* 3469*7c478bd9Sstevel@tonic-gate * hat_softlock isn't supported anymore 3470*7c478bd9Sstevel@tonic-gate */ 3471*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3472*7c478bd9Sstevel@tonic-gate faultcode_t 3473*7c478bd9Sstevel@tonic-gate hat_softlock( 3474*7c478bd9Sstevel@tonic-gate hat_t *hat, 3475*7c478bd9Sstevel@tonic-gate caddr_t addr, 3476*7c478bd9Sstevel@tonic-gate size_t *len, 3477*7c478bd9Sstevel@tonic-gate struct page **page_array, 3478*7c478bd9Sstevel@tonic-gate uint_t flags) 3479*7c478bd9Sstevel@tonic-gate { 3480*7c478bd9Sstevel@tonic-gate return (FC_NOSUPPORT); 3481*7c478bd9Sstevel@tonic-gate } 3482*7c478bd9Sstevel@tonic-gate 3483*7c478bd9Sstevel@tonic-gate 3484*7c478bd9Sstevel@tonic-gate 3485*7c478bd9Sstevel@tonic-gate /* 3486*7c478bd9Sstevel@tonic-gate * Routine to expose supported HAT features to platform independent code. 3487*7c478bd9Sstevel@tonic-gate */ 3488*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3489*7c478bd9Sstevel@tonic-gate int 3490*7c478bd9Sstevel@tonic-gate hat_supported(enum hat_features feature, void *arg) 3491*7c478bd9Sstevel@tonic-gate { 3492*7c478bd9Sstevel@tonic-gate switch (feature) { 3493*7c478bd9Sstevel@tonic-gate 3494*7c478bd9Sstevel@tonic-gate case HAT_SHARED_PT: /* this is really ISM */ 3495*7c478bd9Sstevel@tonic-gate return (1); 3496*7c478bd9Sstevel@tonic-gate 3497*7c478bd9Sstevel@tonic-gate case HAT_DYNAMIC_ISM_UNMAP: 3498*7c478bd9Sstevel@tonic-gate return (0); 3499*7c478bd9Sstevel@tonic-gate 3500*7c478bd9Sstevel@tonic-gate case HAT_VMODSORT: 3501*7c478bd9Sstevel@tonic-gate return (1); 3502*7c478bd9Sstevel@tonic-gate 3503*7c478bd9Sstevel@tonic-gate default: 3504*7c478bd9Sstevel@tonic-gate panic("hat_supported() - unknown feature"); 3505*7c478bd9Sstevel@tonic-gate } 3506*7c478bd9Sstevel@tonic-gate return (0); 3507*7c478bd9Sstevel@tonic-gate } 3508*7c478bd9Sstevel@tonic-gate 3509*7c478bd9Sstevel@tonic-gate /* 3510*7c478bd9Sstevel@tonic-gate * Called when a thread is exiting and has been switched to the kernel AS 3511*7c478bd9Sstevel@tonic-gate */ 3512*7c478bd9Sstevel@tonic-gate void 3513*7c478bd9Sstevel@tonic-gate hat_thread_exit(kthread_t *thd) 3514*7c478bd9Sstevel@tonic-gate { 3515*7c478bd9Sstevel@tonic-gate ASSERT(thd->t_procp->p_as == &kas); 3516*7c478bd9Sstevel@tonic-gate hat_switch(thd->t_procp->p_as->a_hat); 3517*7c478bd9Sstevel@tonic-gate } 3518*7c478bd9Sstevel@tonic-gate 3519*7c478bd9Sstevel@tonic-gate /* 3520*7c478bd9Sstevel@tonic-gate * Setup the given brand new hat structure as the new HAT on this cpu's mmu. 3521*7c478bd9Sstevel@tonic-gate */ 3522*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3523*7c478bd9Sstevel@tonic-gate void 3524*7c478bd9Sstevel@tonic-gate hat_setup(hat_t *hat, int flags) 3525*7c478bd9Sstevel@tonic-gate { 3526*7c478bd9Sstevel@tonic-gate kpreempt_disable(); 3527*7c478bd9Sstevel@tonic-gate 3528*7c478bd9Sstevel@tonic-gate hat_switch(hat); 3529*7c478bd9Sstevel@tonic-gate 3530*7c478bd9Sstevel@tonic-gate kpreempt_enable(); 3531*7c478bd9Sstevel@tonic-gate } 3532*7c478bd9Sstevel@tonic-gate 3533*7c478bd9Sstevel@tonic-gate /* 3534*7c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 3535*7c478bd9Sstevel@tonic-gate * 3536*7c478bd9Sstevel@tonic-gate * The address can only be used from a single CPU and can be remapped 3537*7c478bd9Sstevel@tonic-gate * using hat_mempte_remap(). Return the address of the PTE. 3538*7c478bd9Sstevel@tonic-gate * 3539*7c478bd9Sstevel@tonic-gate * We do the htable_create() if necessary and increment the valid count so 3540*7c478bd9Sstevel@tonic-gate * the htable can't disappear. We also hat_devload() the page table into 3541*7c478bd9Sstevel@tonic-gate * kernel so that the PTE is quickly accessed. 