1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */ 27 /* All Rights Reserved */ 28 29 /* 30 * Portions of this source code were derived from Berkeley 4.3 BSD 31 * under license from the Regents of the University of California. 32 */ 33 34 #pragma ident "%Z%%M% %I% %E% SMI" 35 36 /* 37 * UNIX machine dependent virtual memory support. 38 */ 39 40 #include <sys/vm.h> 41 #include <sys/exec.h> 42 #include <sys/cmn_err.h> 43 #include <sys/cpu_module.h> 44 #include <sys/cpu.h> 45 #include <sys/elf_SPARC.h> 46 #include <sys/archsystm.h> 47 #include <vm/hat_sfmmu.h> 48 #include <sys/memnode.h> 49 #include <sys/mem_cage.h> 50 #include <vm/vm_dep.h> 51 52 #if defined(__sparcv9) && defined(SF_ERRATA_57) 53 caddr_t errata57_limit; 54 #endif 55 56 uint_t page_colors = 0; 57 uint_t page_colors_mask = 0; 58 uint_t page_coloring_shift = 0; 59 int consistent_coloring; 60 61 uint_t mmu_page_sizes = DEFAULT_MMU_PAGE_SIZES; 62 uint_t max_mmu_page_sizes = MMU_PAGE_SIZES; 63 uint_t mmu_hashcnt = DEFAULT_MAX_HASHCNT; 64 uint_t max_mmu_hashcnt = MAX_HASHCNT; 65 size_t mmu_ism_pagesize = DEFAULT_ISM_PAGESIZE; 66 67 /* 68 * The sun4u hardware mapping sizes which will always be supported are 69 * 8K, 64K, 512K and 4M. If sun4u based machines need to support other 70 * page sizes, platform or cpu specific routines need to modify the value. 71 * The base pagesize (p_szc == 0) must always be supported by the hardware. 72 */ 73 int mmu_exported_pagesize_mask = (1 << TTE8K) | (1 << TTE64K) | 74 (1 << TTE512K) | (1 << TTE4M); 75 uint_t mmu_exported_page_sizes; 76 77 uint_t szc_2_userszc[MMU_PAGE_SIZES]; 78 uint_t userszc_2_szc[MMU_PAGE_SIZES]; 79 80 extern uint_t vac_colors_mask; 81 extern int vac_shift; 82 83 hw_pagesize_t hw_page_array[] = { 84 {MMU_PAGESIZE, MMU_PAGESHIFT, 0, MMU_PAGESIZE >> MMU_PAGESHIFT}, 85 {MMU_PAGESIZE64K, MMU_PAGESHIFT64K, 0, 86 MMU_PAGESIZE64K >> MMU_PAGESHIFT}, 87 {MMU_PAGESIZE512K, MMU_PAGESHIFT512K, 0, 88 MMU_PAGESIZE512K >> MMU_PAGESHIFT}, 89 {MMU_PAGESIZE4M, MMU_PAGESHIFT4M, 0, MMU_PAGESIZE4M >> MMU_PAGESHIFT}, 90 {MMU_PAGESIZE32M, MMU_PAGESHIFT32M, 0, 91 MMU_PAGESIZE32M >> MMU_PAGESHIFT}, 92 {MMU_PAGESIZE256M, MMU_PAGESHIFT256M, 0, 93 MMU_PAGESIZE256M >> MMU_PAGESHIFT}, 94 {0, 0, 0, 0} 95 }; 96 97 /* 98 * use_text_pgsz64k and use_text_pgsz512k allow the user to turn on these 99 * additional text page sizes for USIII-IV+ and OPL by changing the default 100 * values via /etc/system. 101 */ 102 int use_text_pgsz64K = 0; 103 int use_text_pgsz512K = 0; 104 105 /* 106 * Maximum and default segment size tunables for user heap, stack, private 107 * and shared anonymous memory, and user text and initialized data. 108 */ 109 size_t max_uheap_lpsize = MMU_PAGESIZE4M; 110 size_t default_uheap_lpsize = MMU_PAGESIZE; 111 size_t max_ustack_lpsize = MMU_PAGESIZE4M; 112 size_t default_ustack_lpsize = MMU_PAGESIZE; 113 size_t max_privmap_lpsize = MMU_PAGESIZE4M; 114 size_t max_uidata_lpsize = MMU_PAGESIZE; 115 size_t max_utext_lpsize = MMU_PAGESIZE4M; 116 size_t max_shm_lpsize = MMU_PAGESIZE4M; 117 118 void 119 adjust_data_maxlpsize(size_t ismpagesize) 120 { 121 if (max_uheap_lpsize == MMU_PAGESIZE4M) { 122 max_uheap_lpsize = ismpagesize; 123 } 124 if (max_ustack_lpsize == MMU_PAGESIZE4M) { 125 max_ustack_lpsize = ismpagesize; 126 } 127 if (max_privmap_lpsize == MMU_PAGESIZE4M) { 