1 #include <linux/sysdev.h> 2 #include <linux/cpu.h> 3 #include <linux/smp.h> 4 #include <linux/percpu.h> 5 #include <linux/init.h> 6 #include <linux/sched.h> 7 #include <linux/module.h> 8 #include <linux/nodemask.h> 9 #include <linux/cpumask.h> 10 #include <linux/notifier.h> 11 12 #include <asm/current.h> 13 #include <asm/processor.h> 14 #include <asm/cputable.h> 15 #include <asm/firmware.h> 16 #include <asm/hvcall.h> 17 #include <asm/prom.h> 18 #include <asm/paca.h> 19 #include <asm/lppaca.h> 20 #include <asm/machdep.h> 21 #include <asm/smp.h> 22 23 static DEFINE_PER_CPU(struct cpu, cpu_devices); 24 25 /* SMT stuff */ 26 27 #ifdef CONFIG_PPC_MULTIPLATFORM 28 /* Time in microseconds we delay before sleeping in the idle loop */ 29 DEFINE_PER_CPU(unsigned long, smt_snooze_delay) = { 100 }; 30 31 static ssize_t store_smt_snooze_delay(struct sys_device *dev, const char *buf, 32 size_t count) 33 { 34 struct cpu *cpu = container_of(dev, struct cpu, sysdev); 35 ssize_t ret; 36 unsigned long snooze; 37 38 ret = sscanf(buf, "%lu", &snooze); 39 if (ret != 1) 40 return -EINVAL; 41 42 per_cpu(smt_snooze_delay, cpu->sysdev.id) = snooze; 43 44 return count; 45 } 46 47 static ssize_t show_smt_snooze_delay(struct sys_device *dev, char *buf) 48 { 49 struct cpu *cpu = container_of(dev, struct cpu, sysdev); 50 51 return sprintf(buf, "%lu\n", per_cpu(smt_snooze_delay, cpu->sysdev.id)); 52 } 53 54 static SYSDEV_ATTR(smt_snooze_delay, 0644, show_smt_snooze_delay, 55 store_smt_snooze_delay); 56 57 /* Only parse OF options if the matching cmdline option was not specified */ 58 static int smt_snooze_cmdline; 59 60 static int __init smt_setup(void) 61 { 62 struct device_node *options; 63 const unsigned int *val; 64 unsigned int cpu; 65 66 if (!cpu_has_feature(CPU_FTR_SMT)) 67 return -ENODEV; 68 69 options = of_find_node_by_path("/options"); 70 if (!options) 71 return -ENODEV; 72 73 val = of_get_property(options, "ibm,smt-snooze-delay", NULL); 74 if (!smt_snooze_cmdline && val) { 75 for_each_possible_cpu(cpu) 76 per_cpu(smt_snooze_delay, cpu) = *val; 77 } 78 79 of_node_put(options); 80 return 0; 81 } 82 __initcall(smt_setup); 83 84 static int __init setup_smt_snooze_delay(char *str) 85 { 86 unsigned int cpu; 87 int snooze; 88 89 if (!cpu_has_feature(CPU_FTR_SMT)) 90 return 1; 91 92 smt_snooze_cmdline = 1; 93 94 if (get_option(&str, &snooze)) { 95 for_each_possible_cpu(cpu) 96 per_cpu(smt_snooze_delay, cpu) = snooze; 97 } 98 99 return 1; 100 } 101 __setup("smt-snooze-delay=", setup_smt_snooze_delay); 102 103 #endif /* CONFIG_PPC_MULTIPLATFORM */ 104 105 /* 106 * Enabling PMCs will slow partition context switch times so we only do 107 * it the first time we write to the PMCs. 108 */ 109 110 static