3542*7c478bd9Sstevel@tonic-gate */ 3543*7c478bd9Sstevel@tonic-gate void * 3544*7c478bd9Sstevel@tonic-gate hat_mempte_kern_setup(caddr_t addr, void *pt) 3545*7c478bd9Sstevel@tonic-gate { 3546*7c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 3547*7c478bd9Sstevel@tonic-gate htable_t *ht; 3548*7c478bd9Sstevel@tonic-gate uint_t entry; 3549*7c478bd9Sstevel@tonic-gate x86pte_t oldpte; 3550*7c478bd9Sstevel@tonic-gate caddr_t p = (caddr_t)pt; 3551*7c478bd9Sstevel@tonic-gate 3552*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 3553*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3554*7c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, &oldpte, 0); 3555*7c478bd9Sstevel@tonic-gate if (ht == NULL) { 3556*7c478bd9Sstevel@tonic-gate /* 3557*7c478bd9Sstevel@tonic-gate * Note that we don't need a hat_reserves_exit() check 3558*7c478bd9Sstevel@tonic-gate * for this htable_create(), since that'll be done by the 3559*7c478bd9Sstevel@tonic-gate * hat_devload() just below. 3560*7c478bd9Sstevel@tonic-gate */ 3561*7c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, va, 0, NULL); 3562*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(va, ht); 3563*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 3564*7c478bd9Sstevel@tonic-gate oldpte = x86pte_get(ht, entry); 3565*7c478bd9Sstevel@tonic-gate } 3566*7c478bd9Sstevel@tonic-gate if (PTE_ISVALID(oldpte)) 3567*7c478bd9Sstevel@tonic-gate panic("hat_mempte_setup(): address already mapped" 3568*7c478bd9Sstevel@tonic-gate "ht=%p, entry=%d, pte=" FMT_PTE, ht, entry, oldpte); 3569*7c478bd9Sstevel@tonic-gate 3570*7c478bd9Sstevel@tonic-gate /* 3571*7c478bd9Sstevel@tonic-gate * increment ht_valid_cnt so that the pagetable can't disappear 3572*7c478bd9Sstevel@tonic-gate */ 3573*7c478bd9Sstevel@tonic-gate HTABLE_INC(ht->ht_valid_cnt); 3574*7c478bd9Sstevel@tonic-gate 3575*7c478bd9Sstevel@tonic-gate /* 3576*7c478bd9Sstevel@tonic-gate * now we need to map the page holding the pagetable for va into 3577*7c478bd9Sstevel@tonic-gate * the kernel's address space. 3578*7c478bd9Sstevel@tonic-gate */ 3579*7c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, p, MMU_PAGESIZE, ht->ht_pfn, 3580*7c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | HAT_NOSYNC | HAT_UNORDERED_OK, 3581*7c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 3582*7c478bd9Sstevel@tonic-gate 3583*7c478bd9Sstevel@tonic-gate /* 3584*7c478bd9Sstevel@tonic-gate * return the PTE address to the caller. 3585*7c478bd9Sstevel@tonic-gate */ 3586*7c478bd9Sstevel@tonic-gate htable_release(ht); 3587*7c478bd9Sstevel@tonic-gate p += entry << mmu.pte_size_shift; 3588*7c478bd9Sstevel@tonic-gate return ((void *)p); 3589*7c478bd9Sstevel@tonic-gate } 3590*7c478bd9Sstevel@tonic-gate 3591*7c478bd9Sstevel@tonic-gate /* 3592*7c478bd9Sstevel@tonic-gate * Prepare for a CPU private mapping for the given address. 3593*7c478bd9Sstevel@tonic-gate */ 3594*7c478bd9Sstevel@tonic-gate void * 3595*7c478bd9Sstevel@tonic-gate hat_mempte_setup(caddr_t addr) 3596*7c478bd9Sstevel@tonic-gate { 3597*7c478bd9Sstevel@tonic-gate x86pte_t *p; 3598*7c478bd9Sstevel@tonic-gate 3599*7c478bd9Sstevel@tonic-gate p = vmem_alloc(heap_arena, MMU_PAGESIZE, VM_SLEEP); 3600*7c478bd9Sstevel@tonic-gate return (hat_mempte_kern_setup(addr, p)); 3601*7c478bd9Sstevel@tonic-gate } 3602*7c478bd9Sstevel@tonic-gate 3603*7c478bd9Sstevel@tonic-gate /* 3604*7c478bd9Sstevel@tonic-gate * Release a CPU private mapping for the given address. 3605*7c478bd9Sstevel@tonic-gate * We decrement the htable valid count so it might be destroyed. 3606*7c478bd9Sstevel@tonic-gate */ 3607*7c478bd9Sstevel@tonic-gate void 3608*7c478bd9Sstevel@tonic-gate hat_mempte_release(caddr_t addr, void *pteptr) 3609*7c478bd9Sstevel@tonic-gate { 3610*7c478bd9Sstevel@tonic-gate htable_t *ht; 3611*7c478bd9Sstevel@tonic-gate uintptr_t va = ALIGN2PAGE(pteptr); 3612*7c478bd9Sstevel@tonic-gate 3613*7c478bd9Sstevel@tonic-gate /* 3614*7c478bd9Sstevel@tonic-gate * first invalidate any left over mapping and decrement the 3615*7c478bd9Sstevel@tonic-gate * htable's mapping count 3616*7c478bd9Sstevel@tonic-gate */ 3617*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 3618*7c478bd9Sstevel@tonic-gate *(x86pte_t *)pteptr = 0; 3619*7c478bd9Sstevel@tonic-gate else 3620*7c478bd9Sstevel@tonic-gate *(x86pte32_t *)pteptr = 0; 3621*7c478bd9Sstevel@tonic-gate mmu_tlbflush_entry(addr); 3622*7c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, ALIGN2PAGE(addr), NULL, NULL, 0); 3623*7c478bd9Sstevel@tonic-gate if (ht == NULL) 3624*7c478bd9Sstevel@tonic-gate panic("hat_mempte_release(): invalid address"); 3625*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 3626*7c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 3627*7c478bd9Sstevel@tonic-gate htable_release(ht); 3628*7c478bd9Sstevel@tonic-gate 3629*7c478bd9Sstevel@tonic-gate /* 3630*7c478bd9Sstevel@tonic-gate * now blow away the kernel mapping to the page table page 3631*7c478bd9Sstevel@tonic-gate * XX64 -- see comment in hat_mempte_setup() 3632*7c478bd9Sstevel@tonic-gate */ 3633*7c478bd9Sstevel@tonic-gate hat_unload_callback(kas.a_hat, (caddr_t)va, MMU_PAGESIZE, 3634*7c478bd9Sstevel@tonic-gate HAT_UNLOAD, NULL); 3635*7c478bd9Sstevel@tonic-gate } 3636*7c478bd9Sstevel@tonic-gate 3637*7c478bd9Sstevel@tonic-gate /* 3638*7c478bd9Sstevel@tonic-gate * Apply a temporary CPU private mapping to a page. We flush the TLB only 3639*7c478bd9Sstevel@tonic-gate * on this CPU, so this ought to have been called with preemption disabled. 3640*7c478bd9Sstevel@tonic-gate */ 3641*7c478bd9Sstevel@tonic-gate void 3642*7c478bd9Sstevel@tonic-gate hat_mempte_remap( 3643*7c478bd9Sstevel@tonic-gate pfn_t pfn, 3644*7c478bd9Sstevel@tonic-gate caddr_t addr, 3645*7c478bd9Sstevel@tonic-gate void *pteptr, 3646*7c478bd9Sstevel@tonic-gate uint_t attr, 3647*7c478bd9Sstevel@tonic-gate uint_t flags) 3648*7c478bd9Sstevel@tonic-gate { 3649*7c478bd9Sstevel@tonic-gate uintptr_t va = (uintptr_t)addr; 3650*7c478bd9Sstevel@tonic-gate x86pte_t pte; 3651*7c478bd9Sstevel@tonic-gate 3652*7c478bd9Sstevel@tonic-gate /* 3653*7c478bd9Sstevel@tonic-gate * Remap the given PTE to the new page's PFN. Invalidate only 3654*7c478bd9Sstevel@tonic-gate * on this CPU. 3655*7c478bd9Sstevel@tonic-gate */ 3656*7c478bd9Sstevel@tonic-gate #ifdef DEBUG 3657*7c478bd9Sstevel@tonic-gate htable_t *ht; 3658*7c478bd9Sstevel@tonic-gate uint_t entry; 3659*7c478bd9Sstevel@tonic-gate 3660*7c478bd9Sstevel@tonic-gate ASSERT(IS_PAGEALIGNED(va)); 3661*7c478bd9Sstevel@tonic-gate ASSERT(!IN_VA_HOLE(va)); 3662*7c478bd9Sstevel@tonic-gate ht = htable_getpte(kas.a_hat, va, &entry, NULL, 0); 3663*7c478bd9Sstevel@tonic-gate ASSERT(ht != NULL); 3664*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_level == 0); 3665*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3666*7c478bd9Sstevel@tonic-gate htable_release(ht); 3667*7c478bd9Sstevel@tonic-gate #endif 3668*7c478bd9Sstevel@tonic-gate pte = hati_mkpte(pfn, attr, 0, flags); 3669*7c478bd9Sstevel@tonic-gate if (mmu.pae_hat) 3670*7c478bd9Sstevel@tonic-gate *(x86pte_t *)pteptr = pte; 3671*7c478bd9Sstevel@tonic-gate else 3672*7c478bd9Sstevel@tonic-gate *(x86pte32_t *)pteptr = (x86pte32_t)pte; 3673*7c478bd9Sstevel@tonic-gate mmu_tlbflush_entry(addr); 3674*7c478bd9Sstevel@tonic-gate } 3675*7c478bd9Sstevel@tonic-gate 3676*7c478bd9Sstevel@tonic-gate 3677*7c478bd9Sstevel@tonic-gate 3678*7c478bd9Sstevel@tonic-gate /* 3679*7c478bd9Sstevel@tonic-gate * Hat locking functions 3680*7c478bd9Sstevel@tonic-gate * XXX - these two functions are currently being used by hatstats 3681*7c478bd9Sstevel@tonic-gate * they can be removed by using a per-as mutex for hatstats. 3682*7c478bd9Sstevel@tonic-gate */ 3683*7c478bd9Sstevel@tonic-gate void 3684*7c478bd9Sstevel@tonic-gate hat_enter(hat_t *hat) 3685*7c478bd9Sstevel@tonic-gate { 3686*7c478bd9Sstevel@tonic-gate mutex_enter(&hat->hat_mutex); 3687*7c478bd9Sstevel@tonic-gate } 3688*7c478bd9Sstevel@tonic-gate 3689*7c478bd9Sstevel@tonic-gate void 3690*7c478bd9Sstevel@tonic-gate hat_exit(hat_t *hat) 3691*7c478bd9Sstevel@tonic-gate { 3692*7c478bd9Sstevel@tonic-gate mutex_exit(&hat->hat_mutex); 3693*7c478bd9Sstevel@tonic-gate } 3694*7c478bd9Sstevel@tonic-gate 3695*7c478bd9Sstevel@tonic-gate 3696*7c478bd9Sstevel@tonic-gate /* 3697*7c478bd9Sstevel@tonic-gate * Used by hat_kern_setup() to create initial kernel HAT mappings from 3698*7c478bd9Sstevel@tonic-gate * the boot loader's mappings. 