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/kernel.h> 14 #include <linux/sched.h> 15 #include <linux/init.h> 16 #include <linux/delay.h> 17 #include <linux/cpu.h> 18 #include <linux/smp.h> 19 #include <linux/mm.h> 20 #include <asm/sysinfo.h> 21 22 #define PTF_HORIZONTAL (0UL) 23 #define PTF_VERTICAL (1UL) 24 #define PTF_CHECK (2UL) 25 26 struct mask_info { 27 struct mask_info *next; 28 unsigned char id; 29 cpumask_t mask; 30 }; 31 32 static int topology_enabled = 1; 33 static void topology_work_fn(struct work_struct *work); 34 static struct sysinfo_15_1_x *tl_info; 35 static void set_topology_timer(void); 36 static DECLARE_WORK(topology_work, topology_work_fn); 37 /* topology_lock protects the core linked list */ 38 static DEFINE_SPINLOCK(topology_lock); 39 40 static struct mask_info core_info; 41 cpumask_t cpu_core_map[NR_CPUS]; 42 unsigned char cpu_core_id[NR_CPUS]; 43 unsigned char cpu_socket_id[NR_CPUS]; 44 45 static struct mask_info book_info; 46 cpumask_t cpu_book_map[NR_CPUS]; 47 unsigned char cpu_book_id[NR_CPUS]; 48 49 static cpumask_t cpu_group_map(struct mask_info *info, unsigned int cpu) 50 { 51 cpumask_t mask; 52 53 cpumask_clear(&mask); 54 if (!topology_enabled || !MACHINE_HAS_TOPOLOGY) { 55 cpumask_copy(&mask, cpumask_of(cpu)); 56 return mask; 57 } 58 while (info) { 59 if (cpumask_test_cpu(cpu, &info->mask)) { 60 mask = info->mask; 61 break; 62 } 63 info = info->next; 64 } 65 if (cpumask_empty(&mask)) 66 cpumask_copy(&mask, cpumask_of(cpu)); 67 return mask; 68 } 69 70 static struct mask_info *add_cpus_to_mask(struct topology_cpu *tl_cpu, 71 struct mask_info *book, 72 struct mask_info *core, 73 int one_core_per_cpu) 74 { 75 unsigned int cpu; 76 77 for_each_set_bit(cpu, &tl_cpu->mask[0], TOPOLOGY_CPU_BITS) { 78 unsigned int rcpu; 79 int lcpu; 80 81 rcpu = TOPOLOGY_CPU_BITS - 1 - cpu + tl_cpu->origin; 82 lcpu = smp_find_processor_id(rcpu); 83 if (lcpu >= 0) { 84 cpumask_set_cpu(lcpu, &book->mask); 85 cpu_book_id[lcpu] = book->id; 86 cpumask_set_cpu(lcpu, &core->mask); 87 cpu_core_id[lcpu] = rcpu; 88 if (one_core_per_cpu) { 89 cpu_socket_id[lcpu] = rcpu; 90 core = core->next; 91 } else { 92 cpu_socket_id[lcpu] = core->id; 93 } 94 smp_cpu_set_polarization(lcpu, tl_cpu->pp); 95 } 96 } 97 return core; 98 } 99 100 static void clear_masks(void) 101 { 102 struct mask_info *info; 103 104 info = &core_info; 105 while (info) { 106 cpumask_clear(&info->mask); 107 info = info->next; 108 } 109 info = &book_info; 110 while (info) { 111 cpumask_clear(&info->mask); 112 info = info->next; 113 } 114 } 115 116 static union topology_entry *next_tle(union topology_entry *tle) 117 { 118 if (!tle->nl) 119 return (union topology_entry *)((struct topology_cpu *)tle + 1); 120 return (union topology_entry *)((struct topology_container *)tle + 1); 121 } 122 123 static void __tl_to_cores_generic(struct sysinfo_15_1_x *info) 124 { 125 struct mask_info *core = &core_info; 126 struct mask_info *book = &book_info; 127 union topology_entry *tle, *end; 128 129 tle = info->tle; 130 end = (union topology_entry *)((unsigned long)info + info->length); 131 while (tle < end) { 132 switch (tle->nl) { 133 case 2: 134 book = book->next; 135 book->id = tle->container.id; 136 break; 137 case 1: 138 core = core->next; 139 core->id = tle->container.id; 140 break; 141 case 0: 142 add_cpus_to_mask(&tle->cpu, book, core, 0); 143 break; 144 default: 145 clear_masks(); 146 return; 147 } 148 tle = next_tle(tle); 149 } 150 } 151 152 static void __tl_to_cores_z10(struct sysinfo_15_1_x *info) 153 { 154 struct mask_info *core = &core_info; 155 struct mask_info *book = &book_info; 156 union topology_entry *tle, *end; 157 158 tle = info->tle; 159 end = (union topology_entry *)((unsigned long)info + info->length); 160 while (tle < end) { 161 switch (tle->nl) { 162 case 1: 163 book = book->next; 164 book->id = tle->container.id; 165 break; 166 case 0: 167 core = add_cpus_to_mask(&tle->cpu, book, core, 1); 168 break; 169 default: 170 clear_masks(); 171 return; 172 } 173 tle = next_tle(tle); 174 } 175 } 176 177 static void tl_to_cores(struct sysinfo_15_1_x *info) 178 { 179 struct cpuid cpu_id; 180 181 get_cpu_id(&cpu_id); 182 spin_lock_irq(&topology_lock); 183 clear_masks(); 184 switch (cpu_id.machine) { 185 case 0x2097: 186 case 0x2098: 187 __tl_to_cores_z10(info); 188 break; 189 default: 190 __tl_to_cores_generic(info); 191 } 192 spin_unlock_irq(&topology_lock); 193 } 194 195 static void topology_update_polarization_simple(void) 196 { 197 int cpu; 198 199 mutex_lock(&smp_cpu_state_mutex); 200 for_each_possible_cpu(cpu) 201 smp_cpu_set_polarization(cpu, POLARIZATION_HRZ); 202 mutex_unlock(&smp_cpu_state_mutex); 203 } 204 205 static int ptf(unsigned long fc) 206 { 207 int rc; 208 209 asm volatile( 210 " .insn rre,0xb9a20000,%1,%1\n" 211 " ipm %0\n" 212 " srl %0,28\n" 213 : "=d" (rc) 214 : "d" (fc) : "cc"); 215 return rc; 216 } 217 218 int topology_set_cpu_management(int fc) 219 { 220 int cpu, rc; 221 222 if (!MACHINE_HAS_TOPOLOGY) 223 return -EOPNOTSUPP; 224 if (fc) 225 rc = ptf(PTF_VERTICAL); 226 else 227 rc = ptf(PTF_HORIZONTAL); 228 if (rc) 229 return -EBUSY; 230 for_each_possible_cpu(cpu) 231 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN); 232 return rc; 233 } 234 235 static void update_cpu_core_map(void) 236 { 237 unsigned long flags; 238 int cpu; 239 240 spin_lock_irqsave(&topology_lock, flags); 241 for_each_possible_cpu(cpu) { 242 cpu_core_map[cpu] = cpu_group_map(&core_info, cpu); 243 cpu_book_map[cpu] = cpu_group_map(&book_info, cpu); 244 } 245 spin_unlock_irqrestore(&topology_lock, flags); 246 } 247 248 void store_topology(struct sysinfo_15_1_x *info) 249 { 250 if (topology_max_mnest >= 3) 251 stsi(info, 15, 1, 3); 252 else 253 stsi(info, 15, 1, 2); 254 } 255 256 int arch_update_cpu_topology(void) 257 { 258 struct sysinfo_15_1_x *info = tl_info; 259 struct device *dev; 260 int cpu; 261 262 if (!MACHINE_HAS_TOPOLOGY) { 263 update_cpu_core_map(); 264 topology_update_polarization_simple(); 265 return 0; 266 } 267 store_topology(info); 268 tl_to_cores(info); 269 update_cpu_core_map(); 270 for_each_online_cpu(cpu) { 271 dev = get_cpu_device(cpu); 272 kobject_uevent(&dev->kobj, KOBJ_CHANGE); 273 } 274 return 1; 275 } 276 277 static void topology_work_fn(struct work_struct *work) 278 { 279 rebuild_sched_domains(); 280 } 281 282 void topology_schedule_update(void) 283 { 284 schedule_work(&topology_work); 285 } 286 287 static void topology_timer_fn(unsigned long ignored) 288 { 289 if (ptf(PTF_CHECK)) 290 topology_schedule_update(); 291 set_topology_timer(); 292 } 293 294 static struct timer_list topology_timer = 295 TIMER_DEFERRED_INITIALIZER(topology_timer_fn, 0, 0); 296 297 static atomic_t topology_poll = ATOMIC_INIT(0); 298 299 static void set_topology_timer(void) 300 { 301 if (atomic_add_unless(&topology_poll, -1, 0)) 302 mod_timer(&topology_timer, jiffies + HZ / 10); 303 else 304 mod_timer(&topology_timer, jiffies + HZ * 60); 305 } 306 307 void topology_expect_change(void) 308 { 309 if (!MACHINE_HAS_TOPOLOGY) 310 return; 311 /* This is racy, but it doesn't matter since it is just a heuristic. 312 * Worst case is that we poll in a higher frequency for a bit longer. 