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 2010 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/systm.h> 27 #include <sys/platform_module.h> 28 #include <sys/sysmacros.h> 29 #include <sys/atomic.h> 30 #include <sys/memlist.h> 31 #include <sys/memnode.h> 32 #include <vm/vm_dep.h> 33 34 int max_mem_nodes = 1; /* max memory nodes on this system */ 35 36 struct mem_node_conf mem_node_config[MAX_MEM_NODES]; 37 int mem_node_pfn_shift; 38 /* 39 * num_memnodes should be updated atomically and always >= 40 * the number of bits in memnodes_mask or the algorithm may fail. 41 */ 42 uint16_t num_memnodes; 43 mnodeset_t memnodes_mask; /* assumes 8*(sizeof(mnodeset_t)) >= MAX_MEM_NODES */ 44 45 /* 46 * If set, mem_node_physalign should be a power of two, and 47 * should reflect the minimum address alignment of each node. 48 */ 49 uint64_t mem_node_physalign; 50 51 /* 52 * Platform hooks we will need. 53 */ 54 55 #pragma weak plat_build_mem_nodes 56 #pragma weak plat_slice_add 57 #pragma weak plat_slice_del 58 59 /* 60 * Adjust the memnode config after a DR operation. 61 * 62 * It is rather tricky to do these updates since we can't 63 * protect the memnode structures with locks, so we must 64 * be mindful of the order in which updates and reads to 65 * these values can occur. 66 */ 67 void 68 mem_node_add_slice(pfn_t start, pfn_t end) 69 { 70 int mnode; 71 mnodeset_t newmask, oldmask; 72 73 /* 74 * DR will pass us the first pfn that is allocatable. 75 * We need to round down to get the real start of 76 * the slice. 77 */ 78 if (mem_node_physalign) { 79 start &= ~(btop(mem_node_physalign) - 1); 80 end = roundup(end, btop(mem_node_physalign)) - 1; 81 } 82 83 mnode = PFN_2_MEM_NODE(start); 84 ASSERT(mnode < max_mem_nodes); 85 86 if (cas32((uint32_t *)&mem_node_config[mnode].exists, 0, 1)) { 87 /* 88 * Add slice to existing node. 89 */ 90 if (start < mem_node_config[mnode].physbase) 91 mem_node_config[mnode].physbase = start; 92 if (end > mem_node_config[mnode].physmax) 93 mem_node_config[mnode].physmax = end; 94 } else { 95 mem_node_config[mnode].physbase = start; 96 mem_node_config[mnode].physmax = end; 97 atomic_add_16(&num_memnodes, 1); 98 do { 99 oldmask = memnodes_mask; 100 newmask = memnodes_mask | (1ull << mnode); 101 } while (cas64(&memnodes_mask, oldmask, newmask) != oldmask); 102 } 103 /* 104 * Let the common lgrp framework know about the new memory 105 */ 106 lgrp_config(LGRP_CONFIG_MEM_ADD, mnode, MEM_NODE_2_LGRPHAND(mnode)); 107 } 108 109 /* 110 * Remove a PFN range from a memnode. On some platforms, 111 * the memnode will be created with physbase at the first 112 * allocatable PFN, but later deleted with the MC slice 113 * base address converted to a PFN, in which case we need 114 * to assume physbase and up. 115 */ 116 void 117 mem_node_del_slice(pfn_t start, pfn_t end) 118 { 119 int mnode; 120 pgcnt_t delta_pgcnt, node_size; 121 mnodeset_t omask, nmask; 122 123 if (mem_node_physalign) { 124 start &= ~(btop(mem_node_physalign) - 1); 125 end = roundup(end, btop(mem_node_physalign)) - 1; 126 } 127 mnode = PFN_2_MEM_NODE(start); 128 129 ASSERT(mnode < max_mem_nodes); 130 ASSERT(mem_node_config[mnode].exists == 1); 131 132 delta_pgcnt = end - start; 133 node_size = mem_node_config[mnode].physmax - 134 mem_node_config[mnode].physbase; 135 136 if (node_size > delta_pgcnt) { 137 /* 138 * Subtract the slice from the memnode. 139 */ 140 if (start <= mem_node_config[mnode].physbase) 141 mem_node_config[mnode].physbase = end + 1; 142 ASSERT(end <= mem_node_config[mnode].physmax); 143 if (end == mem_node_config[mnode].physmax) 144 mem_node_config[mnode].physmax = start - 1; 145 } else { 146 147 /* 148 * Let the common lgrp framework know the mnode is 149 * leaving 150 */ 151 lgrp_config(LGRP_CONFIG_MEM_DEL, mnode, 152 MEM_NODE_2_LGRPHAND(mnode)); 153 154 /* 155 * Delete the whole node. 156 */ 157 ASSERT(MNODE_PGCNT(mnode) == 0); 158 do { 159 omask = memnodes_mask; 160 nmask = omask & ~(1ull << mnode); 161 } while (cas64(&memnodes_mask, omask, nmask) != omask); 162 