3699*7c478bd9Sstevel@tonic-gate * 3700*7c478bd9Sstevel@tonic-gate * - size is either PAGESIZE or some multiple of a level one pagesize 3701*7c478bd9Sstevel@tonic-gate * - there may not be page_t's for every pfn. (ie. the nucleus pages) 3702*7c478bd9Sstevel@tonic-gate * - pfn's are continguous for the given va range (va to va + size * cnt) 3703*7c478bd9Sstevel@tonic-gate */ 3704*7c478bd9Sstevel@tonic-gate void 3705*7c478bd9Sstevel@tonic-gate hati_kern_setup_load( 3706*7c478bd9Sstevel@tonic-gate uintptr_t va, /* starting va of range to map */ 3707*7c478bd9Sstevel@tonic-gate size_t size, /* either PAGESIZE or multiple of large page size */ 3708*7c478bd9Sstevel@tonic-gate pfn_t pfn, /* starting PFN */ 3709*7c478bd9Sstevel@tonic-gate pgcnt_t cnt, /* number of mappings, (cnt * size) == total size */ 3710*7c478bd9Sstevel@tonic-gate uint_t prot) /* protections (PROT_READ, PROT_WRITE, PROT_EXEC) */ 3711*7c478bd9Sstevel@tonic-gate { 3712*7c478bd9Sstevel@tonic-gate level_t level = (size == MMU_PAGESIZE ? 0 : 1); 3713*7c478bd9Sstevel@tonic-gate size_t bytes = size * cnt; 3714*7c478bd9Sstevel@tonic-gate size_t pgsize = LEVEL_SIZE(level); 3715*7c478bd9Sstevel@tonic-gate page_t *pp; 3716*7c478bd9Sstevel@tonic-gate uint_t flags = HAT_LOAD; 3717*7c478bd9Sstevel@tonic-gate 3718*7c478bd9Sstevel@tonic-gate /* 3719*7c478bd9Sstevel@tonic-gate * We're only going to throw away mappings below kernelbase or in 3720*7c478bd9Sstevel@tonic-gate * boot's special double-mapping region, so set noconsist to avoid 3721*7c478bd9Sstevel@tonic-gate * using hments 3722*7c478bd9Sstevel@tonic-gate */ 3723*7c478bd9Sstevel@tonic-gate if (BOOT_VA(va)) 3724*7c478bd9Sstevel@tonic-gate flags |= HAT_LOAD_NOCONSIST; 3725*7c478bd9Sstevel@tonic-gate 3726*7c478bd9Sstevel@tonic-gate prot |= HAT_STORECACHING_OK; 3727*7c478bd9Sstevel@tonic-gate while (bytes != 0) { 3728*7c478bd9Sstevel@tonic-gate ASSERT(bytes >= pgsize); 3729*7c478bd9Sstevel@tonic-gate 3730*7c478bd9Sstevel@tonic-gate pp = NULL; 3731*7c478bd9Sstevel@tonic-gate if (pf_is_memory(pfn) && !BOOT_VA(va) && level == 0) 3732*7c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(pfn); 3733*7c478bd9Sstevel@tonic-gate 3734*7c478bd9Sstevel@tonic-gate hati_load_common(kas.a_hat, va, pp, prot, flags, level, pfn); 3735*7c478bd9Sstevel@tonic-gate 3736*7c478bd9Sstevel@tonic-gate va += pgsize; 3737*7c478bd9Sstevel@tonic-gate pfn += mmu_btop(pgsize); 3738*7c478bd9Sstevel@tonic-gate bytes -= pgsize; 3739*7c478bd9Sstevel@tonic-gate } 3740*7c478bd9Sstevel@tonic-gate } 3741*7c478bd9Sstevel@tonic-gate 3742*7c478bd9Sstevel@tonic-gate /* 3743*7c478bd9Sstevel@tonic-gate * HAT part of cpu intialization. 3744*7c478bd9Sstevel@tonic-gate */ 3745*7c478bd9Sstevel@tonic-gate void 3746*7c478bd9Sstevel@tonic-gate hat_cpu_online(struct cpu *cpup) 3747*7c478bd9Sstevel@tonic-gate { 3748*7c478bd9Sstevel@tonic-gate if (cpup != CPU) { 3749*7c478bd9Sstevel@tonic-gate x86pte_cpu_init(cpup, NULL); 3750*7c478bd9Sstevel@tonic-gate hat_vlp_setup(cpup); 3751*7c478bd9Sstevel@tonic-gate } 3752*7c478bd9Sstevel@tonic-gate CPUSET_ATOMIC_ADD(khat_cpuset, cpup->cpu_id); 3753*7c478bd9Sstevel@tonic-gate } 3754*7c478bd9Sstevel@tonic-gate 3755*7c478bd9Sstevel@tonic-gate /* 3756*7c478bd9Sstevel@tonic-gate * Function called after all CPUs are brought online. 3757*7c478bd9Sstevel@tonic-gate * Used to remove low address boot mappings. 3758*7c478bd9Sstevel@tonic-gate */ 3759*7c478bd9Sstevel@tonic-gate void 3760*7c478bd9Sstevel@tonic-gate clear_boot_mappings(uintptr_t low, uintptr_t high) 3761*7c478bd9Sstevel@tonic-gate { 3762*7c478bd9Sstevel@tonic-gate uintptr_t vaddr = low; 3763*7c478bd9Sstevel@tonic-gate htable_t *ht = NULL; 3764*7c478bd9Sstevel@tonic-gate level_t level; 3765*7c478bd9Sstevel@tonic-gate uint_t entry; 3766*7c478bd9Sstevel@tonic-gate x86pte_t pte; 3767*7c478bd9Sstevel@tonic-gate 3768*7c478bd9Sstevel@tonic-gate /* 3769*7c478bd9Sstevel@tonic-gate * On 1st CPU we can unload the prom mappings, basically we blow away 3770*7c478bd9Sstevel@tonic-gate * all virtual mappings under kernelbase. 