| 1 | /* |
| 2 | * Copyright (c) 2007-2010 Apple Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * This file contains Original Code and/or Modifications of Original Code |
| 7 | * as defined in and that are subject to the Apple Public Source License |
| 8 | * Version 2.0 (the 'License'). You may not use this file except in |
| 9 | * compliance with the License. The rights granted to you under the License |
| 10 | * may not be used to create, or enable the creation or redistribution of, |
| 11 | * unlawful or unlicensed copies of an Apple operating system, or to |
| 12 | * circumvent, violate, or enable the circumvention or violation of, any |
| 13 | * terms of an Apple operating system software license agreement. |
| 14 | * |
| 15 | * Please obtain a copy of the License at |
| 16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
| 17 | * |
| 18 | * The Original Code and all software distributed under the License are |
| 19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
| 23 | * Please see the License for the specific language governing rights and |
| 24 | * limitations under the License. |
| 25 | * |
| 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
| 27 | */ |
| 28 | |
| 29 | #include <mach/machine.h> |
| 30 | #include <mach/processor.h> |
| 31 | #include <kern/kalloc.h> |
| 32 | #include <i386/cpu_affinity.h> |
| 33 | #include <i386/cpu_topology.h> |
| 34 | #include <i386/cpu_threads.h> |
| 35 | #include <i386/machine_cpu.h> |
| 36 | #include <i386/bit_routines.h> |
| 37 | #include <i386/cpu_data.h> |
| 38 | #include <i386/lapic.h> |
| 39 | #include <i386/machine_routines.h> |
| 40 | #include <stddef.h> |
| 41 | |
| 42 | __private_extern__ void qsort( |
| 43 | void * array, |
| 44 | size_t nmembers, |
| 45 | size_t member_size, |
| 46 | int (*)(const void *, const void *)); |
| 47 | |
| 48 | static int lapicid_cmp(const void *x, const void *y); |
| 49 | static x86_affinity_set_t *find_cache_affinity(x86_cpu_cache_t *L2_cachep); |
| 50 | |
| 51 | x86_affinity_set_t *x86_affinities = NULL; |
| 52 | static int x86_affinity_count = 0; |
| 53 | |
| 54 | extern cpu_data_t cpshadows[]; |
| 55 | /* Re-sort double-mapped CPU data shadows after topology discovery sorts the |
| 56 | * primary CPU data structures by physical/APIC CPU ID. |
| 57 | */ |
| 58 | static void cpu_shadow_sort(int ncpus) { |
| 59 | for (int i = 0; i < ncpus; i++) { |
| 60 | cpu_data_t *cpup = cpu_datap(i); |
| 61 | ptrdiff_t coff = cpup - cpu_datap(0); |
| 62 | |
| 63 | cpup->cd_shadow = &cpshadows[coff]; |
| 64 | } |
| 65 | } |
| 66 | |
| 67 | /* |
| 68 | * cpu_topology_sort() is called after all processors have been registered but |
| 69 | * before any non-boot processor id started. We establish canonical logical |
| 70 | * processor numbering - logical cpus must be contiguous, zero-based and |
| 71 | * assigned in physical (local apic id) order. This step is required because |
| 72 | * the discovery/registration order is non-deterministic - cores are registered |
| 73 | * in differing orders over boots. Enforcing canonical numbering simplifies |
| 74 | * identification of processors. |
| 75 | */ |
| 76 | void |
| 77 | cpu_topology_sort(int ncpus) |
| 78 | { |
| 79 | int i; |
| 80 | boolean_t istate; |
| 81 | processor_t lprim = NULL; |
| 82 | |
| 83 | assert(machine_info.physical_cpu == 1); |
| 84 | assert(machine_info.logical_cpu == 1); |
| 85 | assert(master_cpu == 0); |
| 86 | assert(cpu_number() == 0); |
| 87 | assert(cpu_datap(0)->cpu_number == 0); |