128 max_privmap_lpsize = ismpagesize; 129 } 130 if (max_shm_lpsize == MMU_PAGESIZE4M) { 131 max_shm_lpsize = ismpagesize; 132 } 133 } 134 135 /* 136 * map_addr_proc() is the routine called when the system is to 137 * choose an address for the user. We will pick an address 138 * range which is just below the current stack limit. The 139 * algorithm used for cache consistency on machines with virtual 140 * address caches is such that offset 0 in the vnode is always 141 * on a shm_alignment'ed aligned address. Unfortunately, this 142 * means that vnodes which are demand paged will not be mapped 143 * cache consistently with the executable images. When the 144 * cache alignment for a given object is inconsistent, the 145 * lower level code must manage the translations so that this 146 * is not seen here (at the cost of efficiency, of course). 147 * 148 * addrp is a value/result parameter. 149 * On input it is a hint from the user to be used in a completely 150 * machine dependent fashion. For MAP_ALIGN, addrp contains the 151 * minimal alignment. 152 * 153 * On output it is NULL if no address can be found in the current 154 * processes address space or else an address that is currently 155 * not mapped for len bytes with a page of red zone on either side. 156 * If vacalign is true, then the selected address will obey the alignment 157 * constraints of a vac machine based on the given off value. 158 */ 159 /*ARGSUSED4*/ 160 void 161 map_addr_proc(caddr_t *addrp, size_t len, offset_t off, int vacalign, 162 caddr_t userlimit, struct proc *p, uint_t flags) 163 { 164 struct as *as = p->p_as; 165 caddr_t addr; 166 caddr_t base; 167 size_t slen; 168 uintptr_t align_amount; 169 int allow_largepage_alignment = 1; 170 171 base = p->p_brkbase; 172 if (userlimit < as->a_userlimit) { 173 /* 174 * This happens when a program wants to map something in 175 * a range that's accessible to a program in a smaller 176 * address space. For example, a 64-bit program might 177 * be calling mmap32(2) to guarantee that the returned 178 * address is below 4Gbytes. 179 */ 180 ASSERT(userlimit > base); 181 slen = userlimit - base; 182 } else { 183 slen = p->p_usrstack - base - (((size_t)rctl_enforced_value( 184 rctlproc_legacy[RLIMIT_STACK], p->p_rctls, p) + PAGEOFFSET) 185 & PAGEMASK); 186 } 187 len = (len + PAGEOFFSET) & PAGEMASK; 188 189 /* 190 * Redzone for each side of the request. This is done to leave 191 * one page unmapped between segments. This is not required, but 192 * it's useful for the user because if their program strays across 193 * a segment boundary, it will catch a fault immediately making 194 * debugging a little easier. 195 */ 196 len += (2 * PAGESIZE); 197 198 /* 199 * If the request is larger than the size of a particular 200 * mmu level, then we use that level to map the request. 201 * But this requires that both the virtual and the physical 202 * addresses be aligned with respect to that level, so we 203 * do the virtual bit of nastiness here. 204 * 205 * For 32-bit processes, only those which have specified 206 * MAP_ALIGN or an addr will be aligned on a page size > 4MB. Otherwise 207 * we can potentially waste up to 256MB of the 4G process address 208 * space just for alignment. 