1 /* 2 * Copyright IBM Corp. 2007, 2011 3 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com> 4 */ 5 6 #define KMSG_COMPONENT "cpu" 7 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 8 9 #include <linux/workqueue.h> 10 #include <linux/bootmem.h> 11 #include <linux/cpuset.h> 12 #include <linux/device.h> 13 #include <linux/export.h> 14 #include <linux/kernel.h> 15 #include <linux/sched.h> 16 #include <linux/init.h> 17 #include <linux/delay.h> 18 #include <linux/cpu.h> 19 #include <linux/smp.h> 20 #include <linux/mm.h> 21 #include <asm/sysinfo.h> 22 23 #define PTF_HORIZONTAL (0UL) 24 #define PTF_VERTICAL (1UL) 25 #define PTF_CHECK (2UL) 26 27 struct mask_info { 28 struct mask_info *next; 29 unsigned char id; 30 cpumask_t mask; 31 }; 32 33 static void set_topology_timer(void); 34 static void topology_work_fn(struct work_struct *work); 35 static struct sysinfo_15_1_x *tl_info; 36 37 static int topology_enabled = 1; 38 static DECLARE_WORK(topology_work, topology_work_fn); 39 40 /* topology_lock protects the socket and book linked lists */ 41 static DEFINE_SPINLOCK(topology_lock); 42 static struct mask_info socket_info; 43 static struct mask_info book_info; 44 45 struct cpu_topology_s390 cpu_topology[NR_CPUS]; 46 EXPORT_SYMBOL_GPL(cpu_topology); 47 48 static cpumask_t cpu_group_map(struct mask_info *info, unsigned int cpu) 49 { 50 cpumask_t mask; 51 52 cpumask_copy(&mask, cpumask_of(cpu)); 53 if (!topology_enabled || !MACHINE_HAS_TOPOLOGY) 54 return mask; 55 for (; info; info = info->next) { 56 if (cpumask_test_cpu(cpu, &info->mask)) 57 return info->mask; 58 } 59 return mask; 60 } 61 62 static struct mask_info *add_cpus_to_mask(struct topology_cpu *tl_cpu, 63 struct mask_info *book, 64 struct mask_info *socket, 65 int one_socket_per_cpu) 66 { 67 unsigned int cpu; 68 69 for_each_set_bit(cpu, &tl_cpu->mask[0], TOPOLOGY_CPU_BITS) { 70 unsigned int rcpu; 71 int lcpu; 72 73 rcpu = TOPOLOGY_CPU_BITS - 1 - cpu + tl_cpu->origin; 74 lcpu = smp_find_processor_id(rcpu); 75 if (lcpu < 0) 76 continue; 77 cpumask_set_cpu(lcpu, &book->mask); 78 cpu_topology[lcpu].book_id = book->id; 79 cpumask_set_cpu(lcpu, &socket->mask); 80 cpu_topology[lcpu].core_id = rcpu; 81 if (one_socket_per_cpu) { 82 cpu_topology[lcpu].socket_id = rcpu; 83 socket = socket->next; 84 } else { 85 cpu_topology[lcpu].socket_id = socket->id; 86 } 87 smp_cpu_set_polarization(lcpu, tl_cpu->pp); 88 } 89 return socket; 90 } 91 92 static void clear_masks(void) 93 { 94 struct mask_info *info; 95 96 info = &socket_info; 97 while (info) { 98 cpumask_clear(&info->mask); 99 info = info->next; 100 } 101 info = &book_info; 102 while (info) { 103 cpumask_clear(&info->mask); 104 info = info->next; 105 } 106 } 107 108 static union topology_entry *next_tle(union topology_entry *tle) 109 { 110 if (!tle->nl) 111 return (union topology_entry *)((struct topology_cpu *)tle + 1); 112 return (union topology_entry *)((struct topology_container *)tle + 1); 113 } 114 115 static void __tl_to_masks_generic(struct sysinfo_15_1_x *info) 116 { 117 struct mask_info *socket = &socket_info; 118 struct mask_info *book = &book_info; 119 union topology_entry *tle, *end; 120 121 tle = info->tle; 122 end = (union topology_entry *)((unsigned long)info + info->length); 123 while (tle < end) { 124 switch (tle->nl) { 125 case 2: 126 book = book->next; 127 book->id = tle->container.id; 128 break; 129 case 1: 130 socket = socket->next; 131 socket->id = tle->container.id; 132 break; 133 case 0: 134 add_cpus_to_mask(&tle->cpu, book, socket, 0); 135 break; 136 default: 137 clear_masks(); 138 return; 139 } 140 tle = next_tle(tle); 141 } 142 } 143 144 static void __tl_to_masks_z10(struct sysinfo_15_1_x *info) 145 { 146 struct mask_info *socket = &socket_info; 147 struct mask_info *book = &book_info; 148 union topology_entry *tle, *end; 149 150 tle = info->tle; 151 end = (union topology_entry *)((unsigned long)info + info->length); 152 while (tle < end) { 153 switch (tle->nl) { 154 case 1: 155 book = book->next; 