arch/powerpc/platforms/pseries/smp.c

   1 /*
   2  * SMP support for pSeries machines.
   3  *
   4  * Dave Engebretsen, Peter Bergner, and
   5  * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com
   6  *
   7  * Plus various changes from other IBM teams...
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  */
  14
  15 #undef DEBUG
  16
  17 #include <linux/config.h>
  18 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/sched.h>
  21 #include <linux/smp.h>
  22 #include <linux/interrupt.h>
  23 #include <linux/delay.h>
  24 #include <linux/init.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/cache.h>
  27 #include <linux/err.h>
  28 #include <linux/sysdev.h>
  29 #include <linux/cpu.h>
  30
  31 #include <asm/ptrace.h>
  32 #include <asm/atomic.h>
  33 #include <asm/irq.h>
  34 #include <asm/page.h>
  35 #include <asm/pgtable.h>
  36 #include <asm/io.h>
  37 #include <asm/prom.h>
  38 #include <asm/smp.h>
  39 #include <asm/paca.h>
  40 #include <asm/time.h>
  41 #include <asm/machdep.h>
  42 #include "xics.h"
  43 #include <asm/cputable.h>
  44 #include <asm/firmware.h>
  45 #include <asm/system.h>
  46 #include <asm/rtas.h>
  47 #include <asm/pSeries_reconfig.h>
  48 #include <asm/mpic.h>
  49 #include <asm/vdso_datapage.h>
  50
  51 #include "plpar_wrappers.h"
  52
  53 #ifdef DEBUG
  54 #include <asm/udbg.h>
  55 #define DBG(fmt...) udbg_printf(fmt)
  56 #else
  57 #define DBG(fmt...)
  58 #endif
  59
  60 /*
  61  * The primary thread of each non-boot processor is recorded here before
  62  * smp init.
  63  */
  64 static cpumask_t of_spin_map;
  65
  66 extern void pSeries_secondary_smp_init(unsigned long);
  67
  68 #ifdef CONFIG_HOTPLUG_CPU
  69
  70 /* Get state of physical CPU.
  71  * Return codes:
  72  *      0       - The processor is in the RTAS stopped state
  73  *      1       - stop-self is in progress
  74  *      2       - The processor is not in the RTAS stopped state
  75  *      -1      - Hardware Error
  76  *      -2      - Hardware Busy, Try again later.
  77  */
  78 static int query_cpu_stopped(unsigned int pcpu)
  79 {
  80         int cpu_status;
  81         int status, qcss_tok;
  82
  83         qcss_tok = rtas_token("query-cpu-stopped-state");
  84         if (qcss_tok == RTAS_UNKNOWN_SERVICE)
  85                 return -1;
  86         status = rtas_call(qcss_tok, 1, 2, &cpu_status, pcpu);
  87         if (status != 0) {
  88                 printk(KERN_ERR
  89                        "RTAS query-cpu-stopped-state failed: %i\n", status);
  90                 return status;
  91         }
  92
  93         return cpu_status;
  94 }
  95
  96 static int pSeries_cpu_disable(void)
  97 {
  98         int cpu = smp_processor_id();
  99
 100         cpu_clear(cpu, cpu_online_map);
 101         vdso_data->processorCount--;
 102
 103         /*fix boot_cpuid here*/
 104         if (cpu == boot_cpuid)
 105                 boot_cpuid = any_online_cpu(cpu_online_map);
 106
 107         /* FIXME: abstract this to not be platform specific later on */
 108         xics_migrate_irqs_away();
 109         return 0;
 110 }
 111
 112 static void pSeries_cpu_die(unsigned int cpu)
 113 {
 114         int tries;
 115         int cpu_status;
 116         unsigned int pcpu = get_hard_smp_processor_id(cpu);
 117
 118         for (tries = 0; tries < 25; tries++) {
 119                 cpu_status = query_cpu_stopped(pcpu);
 120                 if (cpu_status == 0 || cpu_status == -1)
 121                         break;
 122                 msleep(200);
 123         }
 124         if (cpu_status != 0) {
 125                 printk("Querying DEAD? cpu %i (%i) shows %i\n",
 126                        cpu, pcpu, cpu_status);
 127         }
 128
 129         /* Isolation and deallocation are definatly done by
 130          * drslot_chrp_cpu.  If they were not they would be
 131          * done here.  Change isolate state to Isolate and
 132          * change allocation-state to Unusable.
