| 1 | /* $NetBSD: subr_percpu.c,v 1.17 2014/11/27 15:00:00 uebayasi Exp $ */ |
| 2 | |
| 3 | /*- |
| 4 | * Copyright (c)2007,2008 YAMAMOTO Takashi, |
| 5 | * All rights reserved. |
| 6 | * |
| 7 | * Redistribution and use in source and binary forms, with or without |
| 8 | * modification, are permitted provided that the following conditions |
| 9 | * are met: |
| 10 | * 1. Redistributions of source code must retain the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer. |
| 12 | * 2. Redistributions in binary form must reproduce the above copyright |
| 13 | * notice, this list of conditions and the following disclaimer in the |
| 14 | * documentation and/or other materials provided with the distribution. |
| 15 | * |
| 16 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND |
| 17 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 18 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 19 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
| 20 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 21 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 22 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 23 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 24 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 25 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 26 | * SUCH DAMAGE. |
| 27 | */ |
| 28 | |
| 29 | /* |
| 30 | * per-cpu storage. |
| 31 | */ |
| 32 | |
| 33 | #include <sys/cdefs.h> |
| 34 | __KERNEL_RCSID(0, "$NetBSD: subr_percpu.c,v 1.17 2014/11/27 15:00:00 uebayasi Exp $" ); |
| 35 | |
| 36 | #include <sys/param.h> |
| 37 | #include <sys/cpu.h> |
| 38 | #include <sys/kmem.h> |
| 39 | #include <sys/kernel.h> |
| 40 | #include <sys/mutex.h> |
| 41 | #include <sys/percpu.h> |
| 42 | #include <sys/rwlock.h> |
| 43 | #include <sys/vmem.h> |
| 44 | #include <sys/xcall.h> |
| 45 | |
| 46 | #define PERCPU_QUANTUM_SIZE (ALIGNBYTES + 1) |
| 47 | #define PERCPU_QCACHE_MAX 0 |
| 48 | #define PERCPU_IMPORT_SIZE 2048 |
| 49 | |
| 50 | #if defined(DIAGNOSTIC) |
| 51 | #define MAGIC 0x50435055 /* "PCPU" */ |
| 52 | #define percpu_encrypt(pc) ((pc) ^ MAGIC) |
| 53 | #define percpu_decrypt(pc) ((pc) ^ MAGIC) |
| 54 | #else /* defined(DIAGNOSTIC) */ |
| 55 | #define percpu_encrypt(pc) (pc) |
| 56 | #define percpu_decrypt(pc) (pc) |
| 57 | #endif /* defined(DIAGNOSTIC) */ |
| 58 | |
| 59 | static krwlock_t percpu_swap_lock __cacheline_aligned; |
| 60 | static kmutex_t percpu_allocation_lock __cacheline_aligned; |
| 61 | static vmem_t * percpu_offset_arena __cacheline_aligned; |
| 62 | static unsigned int percpu_nextoff __cacheline_aligned; |
| 63 | |
| 64 | static percpu_cpu_t * |
| 65 | cpu_percpu(struct cpu_info *ci) |
| 66 | { |
| 67 | |
| 68 | return &ci->ci_data.cpu_percpu; |
| 69 | } |
| 70 | |
| 71 | static unsigned int |
| 72 | percpu_offset(percpu_t *pc) |
| 73 | { |