3771*7c478bd9Sstevel@tonic-gate */ 3772*7c478bd9Sstevel@tonic-gate while (vaddr < high) { 3773*7c478bd9Sstevel@tonic-gate pte = htable_walk(kas.a_hat, &ht, &vaddr, high); 3774*7c478bd9Sstevel@tonic-gate if (ht == NULL) 3775*7c478bd9Sstevel@tonic-gate break; 3776*7c478bd9Sstevel@tonic-gate 3777*7c478bd9Sstevel@tonic-gate level = ht->ht_level; 3778*7c478bd9Sstevel@tonic-gate entry = htable_va2entry(vaddr, ht); 3779*7c478bd9Sstevel@tonic-gate ASSERT(level <= mmu.max_page_level); 3780*7c478bd9Sstevel@tonic-gate ASSERT(PTE_ISPAGE(pte, level)); 3781*7c478bd9Sstevel@tonic-gate 3782*7c478bd9Sstevel@tonic-gate /* 3783*7c478bd9Sstevel@tonic-gate * Unload the mapping from the page tables. 3784*7c478bd9Sstevel@tonic-gate */ 3785*7c478bd9Sstevel@tonic-gate (void) x86pte_set(ht, entry, 0, NULL); 3786*7c478bd9Sstevel@tonic-gate ASSERT(ht->ht_valid_cnt > 0); 3787*7c478bd9Sstevel@tonic-gate HTABLE_DEC(ht->ht_valid_cnt); 3788*7c478bd9Sstevel@tonic-gate PGCNT_DEC(ht->ht_hat, ht->ht_level); 3789*7c478bd9Sstevel@tonic-gate 3790*7c478bd9Sstevel@tonic-gate vaddr += LEVEL_SIZE(ht->ht_level); 3791*7c478bd9Sstevel@tonic-gate } 3792*7c478bd9Sstevel@tonic-gate if (ht) 3793*7c478bd9Sstevel@tonic-gate htable_release(ht); 3794*7c478bd9Sstevel@tonic-gate 3795*7c478bd9Sstevel@tonic-gate /* 3796*7c478bd9Sstevel@tonic-gate * cross call for a complete invalidate. 3797*7c478bd9Sstevel@tonic-gate */ 3798*7c478bd9Sstevel@tonic-gate hat_demap(kas.a_hat, DEMAP_ALL_ADDR); 3799*7c478bd9Sstevel@tonic-gate } 3800*7c478bd9Sstevel@tonic-gate 3801*7c478bd9Sstevel@tonic-gate /* 3802*7c478bd9Sstevel@tonic-gate * Initialize a special area in the kernel that always holds some PTEs for 3803*7c478bd9Sstevel@tonic-gate * faster performance. This always holds segmap's PTEs. 3804*7c478bd9Sstevel@tonic-gate * In the 32 bit kernel this maps the kernel heap too. 3805*7c478bd9Sstevel@tonic-gate */ 3806*7c478bd9Sstevel@tonic-gate void 3807*7c478bd9Sstevel@tonic-gate hat_kmap_init(uintptr_t base, size_t len) 3808*7c478bd9Sstevel@tonic-gate { 3809*7c478bd9Sstevel@tonic-gate uintptr_t map_addr; /* base rounded down to large page size */ 3810*7c478bd9Sstevel@tonic-gate uintptr_t map_eaddr; /* base + len rounded up */ 3811*7c478bd9Sstevel@tonic-gate size_t map_len; 3812*7c478bd9Sstevel@tonic-gate caddr_t ptes; /* mapping area in kernel as for ptes */ 3813*7c478bd9Sstevel@tonic-gate size_t window_size; /* size of mapping area for ptes */ 3814*7c478bd9Sstevel@tonic-gate ulong_t htable_cnt; /* # of page tables to cover map_len */ 3815*7c478bd9Sstevel@tonic-gate ulong_t i; 3816*7c478bd9Sstevel@tonic-gate htable_t *ht; 3817*7c478bd9Sstevel@tonic-gate 3818*7c478bd9Sstevel@tonic-gate /* 3819*7c478bd9Sstevel@tonic-gate * we have to map in an area that matches an entire page table 3820*7c478bd9Sstevel@tonic-gate */ 3821*7c478bd9Sstevel@tonic-gate map_addr = base & LEVEL_MASK(1); 3822*7c478bd9Sstevel@tonic-gate map_eaddr = (base + len + LEVEL_SIZE(1) - 1) & LEVEL_MASK(1); 3823*7c478bd9Sstevel@tonic-gate map_len = map_eaddr - map_addr; 3824*7c478bd9Sstevel@tonic-gate window_size = mmu_btop(map_len) * mmu.pte_size; 3825*7c478bd9Sstevel@tonic-gate htable_cnt = mmu_btop(map_len) / mmu.ptes_per_table; 3826*7c478bd9Sstevel@tonic-gate 3827*7c478bd9Sstevel@tonic-gate /* 3828*7c478bd9Sstevel@tonic-gate * allocate vmem for the kmap_ptes 3829*7c478bd9Sstevel@tonic-gate */ 3830*7c478bd9Sstevel@tonic-gate ptes = vmem_xalloc(heap_arena, window_size, MMU_PAGESIZE, 0, 3831*7c478bd9Sstevel@tonic-gate 0, NULL, NULL, VM_SLEEP); 3832*7c478bd9Sstevel@tonic-gate mmu.kmap_htables = 3833*7c478bd9Sstevel@tonic-gate kmem_alloc(htable_cnt * sizeof (htable_t *), KM_SLEEP); 3834*7c478bd9Sstevel@tonic-gate 3835*7c478bd9Sstevel@tonic-gate /* 3836*7c478bd9Sstevel@tonic-gate * Map the page tables that cover kmap into the allocated range. 