DEFINE_PER_CPU(char, pmcs_enabled); 111 112 void ppc64_enable_pmcs(void) 113 { 114 /* Only need to enable them once */ 115 if (__get_cpu_var(pmcs_enabled)) 116 return; 117 118 __get_cpu_var(pmcs_enabled) = 1; 119 120 if (ppc_md.enable_pmcs) 121 ppc_md.enable_pmcs(); 122 } 123 EXPORT_SYMBOL(ppc64_enable_pmcs); 124 125 /* XXX convert to rusty's on_one_cpu */ 126 static unsigned long run_on_cpu(unsigned long cpu, 127 unsigned long (*func)(unsigned long), 128 unsigned long arg) 129 { 130 cpumask_t old_affinity = current->cpus_allowed; 131 unsigned long ret; 132 133 /* should return -EINVAL to userspace */ 134 if (set_cpus_allowed(current, cpumask_of_cpu(cpu))) 135 return 0; 136 137 ret = func(arg); 138 139 set_cpus_allowed(current, old_affinity); 140 141 return ret; 142 } 143 144 #define SYSFS_PMCSETUP(NAME, ADDRESS) \ 145 static unsigned long read_##NAME(unsigned long junk) \ 146 { \ 147 return mfspr(ADDRESS); \ 148 } \ 149 static unsigned long write_##NAME(unsigned long val) \ 150 { \ 151 ppc64_enable_pmcs(); \ 152 mtspr(ADDRESS, val); \ 153 return 0; \ 154 } \ 155 static ssize_t show_##NAME(struct sys_device *dev, char *buf) \ 156 { \ 157 struct cpu *cpu = container_of(dev, struct cpu, sysdev); \ 158 unsigned long val = run_on_cpu(cpu->sysdev.id, read_##NAME, 0); \ 159 return sprintf(buf, "%lx\n", val); \ 160 } \ 161 static ssize_t __used \ 162 store_##NAME(struct sys_device *dev, const char *buf, size_t count) \ 163 { \ 164 struct cpu *cpu = container_of(dev, struct cpu, sysdev); \ 165 unsigned long val; \ 166 int ret = sscanf(buf, "%lx", &val); \ 167 if (ret != 1) \ 168 return -EINVAL; \ 169 run_on_cpu(cpu->sysdev.id, write_##NAME, val); \ 170 return count; \ 171 } 172 173 174 /* Let's define all possible registers, we'll only hook up the ones 175 * that are implemented on the current processor 176 */ 177 178 SYSFS_PMCSETUP(mmcr0, SPRN_MMCR0); 179 SYSFS_PMCSETUP(mmcr1, SPRN_MMCR1); 180 SYSFS_PMCSETUP(mmcra, SPRN_MMCRA); 181 SYSFS_PMCSETUP(pmc1, SPRN_PMC1); 182 SYSFS_PMCSETUP(pmc2, SPRN_PMC2); 183 SYSFS_PMCSETUP(pmc3, SPRN_PMC3); 184 SYSFS_PMCSETUP(pmc4, SPRN_PMC4); 185 SYSFS_PMCSETUP(pmc5, SPRN_PMC5); 186 SYSFS_PMCSETUP(pmc6, SPRN_PMC6); 187 SYSFS_PMCSETUP(pmc7, SPRN_PMC7); 188 SYSFS_PMCSETUP(pmc8, SPRN_PMC8); 189 SYSFS_PMCSETUP(purr, SPRN_PURR); 190 SYSFS_PMCSETUP(spurr, SPRN_SPURR); 191 SYSFS_PMCSETUP(dscr, SPRN_DSCR); 192 193 SYSFS_PMCSETUP(pa6t_pmc0, SPRN_PA6T_PMC0); 194 SYSFS_PMCSETUP(pa6t_pmc1, SPRN_PA6T_PMC1); 195 SYSFS_PMCSETUP(pa6t_pmc2, SPRN_PA6T_PMC2); 196 SYSFS_PMCSETUP(pa6t_pmc3, SPRN_PA6T_PMC3); 197 SYSFS_PMCSETUP(pa6t_pmc4, SPRN_PA6T_PMC4); 198 SYSFS_PMCSETUP(pa6t_pmc5, SPRN_PA6T_PMC5); 199 200 #ifdef CONFIG_DEBUG_KERNEL 201 SYSFS_PMCSETUP(hid0, SPRN_HID0); 202 SYSFS_PMCSETUP(hid1, SPRN_HID1); 203 SYSFS_PMCSETUP(hid4, SPRN_HID4); 204 SYSFS_PMCSETUP(hid5, SPRN_HID5); 205 SYSFS_PMCSETUP(ima0, SPRN_PA6T_IMA0); 206 SYSFS_PMCSETUP(ima1, SPRN_PA6T_IMA1); 207 SYSFS_PMCSETUP(ima2, SPRN_PA6T_IMA2); 208 SYSFS_PMCSETUP(ima3, SPRN_PA6T_IMA3); 209 SYSFS_PMCSETUP(ima4, SPRN_PA6T_IMA4); 210 SYSFS_PMCSETUP(ima5, SPRN_PA6T_IMA5); 211 SYSFS_PMCSETUP(ima6, SPRN_PA6T_IMA6); 212 SYSFS_PMCSETUP(ima7, SPRN_PA6T_IMA7); 213 SYSFS_PMCSETUP(ima8, SPRN_PA6T_IMA8); 214 SYSFS_PMCSETUP(ima9, SPRN_PA6T_IMA9); 215 SYSFS_PMCSETUP(imaat, SPRN_PA6T_IMAAT); 216 SYSFS_PMCSETUP(btcr, SPRN_PA6T_BTCR); 217 SYSFS_PMCSETUP(pccr, SPRN_PA6T_PCCR); 218 SYSFS_PMCSETUP(rpccr, SPRN_PA6T_RPCCR); 219 SYSFS_PMCSETUP(der, SPRN_PA6T_DER); 220 SYSFS_PMCSETUP(mer, SPRN_PA6T_MER); 221 SYSFS_PMCSETUP(ber, SPRN_PA6T_BER); 222 SYSFS_PMCSETUP(ier, SPRN_PA6T_IER); 223 SYSFS_PMCSETUP(sier, SPRN_PA6T_SIER); 224 SYSFS_PMCSETUP(siar, SPRN_PA6T_SIAR); 225 SYSFS_PMCSETUP(tsr0, SPRN_PA6T_TSR0); 226 SYSFS_PMCSETUP(tsr1, SPRN_PA6T_TSR1); 227 SYSFS_PMCSETUP(tsr2, SPRN_PA6T_TSR2); 228 SYSFS_PMCSETUP(tsr3, SPRN_PA6T_TSR3); 229 #endif /* CONFIG_DEBUG_KERNEL */ 230 231 static SYSDEV_ATTR(mmcra, 0600, show_mmcra, store_mmcra); 232 static SYSDEV_ATTR(spurr, 0600, show_spurr, NULL); 233 static SYSDEV_ATTR(dscr, 0600, show_dscr, store_dscr); 234 static SYSDEV_ATTR(purr, 0600, show_purr, store_purr); 235 236 static struct sysdev_attribute ibm_common_attrs[] = { 237 _SYSDEV_ATTR(mmcr0, 0600, show_mmcr0, store_mmcr0), 238 _SYSDEV_ATTR(mmcr1, 0600, show_mmcr1, store_mmcr1), 239 }; 240 241 static struct sysdev_attribute ibm_pmc_attrs[] = { 242 _SYSDEV_ATTR(pmc1, 0600, show_pmc1, store_pmc1), 243 _SYSDEV_ATTR(pmc2, 0600, show_pmc2, store_pmc2), 244 _SYSDEV_ATTR(pmc3, 0600, show_pmc3, store_pmc3), 245 _SYSDEV_ATTR(pmc4, 0600, show_pmc4, store_pmc4), 246 _SYSDEV_ATTR(pmc5, 0600, show_pmc5, store_pmc5), 247 _SYSDEV_ATTR(pmc6, 0600, show_pmc6, store_pmc6), 248 _SYSDEV_ATTR(pmc7, 0600, show_pmc7, store_pmc7), 249 _SYSDEV_ATTR(pmc8, 0600, show_pmc8, store_pmc8), 250 }; 251 252 static struct sysdev_attribute pa6t_attrs[] = { 253 _SYSDEV_ATTR(mmcr0, 0600, show_mmcr0, store_mmcr0), 254 _SYSDEV_ATTR(mmcr1, 0600, show_mmcr1, store_mmcr1), 255 _SYSDEV_ATTR(pmc0, 0600, show_pa6t_pmc0, store_pa6t_pmc0), 256 _SYSDEV_ATTR(pmc1, 0600, show_pa6t_pmc1, store_pa6t_pmc1), 257 _SYSDEV_ATTR(pmc2, 0600, show_pa6t_pmc2, store_pa6t_pmc2), 258 _SYSDEV_ATTR(pmc3, 0600, show_pa6t_pmc3, store_pa6t_pmc3), 259 _SYSDEV_ATTR(pmc4, 0600, show_pa6t_pmc4, store_pa6t_pmc4), 260 _SYSDEV_ATTR(pmc5, 0600, show_pa6t_pmc5, store_pa6t_pmc5), 261 #ifdef CONFIG_DEBUG_KERNEL 262 _SYSDEV_ATTR(hid0, 0600, show_hid0, store_hid0), 263 _SYSDEV_ATTR(hid1, 0600, show_hid1, store_hid1), 264 _SYSDEV_ATTR(hid4, 