| 74 | const unsigned int off = percpu_decrypt((uintptr_t)pc); |
| 75 | |
| 76 | KASSERT(off < percpu_nextoff); |
| 77 | return off; |
| 78 | } |
| 79 | |
| 80 | /* |
| 81 | * percpu_cpu_swap: crosscall handler for percpu_cpu_enlarge |
| 82 | */ |
| 83 | |
| 84 | static void |
| 85 | percpu_cpu_swap(void *p1, void *p2) |
| 86 | { |
| 87 | struct cpu_info * const ci = p1; |
| 88 | percpu_cpu_t * const newpcc = p2; |
| 89 | percpu_cpu_t * const pcc = cpu_percpu(ci); |
| 90 | |
| 91 | KASSERT(ci == curcpu() || !mp_online); |
| 92 | |
| 93 | /* |
| 94 | * swap *pcc and *newpcc unless anyone has beaten us. |
| 95 | */ |
| 96 | rw_enter(&percpu_swap_lock, RW_WRITER); |
| 97 | if (newpcc->pcc_size > pcc->pcc_size) { |
| 98 | percpu_cpu_t tmp; |
| 99 | int s; |
| 100 | |
| 101 | tmp = *pcc; |
| 102 | |
| 103 | /* |
| 104 | * block interrupts so that we don't lose their modifications. |
| 105 | */ |
| 106 | |
| 107 | s = splhigh(); |
| 108 | |
| 109 | /* |
| 110 | * copy data to new storage. |
| 111 | */ |
| 112 | |
| 113 | memcpy(newpcc->pcc_data, pcc->pcc_data, pcc->pcc_size); |
| 114 | |
| 115 | /* |
| 116 | * this assignment needs to be atomic for percpu_getptr_remote. |
| 117 | */ |
| 118 | |
| 119 | pcc->pcc_data = newpcc->pcc_data; |
| 120 | |
| 121 | splx(s); |
| 122 | |
| 123 | pcc->pcc_size = newpcc->pcc_size; |
| 124 | *newpcc = tmp; |
| 125 | } |
| 126 | rw_exit(&percpu_swap_lock); |
| 127 | } |
| 128 | |
| 129 | /* |
| 130 | * percpu_cpu_enlarge: ensure that percpu_cpu_t of each cpus have enough space |
| 131 | */ |
| 132 | |
| 133 | static void |
| 134 | percpu_cpu_enlarge(size_t size) |
| 135 | { |
| 136 | CPU_INFO_ITERATOR cii; |
| 137 | struct cpu_info *ci; |
| 138 | |
| 139 | for (CPU_INFO_FOREACH(cii, ci)) { |
| 140 | percpu_cpu_t pcc; |
| 141 | |
| 142 | pcc.pcc_data = kmem_alloc(size, KM_SLEEP); /* XXX cacheline */ |
| 143 | pcc.pcc_size = size; |
| 144 | if (!mp_online) { |
| 145 | percpu_cpu_swap(ci, &pcc); |
| 146 | } else { |
| 147 | uint64_t where; |
| 148 | |
| 149 | where = xc_unicast(0, percpu_cpu_swap, ci, &pcc, ci); |
| 150 | xc_wait(where); |
| 151 | } |
| 152 | KASSERT(pcc.pcc_size < size); |
| 153 | if (pcc.pcc_data != NULL) { |
| 154 | kmem_free(pcc.pcc_data, pcc.pcc_size); |
| 155 | } |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * percpu_backend_alloc: vmem import callback for percpu_offset_arena |
| 161 | */ |
| 162 | |
| 163 | static int |
| 164 | percpu_backend_alloc(vmem_t *dummy, vmem_size_t size, vmem_size_t *resultsize, |
| 165 | vm_flag_t vmflags, vmem_addr_t *addrp) |
| 166 | { |
| 167 | unsigned int offset; |
| 168 | unsigned int nextoff; |
| 169 | |
| 170 | ASSERT_SLEEPABLE(); |
| 171 | KASSERT(dummy == NULL); |
| 172 | |
| 173 | if ((vmflags & VM_NOSLEEP) != 0) |
| 174 | return ENOMEM; |
| 175 | |
| 176 | size = roundup(size, PERCPU_IMPORT_SIZE); |