3837*7c478bd9Sstevel@tonic-gate * Note we don't ever htable_release() the kmap page tables - they 3838*7c478bd9Sstevel@tonic-gate * can't ever be stolen, freed, etc. 3839*7c478bd9Sstevel@tonic-gate */ 3840*7c478bd9Sstevel@tonic-gate for (i = 0; i < htable_cnt; ++i) { 3841*7c478bd9Sstevel@tonic-gate ht = htable_create(kas.a_hat, map_addr + i * LEVEL_SIZE(1), 3842*7c478bd9Sstevel@tonic-gate 0, NULL); 3843*7c478bd9Sstevel@tonic-gate mmu.kmap_htables[i] = ht; 3844*7c478bd9Sstevel@tonic-gate 3845*7c478bd9Sstevel@tonic-gate hat_devload(kas.a_hat, ptes + i * MMU_PAGESIZE, MMU_PAGESIZE, 3846*7c478bd9Sstevel@tonic-gate ht->ht_pfn, 3847*7c478bd9Sstevel@tonic-gate PROT_READ | PROT_WRITE | HAT_NOSYNC | HAT_UNORDERED_OK, 3848*7c478bd9Sstevel@tonic-gate HAT_LOAD | HAT_LOAD_NOCONSIST); 3849*7c478bd9Sstevel@tonic-gate 3850*7c478bd9Sstevel@tonic-gate } 3851*7c478bd9Sstevel@tonic-gate 3852*7c478bd9Sstevel@tonic-gate /* 3853*7c478bd9Sstevel@tonic-gate * set information in mmu to activate handling of kmap 3854*7c478bd9Sstevel@tonic-gate */ 3855*7c478bd9Sstevel@tonic-gate mmu.kmap_addr = base; 3856*7c478bd9Sstevel@tonic-gate mmu.kmap_eaddr = base + len; 3857*7c478bd9Sstevel@tonic-gate mmu.kmap_ptes = 3858*7c478bd9Sstevel@tonic-gate (x86pte_t *)(ptes + mmu.pte_size * mmu_btop(base - map_addr)); 3859*7c478bd9Sstevel@tonic-gate } 3860*7c478bd9Sstevel@tonic-gate 3861*7c478bd9Sstevel@tonic-gate /* 3862*7c478bd9Sstevel@tonic-gate * Atomically update a new translation for a single page. If the 3863*7c478bd9Sstevel@tonic-gate * currently installed PTE doesn't match the value we expect to find, 3864*7c478bd9Sstevel@tonic-gate * it's not updated and we return the PTE we found. 3865*7c478bd9Sstevel@tonic-gate * 3866*7c478bd9Sstevel@tonic-gate * If activating nosync or NOWRITE and the page was modified we need to sync 3867*7c478bd9Sstevel@tonic-gate * with the page_t. Also sync with page_t if clearing ref/mod bits. 3868*7c478bd9Sstevel@tonic-gate */ 3869*7c478bd9Sstevel@tonic-gate static x86pte_t 3870*7c478bd9Sstevel@tonic-gate hati_update_pte(htable_t *ht, uint_t entry, x86pte_t expected, x86pte_t new) 3871*7c478bd9Sstevel@tonic-gate { 3872*7c478bd9Sstevel@tonic-gate page_t *pp; 3873*7c478bd9Sstevel@tonic-gate uint_t rm = 0; 3874*7c478bd9Sstevel@tonic-gate x86pte_t replaced; 3875*7c478bd9Sstevel@tonic-gate 3876*7c478bd9Sstevel@tonic-gate if (!PTE_GET(expected, PT_NOSYNC | PT_NOCONSIST) && 3877*7c478bd9Sstevel@tonic-gate PTE_GET(expected, PT_MOD | PT_REF) && 3878*7c478bd9Sstevel@tonic-gate (PTE_GET(new, PT_NOSYNC) || !PTE_GET(new, PT_WRITABLE) || 3879*7c478bd9Sstevel@tonic-gate !PTE_GET(new, PT_MOD | PT_REF))) { 3880*7c478bd9Sstevel@tonic-gate 3881*7c478bd9Sstevel@tonic-gate pp = page_numtopp_nolock(PTE2PFN(expected, ht->ht_level)); 3882*7c478bd9Sstevel@tonic-gate ASSERT(pp != NULL); 3883*7c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_MOD)) 3884*7c478bd9Sstevel@tonic-gate rm |= P_MOD; 3885*7c478bd9Sstevel@tonic-gate if (PTE_GET(expected, PT_REF)) 3886*7c478bd9Sstevel@tonic-gate rm |= P_REF; 3887*7c478bd9Sstevel@tonic-gate PTE_CLR(new, PT_MOD | PT_REF); 3888*7c478bd9Sstevel@tonic-gate } 3889*7c478bd9Sstevel@tonic-gate 3890*7c478bd9Sstevel@tonic-gate replaced = x86pte_update(ht, entry, expected, new); 3891*7c478bd9Sstevel@tonic-gate if (replaced != expected) 3892*7c478bd9Sstevel@tonic-gate return (replaced); 3893*7c478bd9Sstevel@tonic-gate 3894*7c478bd9Sstevel@tonic-gate if (rm) { 3895*7c478bd9Sstevel@tonic-gate /* 3896*7c478bd9Sstevel@tonic-gate * sync to all constituent pages of a large page 3897*7c478bd9Sstevel@tonic-gate */ 3898*7c478bd9Sstevel@tonic-gate pgcnt_t pgcnt = page_get_pagecnt(ht->ht_level); 3899*7c478bd9Sstevel@tonic-gate ASSERT(IS_P2ALIGNED(pp->p_pagenum, pgcnt)); 3900*7c478bd9Sstevel@tonic-gate while (pgcnt-- > 0) { 3901*7c478bd9Sstevel@tonic-gate /* 3902*7c478bd9Sstevel@tonic-gate * hat_page_demote() can't decrease 3903*7c478bd9Sstevel@tonic-gate * pszc