0600, show_hid4, store_hid4), 265 _SYSDEV_ATTR(hid5, 0600, show_hid5, store_hid5), 266 _SYSDEV_ATTR(ima0, 0600, show_ima0, store_ima0), 267 _SYSDEV_ATTR(ima1, 0600, show_ima1, store_ima1), 268 _SYSDEV_ATTR(ima2, 0600, show_ima2, store_ima2), 269 _SYSDEV_ATTR(ima3, 0600, show_ima3, store_ima3), 270 _SYSDEV_ATTR(ima4, 0600, show_ima4, store_ima4), 271 _SYSDEV_ATTR(ima5, 0600, show_ima5, store_ima5), 272 _SYSDEV_ATTR(ima6, 0600, show_ima6, store_ima6), 273 _SYSDEV_ATTR(ima7, 0600, show_ima7, store_ima7), 274 _SYSDEV_ATTR(ima8, 0600, show_ima8, store_ima8), 275 _SYSDEV_ATTR(ima9, 0600, show_ima9, store_ima9), 276 _SYSDEV_ATTR(imaat, 0600, show_imaat, store_imaat), 277 _SYSDEV_ATTR(btcr, 0600, show_btcr, store_btcr), 278 _SYSDEV_ATTR(pccr, 0600, show_pccr, store_pccr), 279 _SYSDEV_ATTR(rpccr, 0600, show_rpccr, store_rpccr), 280 _SYSDEV_ATTR(der, 0600, show_der, store_der), 281 _SYSDEV_ATTR(mer, 0600, show_mer, store_mer), 282 _SYSDEV_ATTR(ber, 0600, show_ber, store_ber), 283 _SYSDEV_ATTR(ier, 0600, show_ier, store_ier), 284 _SYSDEV_ATTR(sier, 0600, show_sier, store_sier), 285 _SYSDEV_ATTR(siar, 0600, show_siar, store_siar), 286 _SYSDEV_ATTR(tsr0, 0600, show_tsr0, store_tsr0), 287 _SYSDEV_ATTR(tsr1, 0600, show_tsr1, store_tsr1), 288 _SYSDEV_ATTR(tsr2, 0600, show_tsr2, store_tsr2), 289 _SYSDEV_ATTR(tsr3, 0600, show_tsr3, store_tsr3), 290 #endif /* CONFIG_DEBUG_KERNEL */ 291 }; 292 293 294 static void register_cpu_online(unsigned int cpu) 295 { 296 struct cpu *c = &per_cpu(cpu_devices, cpu); 297 struct sys_device *s = &c->sysdev; 298 struct sysdev_attribute *attrs, *pmc_attrs; 299 int i, nattrs; 300 301 if (!firmware_has_feature(FW_FEATURE_ISERIES) && 302 cpu_has_feature(CPU_FTR_SMT)) 303 sysdev_create_file(s, &attr_smt_snooze_delay); 304 305 /* PMC stuff */ 306 switch (cur_cpu_spec->pmc_type) { 307 case PPC_PMC_IBM: 308 attrs = ibm_common_attrs; 309 nattrs = sizeof(ibm_common_attrs) / sizeof(struct sysdev_attribute); 310 pmc_attrs = ibm_pmc_attrs; 311 break; 312 case PPC_PMC_PA6T: 313 /* PA Semi starts counting at PMC0 */ 314 attrs = pa6t_attrs; 315 nattrs = sizeof(pa6t_attrs) / sizeof(struct sysdev_attribute); 316 pmc_attrs = NULL; 317 break; 318 default: 319 attrs = NULL; 320 nattrs = 0; 321 pmc_attrs = NULL; 322 } 323 324 for (i = 0; i < nattrs; i++) 325 sysdev_create_file(s, &attrs[i]); 326 327 if (pmc_attrs) 328 for (i = 0; i < cur_cpu_spec->num_pmcs; i++) 329 sysdev_create_file(s, &pmc_attrs[i]); 330 331 if (cpu_has_feature(CPU_FTR_MMCRA)) 332 sysdev_create_file(s, &attr_mmcra); 333 334 if (cpu_has_feature(CPU_FTR_PURR)) 335 sysdev_create_file(s, &attr_purr); 336 337 if (cpu_has_feature(CPU_FTR_SPURR)) 338 sysdev_create_file(s, &attr_spurr); 339 340 if (cpu_has_feature(CPU_FTR_DSCR)) 