| 177 | mutex_enter(&percpu_allocation_lock); |
| 178 | offset = percpu_nextoff; |
| 179 | percpu_nextoff = nextoff = percpu_nextoff + size; |
| 180 | mutex_exit(&percpu_allocation_lock); |
| 181 | |
| 182 | percpu_cpu_enlarge(nextoff); |
| 183 | |
| 184 | *resultsize = size; |
| 185 | *addrp = (vmem_addr_t)offset; |
| 186 | return 0; |
| 187 | } |
| 188 | |
| 189 | static void |
| 190 | percpu_zero_cb(void *vp, void *vp2, struct cpu_info *ci) |
| 191 | { |
| 192 | size_t sz = (uintptr_t)vp2; |
| 193 | |
| 194 | memset(vp, 0, sz); |
| 195 | } |
| 196 | |
| 197 | /* |
| 198 | * percpu_zero: initialize percpu storage with zero. |
| 199 | */ |
| 200 | |
| 201 | static void |
| 202 | percpu_zero(percpu_t *pc, size_t sz) |
| 203 | { |
| 204 | |
| 205 | percpu_foreach(pc, percpu_zero_cb, (void *)(uintptr_t)sz); |
| 206 | } |
| 207 | |
| 208 | /* |
| 209 | * percpu_init: subsystem initialization |
| 210 | */ |
| 211 | |
| 212 | void |
| 213 | percpu_init(void) |
| 214 | { |
| 215 | |
| 216 | ASSERT_SLEEPABLE(); |
| 217 | rw_init(&percpu_swap_lock); |
| 218 | mutex_init(&percpu_allocation_lock, MUTEX_DEFAULT, IPL_NONE); |
| 219 | percpu_nextoff = PERCPU_QUANTUM_SIZE; |
| 220 | |
| 221 | percpu_offset_arena = vmem_xcreate("percpu" , 0, 0, PERCPU_QUANTUM_SIZE, |
| 222 | percpu_backend_alloc, NULL, NULL, PERCPU_QCACHE_MAX, VM_SLEEP, |
| 223 | IPL_NONE); |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * percpu_init_cpu: cpu initialization |
| 228 | * |
| 229 | * => should be called before the cpu appears on the list for CPU_INFO_FOREACH. |
| 230 | */ |
| 231 | |
| 232 | void |
| 233 | percpu_init_cpu(struct cpu_info *ci) |
| 234 | { |
| 235 | percpu_cpu_t * const pcc = cpu_percpu(ci); |
| 236 | size_t size = percpu_nextoff; /* XXX racy */ |
| 237 | |
| 238 | ASSERT_SLEEPABLE(); |
| 239 | pcc->pcc_size = size; |
| 240 | if (size) { |
| 241 | pcc->pcc_data = kmem_zalloc(pcc->pcc_size, KM_SLEEP); |
| 242 | } |
| 243 | } |
| 244 | |
| 245 | /* |
| 246 | * percpu_alloc: allocate percpu storage |
| 247 | * |
| 248 | * => called in thread context. |
| 249 | * => considered as an expensive and rare operation. |
| 250 | * => allocated storage is initialized with zeros. |
| 251 | */ |
| 252 | |
| 253 | percpu_t * |
| 254 | percpu_alloc(size_t size) |
| 255 | { |
| 256 | vmem_addr_t offset; |
| 257 | percpu_t *pc; |
| 258 | |
| 259 | ASSERT_SLEEPABLE(); |
| 260 | if (vmem_alloc(percpu_offset_arena, size, VM_SLEEP | VM_BESTFIT, |
| 261 | &offset) != 0) |
| 262 | return NULL; |
| 263 | pc = (percpu_t *)percpu_encrypt((uintptr_t)offset); |
| 264 | percpu_zero(pc, size); |
| 265 | return pc; |
| 266 | } |
| 267 | |
| 268 | /* |
| 269 | * percpu_free: free percpu storage |
| 270 | * |
| 271 | * => called in thread context. |
| 272 | * => considered as an expensive and rare operation. |
| 273 | */ |
| 274 | |
| 275 | void |