| 88 | |
| 89 | /* Lights out for this */ |
| 90 | istate = ml_set_interrupts_enabled(FALSE); |
| 91 | |
| 92 | if (topo_dbg) { |
| 93 | TOPO_DBG("cpu_topology_start() %d cpu%s registered\n" , |
| 94 | ncpus, (ncpus > 1) ? "s" : "" ); |
| 95 | for (i = 0; i < ncpus; i++) { |
| 96 | cpu_data_t *cpup = cpu_datap(i); |
| 97 | TOPO_DBG("\tcpu_data[%d]:%p local apic 0x%x\n" , |
| 98 | i, (void *) cpup, cpup->cpu_phys_number); |
| 99 | } |
| 100 | } |
| 101 | |
| 102 | /* |
| 103 | * Re-order the cpu_data_ptr vector sorting by physical id. |
| 104 | * Skip the boot processor, it's required to be correct. |
| 105 | */ |
| 106 | if (ncpus > 1) { |
| 107 | qsort((void *) &cpu_data_ptr[1], |
| 108 | ncpus - 1, |
| 109 | sizeof(cpu_data_t *), |
| 110 | lapicid_cmp); |
| 111 | } |
| 112 | if (topo_dbg) { |
| 113 | TOPO_DBG("cpu_topology_start() after sorting:\n" ); |
| 114 | for (i = 0; i < ncpus; i++) { |
| 115 | cpu_data_t *cpup = cpu_datap(i); |
| 116 | TOPO_DBG("\tcpu_data[%d]:%p local apic 0x%x\n" , |
| 117 | i, (void *) cpup, cpup->cpu_phys_number); |
| 118 | } |
| 119 | } |
| 120 | |
| 121 | /* |
| 122 | * Finalize logical numbers and map kept by the lapic code. |
| 123 | */ |
| 124 | for (i = 0; i < ncpus; i++) { |
| 125 | cpu_data_t *cpup = cpu_datap(i); |
| 126 | |
| 127 | if (cpup->cpu_number != i) { |
| 128 | kprintf("cpu_datap(%d):%p local apic id 0x%x " |
| 129 | "remapped from %d\n" , |
| 130 | i, cpup, cpup->cpu_phys_number, |
| 131 | cpup->cpu_number); |
| 132 | } |
| 133 | cpup->cpu_number = i; |
| 134 | lapic_cpu_map(cpup->cpu_phys_number, i); |
| 135 | x86_set_logical_topology(&cpup->lcpu, cpup->cpu_phys_number, i); |
| 136 | } |
| 137 | |
| 138 | cpu_shadow_sort(ncpus); |
| 139 | x86_validate_topology(); |
| 140 | |
| 141 | ml_set_interrupts_enabled(istate); |
| 142 | TOPO_DBG("cpu_topology_start() LLC is L%d\n" , topoParms.LLCDepth + 1); |
| 143 | |
| 144 | /* |
| 145 | * Let the CPU Power Management know that the topology is stable. |
| 146 | */ |
| 147 | topoParms.stable = TRUE; |
| 148 | pmCPUStateInit(); |
| 149 | |
| 150 | /* |
| 151 | * Iterate over all logical cpus finding or creating the affinity set |
| 152 | * for their LLC cache. Each affinity set possesses a processor set |
| 153 | * into which each logical processor is added. |
| 154 | */ |
| 155 | TOPO_DBG("cpu_topology_start() creating affinity sets:\n" ); |
| 156 | for (i = 0; i < ncpus; i++) { |
| 157 | cpu_data_t *cpup = cpu_datap(i); |
| 158 | x86_lcpu_t *lcpup = cpu_to_lcpu(i); |
| 159 | x86_cpu_cache_t *LLC_cachep; |
| 160 | x86_affinity_set_t *aset; |
| 161 | |
| 162 | LLC_cachep = lcpup->caches[topoParms.LLCDepth]; |
| 163 | assert(LLC_cachep->type == CPU_CACHE_TYPE_UNIF); |
| 164 | aset = find_cache_affinity(LLC_cachep); |
| 165 | if (aset == NULL) { |
| 166 | aset = (x86_affinity_set_t *) kalloc(sizeof(*aset)); |
| 167 | if (aset == NULL) |
| 168 | panic("cpu_topology_start() failed aset alloc" ); |
| 169 | aset->next = x86_affinities; |
| 170 | x86_affinities = aset; |
| 171 | aset->num = x86_affinity_count++; |
| 172 | aset->cache = LLC_cachep; |
| 173 | aset->pset = (i == master_cpu) ? |
| 174 | processor_pset(master_processor) : |
| 175 | pset_create(pset_node_root()); |
| 176 | if (aset->pset == PROCESSOR_SET_NULL) |
| 177 | panic("cpu_topology_start: pset_create" ); |
| 178 | TOPO_DBG("\tnew set %p(%d) pset %p for cache %p\n" , |
| 179 | aset, aset->num, aset->pset, aset->cache); |
| 180 | } |
| 181 | |