 133          */
 134         paca[cpu].cpu_start = 0;
 135 }
 136
 137 /*
 138  * Update cpu_present_map and paca(s) for a new cpu node.  The wrinkle
 139  * here is that a cpu device node may represent up to two logical cpus
 140  * in the SMT case.  We must honor the assumption in other code that
 141  * the logical ids for sibling SMT threads x and y are adjacent, such
 142  * that x^1 == y and y^1 == x.
 143  */
 144 static int pSeries_add_processor(struct device_node *np)
 145 {
 146         unsigned int cpu;
 147         cpumask_t candidate_map, tmp = CPU_MASK_NONE;
 148         int err = -ENOSPC, len, nthreads, i;
 149         u32 *intserv;
 150
 151         intserv = (u32 *)get_property(np, "ibm,ppc-interrupt-server#s", &len);
 152         if (!intserv)
 153                 return 0;
 154
 155         nthreads = len / sizeof(u32);
 156         for (i = 0; i < nthreads; i++)
 157                 cpu_set(i, tmp);
 158
 159         lock_cpu_hotplug();
 160
 161         BUG_ON(!cpus_subset(cpu_present_map, cpu_possible_map));
 162
 163         /* Get a bitmap of unoccupied slots. */
 164         cpus_xor(candidate_map, cpu_possible_map, cpu_present_map);
 165         if (cpus_empty(candidate_map)) {
 166                 /* If we get here, it most likely means that NR_CPUS is
 167                  * less than the partition's max processors setting.
 168                  */
 169                 printk(KERN_ERR "Cannot add cpu %s; this system configuration"
 170                        " supports %d logical cpus.\n", np->full_name,
 171                        cpus_weight(cpu_possible_map));
 172                 goto out_unlock;
 173         }
 174
 175         while (!cpus_empty(tmp))
 176                 if (cpus_subset(tmp, candidate_map))
 177                         /* Found a range where we can insert the new cpu(s) */
 178                         break;
 179                 else
 180                         cpus_shift_left(tmp, tmp, nthreads);
 181
 182         if (cpus_empty(tmp)) {
 183                 printk(KERN_ERR "Unable to find space in cpu_present_map for"
 184                        " processor %s with %d thread(s)\n", np->name,
 185                        nthreads);
 186                 goto out_unlock;
 187         }
 188
 189         for_each_cpu_mask(cpu, tmp) {
 190                 BUG_ON(cpu_isset(cpu, cpu_present_map));
 191                 cpu_set(cpu, cpu_present_map);
 192                 set_hard_smp_processor_id(cpu, *intserv++);
 193         }
 194         err = 0;
 195 out_unlock:
 196         unlock_cpu_hotplug();
 197         return err;
 198 }
 199
 200 /*
 201  * Update the present map for a cpu node which is going away, and set
 202  * the hard id in the paca(s) to -1 to be consistent with boot time
 203  * convention for non-present cpus.
 204  */
 205 static void pSeries_remove_processor(struct device_node *np)
 206 {
 207         unsigned int cpu;
 208         int len, nthreads, i;
 209         u32 *intserv;
 210
 211         intserv = (u32 *)get_property(np, "ibm,ppc-interrupt-server#s", &len);
 212         if (!intserv)
 213                 return;
 214
 215         nthreads = len / sizeof(u32);
 216
 217         lock_cpu_hotplug();
 218         for (i = 0; i < nthreads; i++) {
 219                 for_each_present_cpu(cpu) {
 220                         if (get_hard_smp_processor_id(cpu) != intserv[i])
 221                                 continue;
 222                         BUG_ON(cpu_online(cpu));
 223                         cpu_clear(cpu, cpu_present_map);
 224                         set_hard_smp_processor_id(cpu, -1);
 225                         break;
 226                 }
 227                 if (cpu == NR_CPUS)
 228                         printk(KERN_WARNING "Could not find cpu to remove "
 229                                "with physical id 0x%x\n", intserv[i]);
 230         }
 231         unlock_cpu_hotplug();
 232 }
 233
 234 static int pSeries_smp_notifier(struct notifier_block *nb, unsigned long action, void *node)
 235 {
 236         int err = NOTIFY_OK;
 237
 238         switch (action) {
 239         case PSERIES_RECONFIG_ADD:
 240                 if (pSeries_add_processor(node))
 241                         err = NOTIFY_BAD;
 242                 break;
 243         case PSERIES_RECONFIG_REMOVE:
 244                 pSeries_remove_processor(node);
 245                 break;
 246         default:
 247                 err = NOTIFY_DONE;
 248                 break;
 249         }
 250         return err;
 251 }
 252
 253 static struct notifier_block pSeries_smp_nb = {
 254         .notifier_call = pSeries_smp_notifier,
 255 };
 256
 257 #endif /* CONFIG_HOTPLUG_CPU */
 258
 259 /**
 260  * smp_startup_cpu() - start the given cpu
 261  *
 262  * At boot time, there is nothing to do for primary threads which were
 263  * started from Open Firmware.  For anything else, call RTAS with the
 264  * appropriate start location.
 265  *
 266  * Returns:
 267  *      0       - failure
 268  *      1       - success
 269  */
 270 static inline int __devinit smp_startup_cpu(unsigned int lcpu)
 271 {
 272         int status;
 273         unsigned long start_here = __pa((u32)*((unsigned long *)
 274                                                pSeries_secondary_smp_init));
 275         unsigned int pcpu;
 276         int start_cpu;
 277
 278         if (cpu_isset(lcpu, of_spin_map))
 279                 /* Already started by OF and sitting in spin loop */
 280                 return 1;
 281
 282         pcpu = get_hard_smp_processor_id(lcpu);
 283
 284         /* Fixup atomic count: it exited inside IRQ handler. */
 285         task_thread_info(paca[lcpu].__current)->preempt_count   = 0;
 286
 287         /*
 288          * If the RTAS start-cpu token does not exist then presume the
 289          * cpu is already spinning.
 290          */
 291         start_cpu = rtas_token("start-cpu");
 292         if (start_cpu == RTAS_UNKNOWN_SERVICE)
 293                 return 1;
 294
 295         status = rtas_call(start_cpu, 3, 1, NULL, pcpu, start_here, lcpu);
 296         if (status != 0) {
 297                 printk(KERN_ERR "start-cpu failed: %i\n", status);
 298                 return 0;
 299         }
 300
 301         return 1;
 302 }
 303
 304 #ifdef CONFIG_XICS
 305 static inline void smp_xics_do_message(int cpu, int msg)
 306 {
 307         set_bit(msg, &xics_ipi_message[cpu].value);
 308         mb();
 309         xics_cause_IPI(cpu);
 310 }
 311
 312 static void smp_xics_message_pass(int target, int msg)
 313 {
 314         unsigned int i;
 315
 316         if (target < NR_CPUS) {
 317                 smp_xics_do_message(target, msg);
 318         } else {
 319                 for_each_online_cpu(i) {
 320                         if (target == MSG_ALL_BUT_SELF
 321                             && i == smp_processor_id())
 322                                 continue;
 323                         smp_xics_do_message(i, msg);
 324                 }
 325         }
 326 }
 327
 328 static int __init smp_xics_probe(void)
 329 {
 330         xics_request_IPIs();
 331
 332         return cpus_weight(cpu_possible_map);
 333 }
 334
 335 static void __devinit smp_xics_setup_cpu(int cpu)
 336 {
 337         if (cpu != boot_cpuid)
 338                 xics_setup_cpu();
 339
 340         if (firmware_has_feature(FW_FEATURE_SPLPAR))
 341                 vpa_init(cpu);
 342
 343         cpu_clear(cpu, of_spin_map);
 344
 345 }
 346 #endif /* CONFIG_XICS */
 347
 348 static DEFINE_SPINLOCK(timebase_lock);
 349 static unsigned long timebase = 0;
 350
 351 static void __devinit pSeries_give_timebase(void)
 352 {
 353         spin_lock(&timebase_lock);
 354         rtas_call(rtas_token("freeze-time-base"), 0, 1, NULL);
 355         timebase = get_tb();
 356         spin_unlock(&timebase_lock);
 357
 358         while (timebase)
 359                 barrier();
 360         rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
 361 }
 362
 363 static void __devinit pSeries_take_timebase(void)
 364 {
 365         while (!timebase)
 366                 barrier();
 367         spin_lock(&timebase_lock);
 368         set_tb(timebase >> 32, timebase & 0xffffffff);
 369         timebase = 0;
 370         spin_unlock(&timebase_lock);
 371 }
 372
 373 static void __devinit smp_pSeries_kick_cpu(int nr)
 374 {
 375         BUG_ON(nr < 0 || nr >= NR_CPUS);
 376
 377         if (!smp_startup_cpu(nr))
 378                 return;
 379
 380         /*
 381          * The processor is currently spinning, waiting for the
 382          * cpu_start field to become non-zero After we set cpu_start,
 383          * the processor will continue on to secondary_start
 384          */
 385         paca[nr].cpu_start = 1;
 386 }
 387
 388 static int smp_pSeries_cpu_bootable(unsigned int nr)
 389 {
 390         /* Special case - we inhibit secondary thread startup
 391          * during boot if the user requests it.  Odd-numbered
 392          * cpus are assumed to be secondary threads.
 393          */
 394         if (system_state < SYSTEM_RUNNING &&
 395             cpu_has_feature(CPU_FTR_SMT) &&
 396             !smt_enabled_at_boot && nr % 2 != 0)
 397                 return 0;
 398
 399         return 1;
 400 }
 401 #ifdef CONFIG_MPIC
 402 static struct smp_ops_t pSeries_mpic_smp_ops = {
 403         .message_pass   = smp_mpic_message_pass,
 404         .probe          = smp_mpic_probe,
 405         .kick_cpu       = smp_pSeries_kick_cpu,
 406         .setup_cpu      = smp_mpic_setup_cpu,
 407 };
 408 #endif
 409 #ifdef CONFIG_XICS
 410 static struct smp_ops_t pSeries_xics_smp_ops = {
 411         .message_pass   = smp_xics_message_pass,
 412         .probe          = smp_xics_probe,
 413         .kick_cpu       = smp_pSeries_kick_cpu,
 414         .setup_cpu      = smp_xics_setup_cpu,
 415         .cpu_bootable   = smp_pSeries_cpu_bootable,
 416 };
 417 #endif
 418
 419 /* This is called very early */
 420 void __init smp_init_pSeries(void)
 421 {
 422         int i;
 423
 424         DBG(" -> smp_init_pSeries()\n");
 425
 426         switch (ppc64_interrupt_controller) {
 427 #ifdef CONFIG_MPIC
 428         case IC_OPEN_PIC:
 429                 smp_ops = &pSeries_mpic_smp_ops;
 430                 break;
 431 #endif
 432 #ifdef CONFIG_XICS
 433         case IC_PPC_XIC:
 434                 smp_ops = &pSeries_xics_smp_ops;
 435                 break;
 436 #endif
 437         default:
 438                 panic("Invalid interrupt controller");
 439         }
 440
 441 #ifdef CONFIG_HOTPLUG_CPU
 442         smp_ops->cpu_disable = pSeries_cpu_disable;
 443         smp_ops->cpu_die = pSeries_cpu_die;
 444
 445         /* Processors can be added/removed only on LPAR */
 446         if (platform_is_lpar())
 447                 pSeries_reconfig_notifier_register(&pSeries_smp_nb);
 448 #endif
 449
 450         /* Mark threads which are still spinning in hold loops. */
 451         if (cpu_has_feature(CPU_FTR_SMT)) {
 452                 for_each_present_cpu(i) {
 453                         if (i % 2 == 0)
 454                                 /*
 455                                  * Even-numbered logical cpus correspond to
 456                                  * primary threads.
 457                                  */
 458                                 cpu_set(i, of_spin_map);
 459                 }
 460         } else {
 461                 of_spin_map = cpu_present_map;
 462         }
 463
 464         cpu_clear(boot_cpuid, of_spin_map);
 465
 466         /* Non-lpar has additional take/give timebase */
 467         if (rtas_token("freeze-time-base") != RTAS_UNKNOWN_SERVICE) {
 468                 smp_ops->give_timebase = pSeries_give_timebase;
 469                 smp_ops->take_timebase = pSeries_take_timebase;
 470         }
 471
 472         DBG(" <- smp_init_pSeries()\n");
 473 }
 474