156 book->id = tle->container.id; 157 break; 158 case 0: 159 socket = add_cpus_to_mask(&tle->cpu, book, socket, 1); 160 break; 161 default: 162 clear_masks(); 163 return; 164 } 165 tle = next_tle(tle); 166 } 167 } 168 169 static void tl_to_masks(struct sysinfo_15_1_x *info) 170 { 171 struct cpuid cpu_id; 172 173 spin_lock_irq(&topology_lock); 174 get_cpu_id(&cpu_id); 175 clear_masks(); 176 switch (cpu_id.machine) { 177 case 0x2097: 178 case 0x2098: 179 __tl_to_masks_z10(info); 180 break; 181 default: 182 __tl_to_masks_generic(info); 183 } 184 spin_unlock_irq(&topology_lock); 185 } 186 187 static void topology_update_polarization_simple(void) 188 { 189 int cpu; 190 191 mutex_lock(&smp_cpu_state_mutex); 192 for_each_possible_cpu(cpu) 193 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ); 194 mutex_unlock(&smp_cpu_state_mutex); 195 } 196 197 static int ptf(unsigned long fc) 198 { 199 int rc; 200 201 asm volatile( 202 " .insn rre,0xb9a20000,%1,%1\n" 203 " ipm %0\n" 204 " srl %0,28\n" 205 : "=d" (rc) 206 : "d" (fc) : "cc"); 207 return rc; 208 } 209 210 int topology_set_cpu_management(int fc) 211 { 212 int cpu, rc; 213 214 if (!MACHINE_HAS_TOPOLOGY) 215 return -EOPNOTSUPP; 216 if (fc) 217 rc = ptf(PTF_VERTICAL); 218 else 219 rc = ptf(PTF_HORIZONTAL); 220 if (rc) 221 return -EBUSY; 222 for_each_possible_cpu(cpu) 223 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN); 224 return rc; 225 } 226 227 static void update_cpu_masks(void) 228 { 229 unsigned long flags; 230 int cpu; 231 232 spin_lock_irqsave(&topology_lock, flags); 233 for_each_possible_cpu(cpu) { 234 cpu_topology[cpu].core_mask = cpu_group_map(&socket_info, cpu); 235 cpu_topology[cpu].book_mask = cpu_group_map(&book_info, cpu); 236 if (!MACHINE_HAS_TOPOLOGY) { 237 cpu_topology[cpu].core_id = cpu; 238 cpu_topology[cpu].socket_id = cpu; 239 cpu_topology[cpu].book_id = cpu; 240 } 241 } 242 spin_unlock_irqrestore(&topology_lock, flags); 243 } 244 245 void store_topology(struct sysinfo_15_1_x *info) 246 { 247 if (topology_max_mnest >= 3) 248 stsi(info, 15, 1, 3); 249 else 250 stsi(info, 15, 1, 2); 251 } 252 253 int arch_update_cpu_topology(void) 254 { 255 struct sysinfo_15_1_x *info = tl_info; 256 struct device *dev; 257 int cpu; 258 259 if (!MACHINE_HAS_TOPOLOGY) { 260 update_cpu_masks(); 261 topology_update_polarization_simple(); 262 return 0; 263 } 264 store_topology(info); 265 tl_to_masks(info); 266 update_cpu_masks(); 267 for_each_online_cpu(cpu) { 268 dev = get_cpu_device(cpu); 269 kobject_uevent(&dev->kobj, KOBJ_CHANGE); 270 } 271 return 1; 272 } 273 274 static void topology_work_fn(struct work_struct *work) 275 { 276 rebuild_sched_domains(); 277 } 278 279 void topology_schedule_update(void) 280 { 281 schedule_work(&topology_work); 282 } 283 284 static void topology_timer_fn(unsigned long ignored) 285 { 286 if (ptf(PTF_CHECK)) 287 topology_schedule_update(); 288 set_topology_timer(); 289 } 290 291 static struct timer_list topology_timer = 292 TIMER_DEFERRED_INITIALIZER(topology_timer_fn, 0, 0); 293 294 static atomic_t topology_poll = ATOMIC_INIT(0); 295 296 static void set_topology_timer(void) 297 { 298 if (atomic_add_unless(&topology_poll, -1, 0)) 299 mod_timer(&topology_timer, jiffies + HZ / 10); 300 else 301 mod_timer(&topology_timer, jiffies + HZ * 60); 302 } 303 304 void topology_expect_change(void) 305 { 306 if (!MACHINE_HAS_TOPOLOGY) 307 return; 308 /* This is racy, but it doesn't matter since it is just a heuristic. 309 * Worst case is that we poll in a higher frequency for a bit longer. 