below this mapping size 3904*7c478bd9Sstevel@tonic-gate * since large mapping existed after we 3905*7c478bd9Sstevel@tonic-gate * took mlist lock. 3906*7c478bd9Sstevel@tonic-gate */ 3907*7c478bd9Sstevel@tonic-gate ASSERT(pp->p_szc >= ht->ht_level); 3908*7c478bd9Sstevel@tonic-gate hat_page_setattr(pp, rm); 3909*7c478bd9Sstevel@tonic-gate ++pp; 3910*7c478bd9Sstevel@tonic-gate } 3911*7c478bd9Sstevel@tonic-gate } 3912*7c478bd9Sstevel@tonic-gate 3913*7c478bd9Sstevel@tonic-gate return (0); 3914*7c478bd9Sstevel@tonic-gate } 3915*7c478bd9Sstevel@tonic-gate 3916*7c478bd9Sstevel@tonic-gate /* 3917*7c478bd9Sstevel@tonic-gate * Kernel Physical Mapping (kpm) facility 3918*7c478bd9Sstevel@tonic-gate * 3919*7c478bd9Sstevel@tonic-gate * Most of the routines needed to support segkpm are almost no-ops on the 3920*7c478bd9Sstevel@tonic-gate * x86 platform. We map in the entire segment when it is created and leave 3921*7c478bd9Sstevel@tonic-gate * it mapped in, so there is no additional work required to set up and tear 3922*7c478bd9Sstevel@tonic-gate * down individual mappings. All of these routines were created to support 3923*7c478bd9Sstevel@tonic-gate * SPARC platforms that have to avoid aliasing in their virtually indexed 3924*7c478bd9Sstevel@tonic-gate * caches. 3925*7c478bd9Sstevel@tonic-gate * 3926*7c478bd9Sstevel@tonic-gate * Most of the routines have sanity checks in them (e.g. verifying that the 3927*7c478bd9Sstevel@tonic-gate * passed-in page is locked). We don't actually care about most of these 3928*7c478bd9Sstevel@tonic-gate * checks on x86, but we leave them in place to identify problems in the 3929*7c478bd9Sstevel@tonic-gate * upper levels. 3930*7c478bd9Sstevel@tonic-gate */ 3931*7c478bd9Sstevel@tonic-gate 3932*7c478bd9Sstevel@tonic-gate /* 3933*7c478bd9Sstevel@tonic-gate * Map in a locked page and return the vaddr. 3934*7c478bd9Sstevel@tonic-gate */ 3935*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3936*7c478bd9Sstevel@tonic-gate caddr_t 3937*7c478bd9Sstevel@tonic-gate hat_kpm_mapin(struct page *pp, struct kpme *kpme) 3938*7c478bd9Sstevel@tonic-gate { 3939*7c478bd9Sstevel@tonic-gate caddr_t vaddr; 3940*7c478bd9Sstevel@tonic-gate 3941*7c478bd9Sstevel@tonic-gate #ifdef DEBUG 3942*7c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 3943*7c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: kpm_enable not set\n"); 3944*7c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 3945*7c478bd9Sstevel@tonic-gate } 3946*7c478bd9Sstevel@tonic-gate 3947*7c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 3948*7c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapin: pp zero or not locked\n"); 3949*7c478bd9Sstevel@tonic-gate return ((caddr_t)NULL); 3950*7c478bd9Sstevel@tonic-gate } 3951*7c478bd9Sstevel@tonic-gate #endif 3952*7c478bd9Sstevel@tonic-gate 3953*7c478bd9Sstevel@tonic-gate vaddr = hat_kpm_page2va(pp, 1); 3954*7c478bd9Sstevel@tonic-gate 3955*7c478bd9Sstevel@tonic-gate return (vaddr); 3956*7c478bd9Sstevel@tonic-gate } 3957*7c478bd9Sstevel@tonic-gate 3958*7c478bd9Sstevel@tonic-gate /* 3959*7c478bd9Sstevel@tonic-gate * Mapout a locked page. 3960*7c478bd9Sstevel@tonic-gate */ 3961*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 3962*7c478bd9Sstevel@tonic-gate void 3963*7c478bd9Sstevel@tonic-gate hat_kpm_mapout(struct page *pp, struct kpme *kpme, caddr_t vaddr) 3964*7c478bd9Sstevel@tonic-gate { 3965*7c478bd9Sstevel@tonic-gate #ifdef DEBUG 3966*7c478bd9Sstevel@tonic-gate if (kpm_enable == 0) { 3967*7c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: kpm_enable not set\n"); 3968*7c478bd9Sstevel@tonic-gate return; 3969*7c478bd9Sstevel@tonic-gate } 3970*7c478bd9Sstevel@tonic-gate 3971*7c478bd9Sstevel@tonic-gate if (IS_KPM_ADDR(vaddr) == 0) { 3972*7c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: no kpm address\n"); 3973*7c478bd9Sstevel@tonic-gate return; 3974*7c478bd9Sstevel@tonic-gate } 3975*7c478bd9Sstevel@tonic-gate 3976*7c478bd9Sstevel@tonic-gate if (pp == NULL || PAGE_LOCKED(pp) == 0) { 3977*7c478bd9Sstevel@tonic-gate cmn_err(CE_WARN, "hat_kpm_mapout: page zero or not locked\n"); 3978*7c478bd9Sstevel@tonic-gate return; 3979*7c478bd9Sstevel@tonic-gate } 3980*7c478bd9Sstevel@tonic-gate #endif 3981*7c478bd9Sstevel@tonic-gate } 3982*7c478bd9Sstevel@tonic-gate 3983*7c478bd9Sstevel@tonic-gate /* 3984*7c478bd9Sstevel@tonic-gate * Return the kpm virtual address for a specific pfn 3985*7c478bd9Sstevel@tonic-gate */ 3986*7c478bd9Sstevel@tonic-gate caddr_t 3987*7c478bd9Sstevel@tonic-gate hat_kpm_pfn2va(pfn_t pfn) 3988*7c478bd9Sstevel@tonic-gate { 3989*7c478bd9Sstevel@tonic-gate uintptr_t vaddr; 3990*7c478bd9Sstevel@tonic-gate 3991*7c478bd9Sstevel@tonic-gate ASSERT(kpm_enable); 3992*7c478bd9Sstevel@tonic-gate 3993*7c478bd9Sstevel@tonic-gate vaddr = (uintptr_t)kpm_vbase + mmu_ptob(pfn); 3994*7c478bd9Sstevel@tonic-gate 3995*7c478bd9Sstevel@tonic-gate return ((caddr_t)vaddr); 3996*7c478bd9Sstevel@tonic-gate } 3997*7c478bd9Sstevel@tonic-gate 3998*7c478bd9Sstevel@tonic-gate /* 3999*7c478bd9Sstevel@tonic-gate * Return the kpm virtual address for the page at pp. 4000*7c478bd9Sstevel@tonic-gate */ 4001*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 4002*7c478bd9Sstevel@tonic-gate caddr_t 4003*7c478bd9Sstevel@tonic-gate hat_kpm_page2va(struct page *pp, int checkswap) 4004*7c478bd9Sstevel@tonic-gate { 4005*7c478bd9Sstevel@tonic-gate return (hat_kpm_pfn2va(pp->p_pagenum)); 4006*7c478bd9Sstevel@tonic-gate } 4007*7c478bd9Sstevel@tonic-gate 4008*7c478bd9Sstevel@tonic-gate /* 4009*7c478bd9Sstevel@tonic-gate * Return the page frame number for the kpm virtual address vaddr. 4010*7c478bd9Sstevel@tonic-gate */ 4011*7c478bd9Sstevel@tonic-gate pfn_t 4012*7c478bd9Sstevel@tonic-gate hat_kpm_va2pfn(caddr_t vaddr) 4013*7c478bd9Sstevel@tonic-gate { 4014*7c478bd9Sstevel@tonic-gate pfn_t pfn; 4015*7c478bd9Sstevel@tonic-gate 4016*7c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 4017*7c478bd9Sstevel@tonic-gate 4018*7c478bd9Sstevel@tonic-gate pfn = (pfn_t)btop(vaddr - kpm_vbase); 4019*7c478bd9Sstevel@tonic-gate 4020*7c478bd9Sstevel@tonic-gate return (pfn); 4021*7c478bd9Sstevel@tonic-gate } 4022*7c478bd9Sstevel@tonic-gate 4023*7c478bd9Sstevel@tonic-gate 4024*7c478bd9Sstevel@tonic-gate /* 4025*7c478bd9Sstevel@tonic-gate * Return the page for the kpm virtual address vaddr. 4026*7c478bd9Sstevel@tonic-gate */ 4027*7c478bd9Sstevel@tonic-gate page_t * 4028*7c478bd9Sstevel@tonic-gate hat_kpm_vaddr2page(caddr_t vaddr) 4029*7c478bd9Sstevel@tonic-gate { 4030*7c478bd9Sstevel@tonic-gate pfn_t pfn; 4031*7c478bd9Sstevel@tonic-gate 4032*7c478bd9Sstevel@tonic-gate ASSERT(IS_KPM_ADDR(vaddr)); 4033*7c478bd9Sstevel@tonic-gate 4034*7c478bd9Sstevel@tonic-gate pfn = hat_kpm_va2pfn(vaddr); 4035*7c478bd9Sstevel@tonic-gate 4036*7c478bd9Sstevel@tonic-gate return (page_numtopp_nolock(pfn)); 4037*7c478bd9Sstevel@tonic-gate } 4038*7c478bd9Sstevel@tonic-gate 4039*7c478bd9Sstevel@tonic-gate /* 4040*7c478bd9Sstevel@tonic-gate * hat_kpm_fault is called from segkpm_fault when we take a page fault on a 4041*7c478bd9Sstevel@tonic-gate * KPM page. This should never happen on x86 4042*7c478bd9Sstevel@tonic-gate */ 4043*7c478bd9Sstevel@tonic-gate int 4044*7c478bd9Sstevel@tonic-gate hat_kpm_fault(hat_t *hat, caddr_t vaddr) 4045*7c478bd9Sstevel@tonic-gate { 4046*7c478bd9Sstevel@tonic-gate panic("pagefault in seg_kpm. hat: 0x%p vaddr: 0x%p", hat, vaddr); 4047*7c478bd9Sstevel@tonic-gate 4048*7c478bd9Sstevel@tonic-gate return (0); 4049*7c478bd9Sstevel@tonic-gate } 4050*7c478bd9Sstevel@tonic-gate 4051*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 4052*7c478bd9Sstevel@tonic-gate void 4053*7c478bd9Sstevel@tonic-gate hat_kpm_mseghash_clear(int nentries) 4054*7c478bd9Sstevel@tonic-gate {} 4055*7c478bd9Sstevel@tonic-gate 4056*7c478bd9Sstevel@tonic-gate /*ARGSUSED*/ 4057*7c478bd9Sstevel@tonic-gate void 4058*7c478bd9Sstevel@tonic-gate hat_kpm_mseghash_update(pgcnt_t inx, struct memseg *msp) 4059*7c478bd9Sstevel@tonic-gate {} 4060