| 276 | percpu_free(percpu_t *pc, size_t size) |
| 277 | { |
| 278 | |
| 279 | ASSERT_SLEEPABLE(); |
| 280 | vmem_free(percpu_offset_arena, (vmem_addr_t)percpu_offset(pc), size); |
| 281 | } |
| 282 | |
| 283 | /* |
| 284 | * percpu_getref: |
| 285 | * |
| 286 | * => safe to be used in either thread or interrupt context |
| 287 | * => disables preemption; must be bracketed with a percpu_putref() |
| 288 | */ |
| 289 | |
| 290 | void * |
| 291 | percpu_getref(percpu_t *pc) |
| 292 | { |
| 293 | |
| 294 | kpreempt_disable(); |
| 295 | return percpu_getptr_remote(pc, curcpu()); |
| 296 | } |
| 297 | |
| 298 | /* |
| 299 | * percpu_putref: |
| 300 | * |
| 301 | * => drops the preemption-disabled count after caller is done with per-cpu |
| 302 | * data |
| 303 | */ |
| 304 | |
| 305 | void |
| 306 | percpu_putref(percpu_t *pc) |
| 307 | { |
| 308 | |
| 309 | kpreempt_enable(); |
| 310 | } |
| 311 | |
| 312 | /* |
| 313 | * percpu_traverse_enter, percpu_traverse_exit, percpu_getptr_remote: |
| 314 | * helpers to access remote cpu's percpu data. |
| 315 | * |
| 316 | * => called in thread context. |
| 317 | * => percpu_traverse_enter can block low-priority xcalls. |
| 318 | * => typical usage would be: |
| 319 | * |
| 320 | * sum = 0; |
| 321 | * percpu_traverse_enter(); |
| 322 | * for (CPU_INFO_FOREACH(cii, ci)) { |
| 323 | * unsigned int *p = percpu_getptr_remote(pc, ci); |
| 324 | * sum += *p; |
| 325 | * } |
| 326 | * percpu_traverse_exit(); |
| 327 | */ |
| 328 | |
| 329 | void |
| 330 | percpu_traverse_enter(void) |
| 331 | { |
| 332 | |
| 333 | ASSERT_SLEEPABLE(); |
| 334 | rw_enter(&percpu_swap_lock, RW_READER); |
| 335 | } |
| 336 | |
| 337 | void |
| 338 | percpu_traverse_exit(void) |
| 339 | { |
| 340 | |
| 341 | rw_exit(&percpu_swap_lock); |
| 342 | } |
| 343 | |
| 344 | void * |
| 345 | percpu_getptr_remote(percpu_t *pc, struct cpu_info *ci) |
| 346 | { |
| 347 | |
| 348 | return &((char *)cpu_percpu(ci)->pcc_data)[percpu_offset(pc)]; |
| 349 | } |
| 350 | |
| 351 | /* |
| 352 | * percpu_foreach: call the specified callback function for each cpus. |
| 353 | * |
| 354 | * => called in thread context. |
| 355 | * => caller should not rely on the cpu iteration order. |
| 356 | * => the callback function should be minimum because it is executed with |
| 357 | * holding a global lock, which can block low-priority xcalls. |
| 358 | * eg. it's illegal for a callback function to sleep for memory allocation. |
| 359 | */ |
| 360 | void |
| 361 | percpu_foreach(percpu_t *pc, percpu_callback_t cb, void *arg) |
| 362 | { |
| 363 | CPU_INFO_ITERATOR cii; |
| 364 | struct cpu_info *ci; |
| 365 | |
| 366 | percpu_traverse_enter(); |
| 367 | for (CPU_INFO_FOREACH(cii, ci)) { |
| 368 | (*cb)(percpu_getptr_remote(pc, ci), arg, ci); |
| 369 | } |
| 370 | percpu_traverse_exit(); |
| 371 | } |
| 372 | |