313 */ 314 if (atomic_read(&topology_poll) > 60) 315 return; 316 atomic_add(60, &topology_poll); 317 set_topology_timer(); 318 } 319 320 static int __init early_parse_topology(char *p) 321 { 322 if (strncmp(p, "off", 3)) 323 return 0; 324 topology_enabled = 0; 325 return 0; 326 } 327 early_param("topology", early_parse_topology); 328 329 static void __init alloc_masks(struct sysinfo_15_1_x *info, 330 struct mask_info *mask, int offset) 331 { 332 int i, nr_masks; 333 334 nr_masks = info->mag[TOPOLOGY_NR_MAG - offset]; 335 for (i = 0; i < info->mnest - offset; i++) 336 nr_masks *= info->mag[TOPOLOGY_NR_MAG - offset - 1 - i]; 337 nr_masks = max(nr_masks, 1); 338 for (i = 0; i < nr_masks; i++) { 339 mask->next = alloc_bootmem(sizeof(struct mask_info)); 340 mask = mask->next; 341 } 342 } 343 344 void __init s390_init_cpu_topology(void) 345 { 346 struct sysinfo_15_1_x *info; 347 int i; 348 349 if (!MACHINE_HAS_TOPOLOGY) 350 return; 351 tl_info = alloc_bootmem_pages(PAGE_SIZE); 352 info = tl_info; 353 store_topology(info); 354 pr_info("The CPU configuration topology of the machine is:"); 355 for (i = 0; i < TOPOLOGY_NR_MAG; i++) 356 printk(KERN_CONT " %d", info->mag[i]); 357 printk(KERN_CONT " / %d\n", info->mnest); 358 alloc_masks(info, &core_info, 1); 359 alloc_masks(info, &book_info, 2); 360 } 361 362 static int cpu_management; 363 364 static ssize_t dispatching_show(struct device *dev, 365 struct device_attribute *attr, 366 char *buf) 367 { 368 ssize_t count; 369 370 mutex_lock(&smp_cpu_state_mutex); 371 count = sprintf(buf, "%d\n", cpu_management); 372 mutex_unlock(&smp_cpu_state_mutex); 373 return count; 374 } 375 376 static ssize_t dispatching_store(struct device *dev, 377 struct device_attribute *attr, 378 const char *buf, 379 size_t count) 380 { 381 int val, rc; 382 char delim; 383 384 if (sscanf(buf, "%d %c", &val, &delim) != 1) 385 return -EINVAL; 386 if (val != 0 && val != 1) 387 return -EINVAL; 388 rc = 0; 389 get_online_cpus(); 390 mutex_lock(&smp_cpu_state_mutex); 391 if (cpu_management == val) 392 goto out; 393 rc = topology_set_cpu_management(val); 394 if (rc) 395 goto out; 396 cpu_management = val; 397 topology_expect_change(); 398 out: 399 mutex_unlock(&smp_cpu_state_mutex); 400 put_online_cpus(); 401 return rc ? rc : count; 402 } 403 static DEVICE_ATTR(dispatching, 0644, dispatching_show, 404 dispatching_store); 405 406 static ssize_t cpu_polarization_show(struct device *dev, 407 struct device_attribute *attr, char *buf) 408 { 409 int cpu = dev->id; 410 ssize_t count; 411 412 mutex_lock(&smp_cpu_state_mutex); 413 switch (smp_cpu_get_polarization(cpu)) { 414 case POLARIZATION_HRZ: 415 count = sprintf(buf, "horizontal\n"); 416 break; 417 case POLARIZATION_VL: 418 count = sprintf(buf, "vertical:low\n"); 419 break; 420 case POLARIZATION_VM: 421 count = sprintf(buf, "vertical:medium\n"); 422 break; 423 case POLARIZATION_VH: 424 count = sprintf(buf, "vertical:high\n"); 425 break; 426 default: 427 count = sprintf(buf, "unknown\n"); 428 break; 429 } 430 mutex_unlock(&smp_cpu_state_mutex); 431 return count; 432 } 433 static DEVICE_ATTR(polarization, 0444, cpu_polarization_show, NULL); 434 435 static struct attribute *topology_cpu_attrs[] = { 436 &dev_attr_polarization.attr, 437 NULL, 438 }; 439 440 static struct attribute_group topology_cpu_attr_group = { 441 .attrs = topology_cpu_attrs, 442 }; 443 444 int topology_cpu_init(struct cpu *cpu) 445 { 446 return sysfs_create_group(&cpu->dev.kobj, &topology_cpu_attr_group); 447 } 448 449 static int __init topology_init(void) 450 { 451 if (!MACHINE_HAS_TOPOLOGY) { 452 topology_update_polarization_simple(); 453 goto out; 454 } 455 set_topology_timer(); 456 out: 457 update_cpu_core_map(); 458 return device_create_file(cpu_subsys.dev_root, &dev_attr_dispatching); 459 } 460 device_initcall(topology_init); 461