341 sysdev_create_file(s, &attr_dscr); 342 } 343 344 #ifdef CONFIG_HOTPLUG_CPU 345 static void unregister_cpu_online(unsigned int cpu) 346 { 347 struct cpu *c = &per_cpu(cpu_devices, cpu); 348 struct sys_device *s = &c->sysdev; 349 struct sysdev_attribute *attrs, *pmc_attrs; 350 int i, nattrs; 351 352 BUG_ON(!c->hotpluggable); 353 354 if (!firmware_has_feature(FW_FEATURE_ISERIES) && 355 cpu_has_feature(CPU_FTR_SMT)) 356 sysdev_remove_file(s, &attr_smt_snooze_delay); 357 358 /* PMC stuff */ 359 switch (cur_cpu_spec->pmc_type) { 360 case PPC_PMC_IBM: 361 attrs = ibm_common_attrs; 362 nattrs = sizeof(ibm_common_attrs) / sizeof(struct sysdev_attribute); 363 pmc_attrs = ibm_pmc_attrs; 364 break; 365 case PPC_PMC_PA6T: 366 /* PA Semi starts counting at PMC0 */ 367 attrs = pa6t_attrs; 368 nattrs = sizeof(pa6t_attrs) / sizeof(struct sysdev_attribute); 369 pmc_attrs = NULL; 370 break; 371 default: 372 attrs = NULL; 373 nattrs = 0; 374 pmc_attrs = NULL; 375 } 376 377 for (i = 0; i < nattrs; i++) 378 sysdev_remove_file(s, &attrs[i]); 379 380 if (pmc_attrs) 381 for (i = 0; i < cur_cpu_spec->num_pmcs; i++) 382 sysdev_remove_file(s, &pmc_attrs[i]); 383 384 if (cpu_has_feature(CPU_FTR_MMCRA)) 385 sysdev_remove_file(s, &attr_mmcra); 386 387 if (cpu_has_feature(CPU_FTR_PURR)) 388 sysdev_remove_file(s, &attr_purr); 389 390 if (cpu_has_feature(CPU_FTR_SPURR)) 391 sysdev_remove_file(s, &attr_spurr); 392 393 if (cpu_has_feature(CPU_FTR_DSCR)) 394 sysdev_remove_file(s, &attr_dscr); 395 } 396 #endif /* CONFIG_HOTPLUG_CPU */ 397 398 static int __cpuinit sysfs_cpu_notify(struct notifier_block *self, 399 unsigned long action, void *hcpu) 400 { 401 unsigned int cpu = (unsigned int)(long)hcpu; 402 403 switch (action) { 404 case CPU_ONLINE: 405 case CPU_ONLINE_FROZEN: 406 register_cpu_online(cpu); 407 break; 408 #ifdef CONFIG_HOTPLUG_CPU 409 case CPU_DEAD: 410 case CPU_DEAD_FROZEN: 411 unregister_cpu_online(cpu); 412 break; 413 #endif 414 } 415 return NOTIFY_OK; 416 } 417 418 static struct notifier_block __cpuinitdata sysfs_cpu_nb = { 419 .notifier_call = sysfs_cpu_notify, 420 }; 421 422 static DEFINE_MUTEX(cpu_mutex); 423 424 int cpu_add_sysdev_attr(struct sysdev_attribute *attr) 425 { 426 int cpu; 427 428 mutex_lock(&cpu_mutex); 429 430 for_each_possible_cpu(cpu) { 431 sysdev_create_file(get_cpu_sysdev(cpu), attr); 432 } 433 434 mutex_unlock(&cpu_mutex); 435 return 0; 436 } 437 EXPORT_SYMBOL_GPL(cpu_add_sysdev_attr); 438 439 int cpu_add_sysdev_attr_group(struct attribute_group *attrs) 440 { 441 int cpu; 442 struct sys_device *sysdev; 443 int ret; 444 445 mutex_lock(&cpu_mutex); 446 447 for_each_possible_cpu(cpu) { 448 sysdev = get_cpu_sysdev(cpu); 449 ret = sysfs_create_group(&sysdev->kobj, attrs); 450 WARN_ON(ret != 0); 451 } 452 453 mutex_unlock(&cpu_mutex); 454 return 0; 455 } 456 EXPORT_SYMBOL_GPL(cpu_add_sysdev_attr_group); 457 458 459 void cpu_remove_sysdev_attr(struct sysdev_attribute *attr) 460 { 461 int cpu; 462 463 mutex_lock(&cpu_mutex); 464 465 for_each_possible_cpu(cpu) { 466 sysdev_remove_file(get_cpu_sysdev(cpu), attr); 467 } 468 469 mutex_unlock(&cpu_mutex); 470 } 471 EXPORT_SYMBOL_GPL(cpu_remove_sysdev_attr); 472 473 void cpu_remove_sysdev_attr_group(struct attribute_group *attrs) 474 { 475 int cpu; 476 struct sys_device *sysdev; 477 478 mutex_lock(&cpu_mutex); 479 480 for_each_possible_cpu(cpu) { 481 sysdev = get_cpu_sysdev(cpu); 482 sysfs_remove_group(&sysdev->kobj, attrs); 483 } 484 485 mutex_unlock(&cpu_mutex); 486 } 487 EXPORT_SYMBOL_GPL(cpu_remove_sysdev_attr_group); 488 489 490 /* NUMA stuff */ 491 492 #ifdef CONFIG_NUMA 493 static void register_nodes(void) 494 { 495 int i; 496 497 for (i = 0; i < MAX_NUMNODES; i++) 498 register_one_node(i); 499 } 500 501 int sysfs_add_device_to_node(struct sys_device *dev, int nid) 502 { 503 struct node *node = &node_devices[nid]; 504 return sysfs_create_link(&node->sysdev.kobj, &dev->kobj, 505 kobject_name(&dev->kobj)); 506 } 507 EXPORT_SYMBOL_GPL(sysfs_add_device_to_node); 508 509 void sysfs_remove_device_from_node(struct sys_device *dev, int nid) 510 { 511 struct node *node = &node_devices[nid]; 512 sysfs_remove_link(&node->sysdev.kobj, kobject_name(&dev->kobj)); 513 } 514 EXPORT_SYMBOL_GPL(sysfs_remove_device_from_node); 515 516 #else 517 static void register_nodes(void) 518 { 519 return; 520 } 521 522 #endif 523 524 /* Only valid if CPU is present. */ 525 static ssize_t show_physical_id(struct sys_device *dev, char *buf) 526 { 527 struct cpu *cpu = container_of(dev, struct cpu, sysdev); 528 529 return sprintf(buf, "%d\n", get_hard_smp_processor_id(cpu->sysdev.id)); 530 } 531 static SYSDEV_ATTR(physical_id, 0444, show_physical_id, NULL); 532 533 static int __init topology_init(void) 534 { 535 int cpu; 536 537 register_nodes(); 538 register_cpu_notifier(&sysfs_cpu_nb); 539 540 for_each_possible_cpu(cpu) { 541 struct cpu *c = &per_cpu(cpu_devices, cpu); 542 543 /* 544 * For now, we just see if the system supports making 545 * the RTAS calls for CPU hotplug. But, there may be a 546 * more comprehensive way to do this for an individual 547 * CPU. For instance, the boot cpu might never be valid 548 * for hotplugging. 549 */ 550 if (ppc_md.cpu_die) 551 c->hotpluggable = 1; 552 553 if (cpu_online(cpu) || c->hotpluggable) { 554 register_cpu(c, cpu); 555 556 sysdev_create_file(&c->sysdev, &attr_physical_id); 557 } 558 559 if (cpu_online(cpu)) 560 register_cpu_online(cpu); 561 } 562 563 return 0; 564 } 565 subsys_initcall(topology_init); 566