atomic_add_16(&num_memnodes, -1); 163 mem_node_config[mnode].exists = 0; 164 } 165 } 166 167 void 168 mem_node_add_range(pfn_t start, pfn_t end) 169 { 170 if (&plat_slice_add != NULL) 171 plat_slice_add(start, end); 172 else 173 mem_node_add_slice(start, end); 174 } 175 176 void 177 mem_node_del_range(pfn_t start, pfn_t end) 178 { 179 if (&plat_slice_del != NULL) 180 plat_slice_del(start, end); 181 else 182 mem_node_del_slice(start, end); 183 } 184 185 void 186 startup_build_mem_nodes(prom_memlist_t *list, size_t nelems) 187 { 188 size_t elem; 189 pfn_t basepfn; 190 pgcnt_t npgs; 191 192 /* LINTED: ASSERT will always true or false */ 193 ASSERT(NBBY * sizeof (mnodeset_t) >= max_mem_nodes); 194 195 if (&plat_build_mem_nodes != NULL) { 196 plat_build_mem_nodes(list, nelems); 197 } else { 198 /* 199 * Boot install lists are arranged <addr, len>, ... 200 */ 201 for (elem = 0; elem < nelems; list++, elem++) { 202 basepfn = btop(list->addr); 203 npgs = btop(list->size); 204 mem_node_add_range(basepfn, basepfn + npgs - 1); 205 } 206 } 207 } 208 209 /* 210 * Allocate an unassigned memnode. 211 */ 212 int 213 mem_node_alloc() 214 { 215 int mnode; 216 mnodeset_t newmask, oldmask; 217 218 /* 219 * Find an unused memnode. Update it atomically to prevent 220 * a first time memnode creation race. 221 */ 222 for (mnode = 0; mnode < max_mem_nodes; mnode++) 223 if (cas32((uint32_t *)&mem_node_config[mnode].exists, 224 0, 1) == 0) 225 break; 226 227 if (mnode >= max_mem_nodes) 228 panic("Out of free memnodes\n"); 229 230 mem_node_config[mnode].physbase = (uint64_t)-1; 231 mem_node_config[mnode].physmax = 0; 232 atomic_add_16(&num_memnodes, 1); 233 do { 234 oldmask = memnodes_mask; 235 newmask = memnodes_mask | (1ull << mnode); 236 } while (cas64(&memnodes_mask, oldmask, newmask) != oldmask); 237 238 return (mnode); 239 } 240 241 /* 242 * Find the intersection between a memnode and a memlist 243 * and returns the number of pages that overlap. 244 * 245 * Grab the memlist lock to protect the list from DR operations. 246 */ 247 pgcnt_t 248 mem_node_memlist_pages(int mnode, struct memlist *mlist) 249 { 250 pfn_t base, end; 251 pfn_t cur_base, cur_end; 252 pgcnt_t npgs = 0; 253 pgcnt_t pages; 254 struct memlist *pmem; 255 256 if (&plat_mem_node_intersect_range != NULL) { 257 memlist_read_lock(); 258 259 for (pmem = mlist; pmem; pmem = pmem->ml_next) { 260 plat_mem_node_intersect_range(btop(pmem->ml_address), 261 btop(pmem->ml_size), mnode, &pages); 262 npgs += pages; 263 } 264 265 memlist_read_unlock(); 266 return (npgs); 267 } 268 269 base = mem_node_config[mnode].physbase; 270 end = mem_node_config[mnode].physmax; 271 272 memlist_read_lock(); 273 274 for (pmem = mlist; pmem; pmem = pmem->ml_next) { 275 cur_base = btop(pmem->ml_address); 276 cur_end = cur_base + btop(pmem->ml_size) - 1; 277 if (end < cur_base || base > cur_end) 278 continue; 279 npgs = npgs + (MIN(cur_end, end) - 280 MAX(cur_base, base)) + 1; 281 } 282 283 memlist_read_unlock(); 284 285 return (npgs); 286 } 287 288 /* 289 * Find MIN(physbase) and MAX(physmax) over all mnodes 290 * 291 * Called during startup and DR to find hpm_counters limits when 292 * interleaved_mnodes is set. 293 * NOTE: there is a race condition with DR if it tries to change more than 294 * one mnode in parallel. Sizing shared hpm_counters depends on finding the 295 * min(physbase) and max(physmax) across all mnodes. Therefore, the caller of 296 * page_ctrs_adjust must ensure that mem_node_config does not change while it 297 * is running. 298 */ 299 void 300 mem_node_max_range(pfn_t *basep, pfn_t *maxp) 301 { 302 int mnode; 303 pfn_t max = 0; 304 pfn_t base = (pfn_t)-1; 305 306 for (mnode = 0; mnode < max_mem_nodes; mnode++) { 307 if (mem_node_config[mnode].exists == 0) 308 continue; 309 if (max < mem_node_config[mnode].physmax) 310 max = mem_node_config[mnode].physmax; 311 if (base > mem_node_config[mnode].physbase) 312 base = mem_node_config[mnode].physbase; 313 } 314 ASSERT(base != (pfn_t)-1 && max != 0); 315 *basep = base; 316 *maxp = max; 317 } 318