| 182 | TOPO_DBG("\tprocessor_init set %p(%d) lcpup %p(%d) cpu %p processor %p\n" , |
| 183 | aset, aset->num, lcpup, lcpup->cpu_num, cpup, cpup->cpu_processor); |
| 184 | |
| 185 | if (i != master_cpu) |
| 186 | processor_init(cpup->cpu_processor, i, aset->pset); |
| 187 | |
| 188 | if (lcpup->core->num_lcpus > 1) { |
| 189 | if (lcpup->lnum == 0) |
| 190 | lprim = cpup->cpu_processor; |
| 191 | |
| 192 | processor_set_primary(cpup->cpu_processor, lprim); |
| 193 | } |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | /* We got a request to start a CPU. Check that this CPU is within the |
| 198 | * max cpu limit set before we do. |
| 199 | */ |
| 200 | kern_return_t |
| 201 | cpu_topology_start_cpu( int cpunum ) |
| 202 | { |
| 203 | int ncpus = machine_info.max_cpus; |
| 204 | int i = cpunum; |
| 205 | |
| 206 | /* Decide whether to start a CPU, and actually start it */ |
| 207 | TOPO_DBG("cpu_topology_start() processor_start():\n" ); |
| 208 | if( i < ncpus) |
| 209 | { |
| 210 | TOPO_DBG("\tlcpu %d\n" , cpu_datap(i)->cpu_number); |
| 211 | processor_start(cpu_datap(i)->cpu_processor); |
| 212 | return KERN_SUCCESS; |
| 213 | } |
| 214 | else |
| 215 | return KERN_FAILURE; |
| 216 | } |
| 217 | |
| 218 | static int |
| 219 | lapicid_cmp(const void *x, const void *y) |
| 220 | { |
| 221 | cpu_data_t *cpu_x = *((cpu_data_t **)(uintptr_t)x); |
| 222 | cpu_data_t *cpu_y = *((cpu_data_t **)(uintptr_t)y); |
| 223 | |
| 224 | TOPO_DBG("lapicid_cmp(%p,%p) (%d,%d)\n" , |
| 225 | x, y, cpu_x->cpu_phys_number, cpu_y->cpu_phys_number); |
| 226 | if (cpu_x->cpu_phys_number < cpu_y->cpu_phys_number) |
| 227 | return -1; |
| 228 | if (cpu_x->cpu_phys_number == cpu_y->cpu_phys_number) |
| 229 | return 0; |
| 230 | return 1; |
| 231 | } |
| 232 | |
| 233 | static x86_affinity_set_t * |
| 234 | find_cache_affinity(x86_cpu_cache_t *l2_cachep) |
| 235 | { |
| 236 | x86_affinity_set_t *aset; |
| 237 | |
| 238 | for (aset = x86_affinities; aset != NULL; aset = aset->next) { |
| 239 | if (l2_cachep == aset->cache) |
| 240 | break; |
| 241 | } |
| 242 | return aset; |
| 243 | } |
| 244 | |
| 245 | int |
| 246 | ml_get_max_affinity_sets(void) |
| 247 | { |
| 248 | return x86_affinity_count; |
| 249 | } |
| 250 | |
| 251 | processor_set_t |
| 252 | ml_affinity_to_pset(uint32_t affinity_num) |
| 253 | { |
| 254 | x86_affinity_set_t *aset; |
| 255 | |
| 256 | for (aset = x86_affinities; aset != NULL; aset = aset->next) { |
| 257 | if (affinity_num == aset->num) |
| 258 | break; |
| 259 | } |
| 260 | return (aset == NULL) ? PROCESSOR_SET_NULL : aset->pset; |
| 261 | } |
| 262 | |
| 263 | uint64_t |
| 264 | ml_cpu_cache_size(unsigned int level) |
| 265 | { |
| 266 | x86_cpu_cache_t *cachep; |
| 267 | |
| 268 | if (level == 0) { |
| 269 | return machine_info.max_mem; |
| 270 | } else if ( 1 <= level && level <= MAX_CACHE_DEPTH) { |
| 271 | cachep = current_cpu_datap()->lcpu.caches[level-1]; |
| 272 | return cachep ? cachep->cache_size : 0; |
| 273 | } else { |
| 274 | return 0; |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | uint64_t |
| 279 | ml_cpu_cache_sharing(unsigned int level) |
| 280 | { |
| 281 | x86_cpu_cache_t *cachep; |
| 282 | |
| 283 | if (level == 0) { |
| 284 | return machine_info.max_cpus; |
| 285 | } else if ( 1 <= level && level <= MAX_CACHE_DEPTH) { |
| 286 | cachep = current_cpu_datap()->lcpu.caches[level-1]; |
| 287 | return cachep ? cachep->nlcpus : 0; |
| 288 | } else { |
| 289 | return 0; |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | |