1 /*- 2 * Copyright (c) 2010 Hudson River Trading LLC 3 * Written by: John H. Baldwin <jhb@FreeBSD.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include "opt_vm.h" 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/bus.h> 36 #include <sys/kernel.h> 37 #include <sys/lock.h> 38 #include <sys/mutex.h> 39 #include <sys/smp.h> 40 #include <sys/vmmeter.h> 41 #include <vm/vm.h> 42 #include <vm/pmap.h> 43 #include <vm/vm_param.h> 44 #include <vm/vm_page.h> 45 #include <vm/vm_phys.h> 46 47 #include <contrib/dev/acpica/include/acpi.h> 48 #include <contrib/dev/acpica/include/aclocal.h> 49 #include <contrib/dev/acpica/include/actables.h> 50 51 #include <machine/intr_machdep.h> 52 #include <x86/apicvar.h> 53 54 #include <dev/acpica/acpivar.h> 55 56 #if MAXMEMDOM > 1 57 struct cpu_info { 58 int enabled:1; 59 int has_memory:1; 60 int domain; 61 } cpus[MAX_APIC_ID + 1]; 62 63 struct mem_affinity mem_info[VM_PHYSSEG_MAX + 1]; 64 int num_mem; 65 66 static ACPI_TABLE_SRAT *srat; 67 static vm_paddr_t srat_physaddr; 68 69 static int domain_pxm[MAXMEMDOM]; 70 static int ndomain; 71 72 static ACPI_TABLE_SLIT *slit; 73 static vm_paddr_t slit_physaddr; 74 static int vm_locality_table[MAXMEMDOM * MAXMEMDOM]; 75 76 static void srat_walk_table(acpi_subtable_handler *handler, void *arg); 77 78 /* 79 * SLIT parsing. 80 */ 81 82 static void 83 slit_parse_table(ACPI_TABLE_SLIT *s) 84 { 85 int i, j; 86 int i_domain, j_domain; 87 int offset = 0; 88 uint8_t e; 89 90 /* 91 * This maps the SLIT data into the VM-domain centric view. 92 * There may be sparse entries in the PXM namespace, so 93 * remap them to a VM-domain ID and if it doesn't exist, 94 * skip it. 95 * 96 * It should result in a packed 2d array of VM-domain 97 * locality information entries. 98 */ 99 100 if (bootverbose) 101 printf("SLIT.Localities: %d\n", (int) s->LocalityCount); 102 for (i = 0; i < s->LocalityCount; i++) { 103 i_domain = acpi_map_pxm_to_vm_domainid(i); 104 if (i_domain < 0) 105 continue; 106 107 if (bootverbose) 108 printf("%d: ", i); 109 for (j = 0; j < s->LocalityCount; j++) { 110 j_domain = acpi_map_pxm_to_vm_domainid(j); 111 if (j_domain < 0) 112 continue; 113 e = s->Entry[i * s->LocalityCount + j]; 114 if (bootverbose) 115 printf("%d ", (int) e); 116 /* 255 == "no locality information" */ 117 if (e == 255) 118 vm_locality_table[offset] = -1; 119 else 120 vm_locality_table[offset] = e; 121 offset++; 122 } 123 if (bootverbose) 124 printf("\n"); 125 } 126 } 127 128 /* 129 * Look for an ACPI System Locality Distance Information Table ("SLIT") 130 */ 131 static int 132 parse_slit(void) 133 { 134 135 if (resource_disabled("slit", 0)) { 136 return (-1); 137 } 138 139 slit_physaddr = acpi_find_table(ACPI_SIG_SLIT); 140 if (slit_physaddr == 0) { 141 return (-1); 142 } 143 144 /* 145 * Make a pass over the table to populate the cpus[] and 146 * mem_info[] tables. 147 */ 148 slit = acpi_map_table(slit_physaddr, ACPI_SIG_SLIT); 149 slit_parse_table(slit); 150 acpi_unmap_table(slit); 151 slit = NULL; 152 153 #ifdef VM_NUMA_ALLOC 154 /* Tell the VM about it! */ 155 mem_locality = vm_locality_table; 156 #endif 157 return (0); 158 } 159 160 /* 161 * SRAT parsing. 162 */ 163 164 /* 165 * Returns true if a memory range overlaps with at least one range in 166 * phys_avail[]. 167 */ 168 static int 169 overlaps_phys_avail(vm_paddr_t start, vm_paddr_t end) 170 { 171 int i; 172 173 for (i = 0; phys_avail[i] != 0 && phys_avail[i + 1] != 0; i += 2) { 174 if (phys_avail[i + 1] < start) 175 continue; 176 if (phys_avail[i] < end) 177 return (1); 178 break; 179 } 180 return (0); 181 182 } 183 184 static void 185 srat_parse_entry(ACPI_SUBTABLE_HEADER *entry, void *arg) 186 { 187 ACPI_SRAT_CPU_AFFINITY *cpu; 188 ACPI_SRAT_X2APIC_CPU_AFFINITY *x2apic; 189 ACPI_SRAT_MEM_AFFINITY *mem; 190 int domain, i, slot; 191 192 switch (entry->Type) { 193 case ACPI_SRAT_TYPE_CPU_AFFINITY: 194 cpu = (ACPI_SRAT_CPU_AFFINITY *)entry; 195 domain = cpu->ProximityDomainLo | 196 cpu->ProximityDomainHi[0] << 8 | 197 cpu->ProximityDomainHi[1] << 16 | 198 cpu->ProximityDomainHi[2] << 24; 199 if (bootverbose) 200 printf("SRAT: Found CPU APIC ID %u domain %d: %s\n", 201 cpu->ApicId, domain, 202 (cpu->Flags & ACPI_SRAT_CPU_ENABLED) ? 203 "enabled" : "disabled"); 204 if (!(cpu->Flags & ACPI_SRAT_CPU_ENABLED)) 205 break; 206 if (cpu->ApicId > MAX_APIC_ID) { 207 printf("SRAT: Ignoring local APIC ID %u (too high)\n", 208 cpu->ApicId); 209 break; 210 } 211 212 if (cpus[cpu->ApicId].enabled) { 213 printf("SRAT: Duplicate local APIC ID %u\n", 214 cpu->ApicId); 215 *(int *)arg = ENXIO; 216 break; 217 } 218 cpus[cpu->ApicId].domain = domain; 219 cpus[cpu->ApicId].enabled = 1; 220 break; 221 case ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY: 222 x2apic = (ACPI_SRAT_X2APIC_CPU_AFFINITY *)entry; 223 if (bootverbose) 224 printf("SRAT: Found CPU APIC ID %u domain %d: %s\n", 225 x2apic->ApicId, x2apic->ProximityDomain, 226 (x2apic->Flags & ACPI_SRAT_CPU_ENABLED) ? 227 "enabled" : "disabled"); 228 if (!(x2apic->Flags & ACPI_SRAT_CPU_ENABLED)) 229 break; 230 if (x2apic->ApicId > MAX_APIC_ID) { 231 printf("SRAT: Ignoring local APIC ID %u (too high)\n", 232 x2apic->ApicId); 233 break; 234 } 235 236 KASSERT(!cpus[x2apic->ApicId].enabled, 237 ("Duplicate local APIC ID %u", x2apic->ApicId)); 238 cpus[x2apic->ApicId].domain = x2apic->ProximityDomain; 239 cpus[x2apic->ApicId].enabled = 1; 240 break; 241 case ACPI_SRAT_TYPE_MEMORY_AFFINITY: 242 mem = (ACPI_SRAT_MEM_AFFINITY *)entry; 243 if (bootverbose) 244 printf( 245 "SRAT: Found memory domain %d addr 0x%jx len 0x%jx: %s\n", 246 mem->ProximityDomain, (uintmax_t)mem->BaseAddress, 247 (uintmax_t)mem->Length, 248 (mem->Flags & ACPI_SRAT_MEM_ENABLED) ? 249 "enabled" : "disabled"); 250 if (!(mem->Flags & ACPI_SRAT_MEM_ENABLED)) 251 break; 252 if (!overlaps_phys_avail(mem->BaseAddress, 253 mem->BaseAddress + mem->Length)) { 254 printf("SRAT: Ignoring memory at addr 0x%jx\n", 255 (uintmax_t)mem->BaseAddress); 256 break; 257 } 258 if (num_mem == VM_PHYSSEG_MAX) { 259 printf("SRAT: Too many memory regions\n"); 260 *(int *)arg = ENXIO; 261 break; 262 } 263 slot = num_mem; 264 for (i = 0; i < num_mem; i++) { 265 if (mem_info[i].end <= mem->BaseAddress) 266 continue; 267 if (mem_info[i].start < 268 (mem->BaseAddress + mem->Length)) { 269 printf("SRAT: Overlapping memory entries\n"); 270 *(int *)arg = ENXIO; 271 return; 272 } 273 slot = i; 274 } 275 for (i = num_mem; i > slot; i--) 276 mem_info[i] = mem_info[i - 1]; 277 mem_info[slot].start = mem->BaseAddress; 278 mem_info[slot].end = mem->BaseAddress + mem->Length; 279 mem_info[slot].domain = mem->ProximityDomain; 280 num_mem++; 281 break; 282 } 283 } 284 285 /* 286 * Ensure each memory domain has at least one CPU and that each CPU 287 * has at least one memory domain. 288 */ 289 static int 290 check_domains(void) 291 { 292 int found, i, j; 293 294 for (i = 0; i < num_mem; i++) { 295 found = 0; 296 for (j = 0; j <= MAX_APIC_ID; j++) 297 if (cpus[j].enabled && 298 cpus[j].domain == mem_info[i].domain) { 299 cpus[j].has_memory = 1; 300 found++; 301 } 302 if (!found) { 303 printf("SRAT: No CPU found for memory domain %d\n", 304 mem_info[i].domain); 305 return (ENXIO); 306 } 307 } 308 for (i = 0; i <= MAX_APIC_ID; i++) 309 if (cpus[i].enabled && !cpus[i].has_memory) { 310 printf("SRAT: No memory found for CPU %d\n", i); 311 return (ENXIO); 312 } 313 return (0); 314 } 315 316 /* 317 * Check that the SRAT memory regions cover all of the regions in 318 * phys_avail[]. 319 */ 320 static int 321 check_phys_avail(void) 322 { 323 vm_paddr_t address; 324 int i, j; 325 326 /* j is the current offset into phys_avail[]. */ 327 address = phys_avail[0]; 328 j = 0; 329 for (i = 0; i < num_mem; i++) { 330 /* 331 * Consume as many phys_avail[] entries as fit in this 332 * region. 333 */ 334 while (address >= mem_info[i].start && 335 address <= mem_info[i].end) { 336 /* 337 * If we cover the rest of this phys_avail[] entry, 338 * advance to the next entry. 339 */ 340 if (phys_avail[j + 1] <= mem_info[i].end) { 341 j += 2; 342 if (phys_avail[j] == 0 && 343 phys_avail[j + 1] == 0) { 344 return (0); 345 } 346 address = phys_avail[j]; 347 } else 348 address = mem_info[i].end + 1; 349 } 350 } 351 printf("SRAT: No memory region found for 0x%jx - 0x%jx\n", 352 (uintmax_t)phys_avail[j], (uintmax_t)phys_avail[j + 1]); 353 return (ENXIO); 354 } 355 356 /* 357 * Renumber the memory domains to be compact and zero-based if not 358 * already. Returns an error if there are too many domains. 359 */ 360 static int 361 renumber_domains(void) 362 { 363 int i, j, slot; 364 365 /* Enumerate all the domains. */ 366 ndomain = 0; 367 for (i = 0; i < num_mem; i++) { 368 /* See if this domain is already known. */ 369 for (j = 0; j < ndomain; j++) { 370 if (domain_pxm[j] >= mem_info[i].domain) 371 break; 372 } 373 if (j < ndomain && domain_pxm[j] == mem_info[i].domain) 374 continue; 375 376 if (ndomain >= MAXMEMDOM) { 377 ndomain = 1; 378 printf("SRAT: Too many memory domains\n"); 379 return (EFBIG); 380 } 381 382 /* Insert the new domain at slot 'j'. */ 383 slot = j; 384 for (j = ndomain; j > slot; j--) 385 domain_pxm[j] = domain_pxm[j - 1]; 386 domain_pxm[slot] = mem_info[i].domain; 387 ndomain++; 388 } 389 390 /* Renumber each domain to its index in the sorted 'domain_pxm' list. */ 391 for (i = 0; i < ndomain; i++) { 392 /* 393 * If the domain is already the right value, no need 394 * to renumber. 