209 */ 210 if (p->p_model == DATAMODEL_ILP32 && ((flags & MAP_ALIGN) == 0 || 211 ((uintptr_t)*addrp) != 0)) { 212 allow_largepage_alignment = 0; 213 } 214 if ((mmu_page_sizes == max_mmu_page_sizes) && 215 allow_largepage_alignment && 216 (len >= MMU_PAGESIZE256M)) { /* 256MB mappings */ 217 align_amount = MMU_PAGESIZE256M; 218 } else if ((mmu_page_sizes == max_mmu_page_sizes) && 219 allow_largepage_alignment && 220 (len >= MMU_PAGESIZE32M)) { /* 32MB mappings */ 221 align_amount = MMU_PAGESIZE32M; 222 } else if (len >= MMU_PAGESIZE4M) { /* 4MB mappings */ 223 align_amount = MMU_PAGESIZE4M; 224 } else if (len >= MMU_PAGESIZE512K) { /* 512KB mappings */ 225 align_amount = MMU_PAGESIZE512K; 226 } else if (len >= MMU_PAGESIZE64K) { /* 64KB mappings */ 227 align_amount = MMU_PAGESIZE64K; 228 } else { 229 /* 230 * Align virtual addresses on a 64K boundary to ensure 231 * that ELF shared libraries are mapped with the appropriate 232 * alignment constraints by the run-time linker. 233 */ 234 align_amount = ELF_SPARC_MAXPGSZ; 235 if ((flags & MAP_ALIGN) && ((uintptr_t)*addrp != 0) && 236 ((uintptr_t)*addrp < align_amount)) 237 align_amount = (uintptr_t)*addrp; 238 } 239 240 /* 241 * 64-bit processes require 1024K alignment of ELF shared libraries. 242 */ 243 if (p->p_model == DATAMODEL_LP64) 244 align_amount = MAX(align_amount, ELF_SPARCV9_MAXPGSZ); 245 #ifdef VAC 246 if (vac && vacalign && (align_amount < shm_alignment)) 247 align_amount = shm_alignment; 248 #endif 249 250 if ((flags & MAP_ALIGN) && ((uintptr_t)*addrp > align_amount)) { 251 align_amount = (uintptr_t)*addrp; 252 } 253 len += align_amount; 254 255 /* 256 * Look for a large enough hole starting below the stack limit. 257 * After finding it, use the upper part. Addition of PAGESIZE is 258 * for the redzone as described above. 259 */ 260 as_purge(as); 261 if (as_gap(as, len, &base, &slen, AH_HI, NULL) == 0) { 262 caddr_t as_addr; 263 264 addr = base + slen - len + PAGESIZE; 265 as_addr = addr; 266 /* 267 * Round address DOWN to the alignment amount, 268 * add the offset, and if this address is less 269 * than the original address, add alignment amount. 270 */ 271 addr = (caddr_t)((uintptr_t)addr & (~(align_amount - 1l))); 272 addr += (long)(off & (align_amount - 1l)); 273 if (addr < as_addr) { 274 addr += align_amount; 275 } 276 277 ASSERT(addr <= (as_addr + align_amount)); 278 ASSERT(((uintptr_t)addr & (align_amount - 1l)) == 279 ((uintptr_t)(off & (align_amount - 1l)))); 280 *addrp = addr; 281 282 #if defined(SF_ERRATA_57) 283 if (AS_TYPE_64BIT(as) && addr < errata57_limit) { 284 *addrp = NULL; 285 } 286 #endif 287 } else { 288 *addrp = NULL; /* no more virtual space */ 289 } 290 } 291 292 /* 293 * Platform-dependent page scrub call. 294 */ 295 void 296 pagescrub(page_t *pp, uint_t off, uint_t len) 297 { 298 /* 299 * For now, we rely on the fact that pagezero() will 300 * always clear UEs. 301 */ 302 pagezero(pp, off, len); 303 } 304 305 /*ARGSUSED*/ 306 void 307 sync_data_memory(caddr_t va, size_t len) 308 { 309 cpu_flush_ecache(); 310 } 311 312 /* 313 * platform specific large pages for kernel heap support 314 */ 315 void 316 mmu_init_kcontext() 317 { 318 extern void set_kcontextreg(); 319 320 if (kcontextreg) 321 set_kcontextreg(); 322 } 323 324 void 325 contig_mem_init(void) 326 { 327 /* not applicable to sun4u */ 328 } 329 330 size_t 331 exec_get_spslew(void) 332 { 333 return (0); 334 } 335