310 */ 311 if (atomic_read(&topology_poll) > 60) 312 return; 313 atomic_add(60, &topology_poll); 314 set_topology_timer(); 315 } 316 317 static int __init early_parse_topology(char *p) 318 { 319 if (strncmp(p, "off", 3)) 320 return 0; 321 topology_enabled = 0; 322 return 0; 323 } 324 early_param("topology", early_parse_topology); 325 326 static void __init alloc_masks(struct sysinfo_15_1_x *info, 327 struct mask_info *mask, int offset) 328 { 329 int i, nr_masks; 330 331 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset]; 332 for (i = 0; i < info->mnest - offset; i++) 333 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i]; 334 nr_masks = max(nr_masks, 1); 335 for (i = 0; i < nr_masks; i++) { 336 mask->next = alloc_bootmem_align( 337 roundup_pow_of_two(sizeof(struct mask_info)), 338 roundup_pow_of_two(sizeof(struct mask_info))); 339 mask = mask->next; 340 } 341 } 342 343 void __init s390_init_cpu_topology(void) 344 { 345 struct sysinfo_15_1_x *info; 346 int i; 347 348 if (!MACHINE_HAS_TOPOLOGY) 349 return; 350 tl_info = alloc_bootmem_pages(PAGE_SIZE); 351 info = tl_info; 352 store_topology(info); 353 pr_info("The CPU configuration topology of the machine is:"); 354 for (i = 0; i < TOPOLOGY_NR_MAG; i++) 355 printk(KERN_CONT " %d", info->mag[i]); 356 printk(KERN_CONT " / %d\n", info->mnest); 357 alloc_masks(info, &socket_info, 1); 358 alloc_masks(info, &book_info, 2); 359 } 360 361 static int cpu_management; 362 363 static ssize_t dispatching_show(struct device *dev, 364 struct device_attribute *attr, 365 char *buf) 366 { 367 ssize_t count; 368 369 mutex_lock(&smp_cpu_state_mutex); 370 count = sprintf(buf, "%d\n", cpu_management); 371 mutex_unlock(&smp_cpu_state_mutex); 372 return count; 373 } 374 375 static ssize_t dispatching_store(struct device *dev, 376 struct device_attribute *attr, 377 const char *buf, 378 size_t count) 379 { 380 int val, rc; 381 char delim; 382 383 if (sscanf(buf, "%d %c", &val, &delim) != 1) 384 return -EINVAL; 385 if (val != 0 && val != 1) 386 return -EINVAL; 387 rc = 0; 388 get_online_cpus(); 389 mutex_lock(&smp_cpu_state_mutex); 390 if (cpu_management == val) 391 goto out; 392 rc = topology_set_cpu_management(val); 393 if (rc) 394 goto out; 395 cpu_management = val; 396 topology_expect_change(); 397 out: 398 mutex_unlock(&smp_cpu_state_mutex); 399 put_online_cpus(); 400 return rc ? rc : count; 401 } 402 static DEVICE_ATTR(dispatching, 0644, dispatching_show, 403 dispatching_store); 404 405 static ssize_t cpu_polarization_show(struct device *dev, 406 struct device_attribute *attr, char *buf) 407 { 408 int cpu = dev->id; 409 ssize_t count; 410 411 mutex_lock(&smp_cpu_state_mutex); 412 switch (smp_cpu_get_polarization(cpu)) { 413 case POLARIZATION_HRZ: 414 count = sprintf(buf, "horizontal\n"); 415 break; 416 case POLARIZATION_VL: 417 count = sprintf(buf, "vertical:low\n"); 418 break; 419 case POLARIZATION_VM: 420 count = sprintf(buf, "vertical:medium\n"); 421 break; 422 case POLARIZATION_VH: 423 count = sprintf(buf, "vertical:high\n"); 424 break; 425 default: 426 count = sprintf(buf, "unknown\n"); 427 break; 428 } 429 mutex_unlock(&smp_cpu_state_mutex); 430 return count; 431 } 432 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL); 433 434 static struct attribute *topology_cpu_attrs[] = { 435 &dev_attr_polarization.attr, 436 NULL, 437 }; 438 439 static struct attribute_group topology_cpu_attr_group = { 440 .attrs = topology_cpu_attrs, 441 }; 442 443 int topology_cpu_init(struct cpu *cpu) 444 { 445 return sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group); 446 } 447 448 const struct cpumask *cpu_coregroup_mask(int cpu) 449 { 450 return &cpu_topology[cpu].core_mask; 451 } 452 453 static const struct cpumask *cpu_book_mask(int cpu) 454 { 455 return &cpu_topology[cpu].book_mask; 456 } 457 458 static struct sched_domain_topology_level s390_topology[] = { 459 { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, 460 { cpu_book_mask, SD_INIT_NAME(BOOK) }, 461 { cpu_cpu_mask, SD_INIT_NAME(DIE) }, 462 { NULL, }, 463 }; 464 465 static int __init topology_init(void) 466 { 467 if (MACHINE_HAS_TOPOLOGY) 468 set_topology_timer(); 469 else 470 topology_update_polarization_simple(); 471 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching); 472 } 473 device_initcall(topology_init); 474 475 static int __init early_topology_init(void) 476 { 477 set_sched_topology(s390_topology); 478 return 0; 479 } 480 early_initcall(early_topology_init); 481