395 */ 396 if (domain_pxm[i] == i) 397 continue; 398 399 /* Walk the cpu[] and mem_info[] arrays to renumber. */ 400 for (j = 0; j < num_mem; j++) 401 if (mem_info[j].domain == domain_pxm[i]) 402 mem_info[j].domain = i; 403 for (j = 0; j <= MAX_APIC_ID; j++) 404 if (cpus[j].enabled && cpus[j].domain == domain_pxm[i]) 405 cpus[j].domain = i; 406 } 407 408 return (0); 409 } 410 411 /* 412 * Look for an ACPI System Resource Affinity Table ("SRAT") 413 */ 414 static int 415 parse_srat(void) 416 { 417 int error; 418 419 if (resource_disabled("srat", 0)) 420 return (-1); 421 422 srat_physaddr = acpi_find_table(ACPI_SIG_SRAT); 423 if (srat_physaddr == 0) 424 return (-1); 425 426 /* 427 * Make a pass over the table to populate the cpus[] and 428 * mem_info[] tables. 429 */ 430 srat = acpi_map_table(srat_physaddr, ACPI_SIG_SRAT); 431 error = 0; 432 srat_walk_table(srat_parse_entry, &error); 433 acpi_unmap_table(srat); 434 srat = NULL; 435 if (error || check_domains() != 0 || check_phys_avail() != 0 || 436 renumber_domains() != 0) { 437 srat_physaddr = 0; 438 return (-1); 439 } 440 441 #ifdef VM_NUMA_ALLOC 442 /* Point vm_phys at our memory affinity table. */ 443 vm_ndomains = ndomain; 444 mem_affinity = mem_info; 445 #endif 446 447 return (0); 448 } 449 450 static void 451 init_mem_locality(void) 452 { 453 int i; 454 455 /* 456 * For now, assume -1 == "no locality information for 457 * this pairing. 458 */ 459 for (i = 0; i < MAXMEMDOM * MAXMEMDOM; i++) 460 vm_locality_table[i] = -1; 461 } 462 463 static void 464 parse_acpi_tables(void *dummy) 465 { 466 467 if (parse_srat() < 0) 468 return; 469 init_mem_locality(); 470 (void) parse_slit(); 471 } 472 SYSINIT(parse_acpi_tables, SI_SUB_VM - 1, SI_ORDER_FIRST, parse_acpi_tables, 473 NULL); 474 475 static void 476 srat_walk_table(acpi_subtable_handler *handler, void *arg) 477 { 478 479 acpi_walk_subtables(srat + 1, (char *)srat + srat->Header.Length, 480 handler, arg); 481 } 482 483 /* 484 * Setup per-CPU domain IDs. 485 */ 486 static void 487 srat_set_cpus(void *dummy) 488 { 489 struct cpu_info *cpu; 490 struct pcpu *pc; 491 u_int i; 492 493 if (srat_physaddr == 0) 494 return; 495 for (i = 0; i < MAXCPU; i++) { 496 if (CPU_ABSENT(i)) 497 continue; 498 pc = pcpu_find(i); 499 KASSERT(pc != NULL, ("no pcpu data for CPU %u", i)); 500 cpu = &cpus[pc->pc_apic_id]; 501 if (!cpu->enabled) 502 panic("SRAT: CPU with APIC ID %u is not known", 503 pc->pc_apic_id); 504 pc->pc_domain = cpu->domain; 505 CPU_SET(i, &cpuset_domain[cpu->domain]); 506 if (bootverbose) 507 printf("SRAT: CPU %u has memory domain %d\n", i, 508 cpu->domain); 509 } 510 } 511 SYSINIT(srat_set_cpus, SI_SUB_CPU, SI_ORDER_ANY, srat_set_cpus, NULL); 512 513 /* 514 * Map a _PXM value to a VM domain ID. 515 * 516 * Returns the domain ID, or -1 if no domain ID was found. 517 */ 518 int 519 acpi_map_pxm_to_vm_domainid(int pxm) 520 { 521 int i; 522 523 for (i = 0; i < ndomain; i++) { 524 if (domain_pxm[i] == pxm) 525 return (i); 526 } 527 528 return (-1); 529 } 530 531 #else /* MAXMEMDOM == 1 */ 532 533 int 534 acpi_map_pxm_to_vm_domainid(int pxm) 535 { 536 537 return (-1); 538 } 539 540 #endif /* MAXMEMDOM > 1 */ 541