arch/s390/kernel/topology.c

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