arch/ppc64/kernel/pSeries_smp.c

   1 /*
   2  * SMP support for pSeries and BPA 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 <asm/xics.h>
  43 #include <asm/cputable.h>
  44 #include <asm/system.h>
  45 #include <asm/rtas.h>
  46 #include <asm/plpar_wrappers.h>
  47 #include <asm/pSeries_reconfig.h>
  48
  49 #include "mpic.h"
  50 #include "bpa_iic.h"
  51
  52 #ifdef DEBUG
  53 #define DBG(fmt...) udbg_printf(fmt)
  54 #else
  55 #define DBG(fmt...)
  56 #endif
  57
  58 /*
  59  * The primary thread of each non-boot processor is recorded here before
  60  * smp init.
  61  */
  62 static cpumask_t of_spin_map;
  63
  64 extern void pSeries_secondary_smp_init(unsigned long);
  65
  66 #ifdef CONFIG_HOTPLUG_CPU
  67
  68 /* Get state of physical CPU.
  69  * Return codes:
  70  *      0       - The processor is in the RTAS stopped state
  71  *      1       - stop-self is in progress
  72  *      2       - The processor is not in the RTAS stopped state
  73  *      -1      - Hardware Error
  74  *      -2      - Hardware Busy, Try again later.
  75  */
  76 static int query_cpu_stopped(unsigned int pcpu)
  77 {
  78         int cpu_status;
  79         int status, qcss_tok;
  80
  81         qcss_tok = rtas_token("query-cpu-stopped-state");
  82         if (qcss_tok == RTAS_UNKNOWN_SERVICE)
  83                 return -1;
  84         status = rtas_call(qcss_tok, 1, 2, &cpu_status, pcpu);
  85         if (status != 0) {
  86                 printk(KERN_ERR
  87                        "RTAS query-cpu-stopped-state failed: %i\n", status);
  88                 return status;
  89         }
  90
  91         return cpu_status;
  92 }
  93
  94 int pSeries_cpu_disable(void)
  95 {
  96         int cpu = smp_processor_id();
  97
  98         cpu_clear(cpu, cpu_online_map);
  99         systemcfg->processorCount--;
 100
 101         /*fix boot_cpuid here*/
 102         if (cpu == boot_cpuid)
 103                 boot_cpuid = any_online_cpu(cpu_online_map);
 104
 105         /* FIXME: abstract this to not be platform specific later on */
 106         xics_migrate_irqs_away();
 107         return 0;
 108 }
 109
 110 void pSeries_cpu_die(unsigned int cpu)
 111 {
 112         int tries;
 113         int cpu_status;
 114         unsigned int pcpu = get_hard_smp_processor_id(cpu);
 115
 116         for (tries = 0; tries < 25; tries++) {
 117                 cpu_status = query_cpu_stopped(pcpu);
 118                 if (cpu_status == 0 || cpu_status == -1)
 119                         break;
 120                 msleep(200);
 121         }
 122         if (cpu_status != 0) {
 123                 printk("Querying DEAD? cpu %i (%i) shows %i\n",
 124                        cpu, pcpu, cpu_status);
 125         }
 126
 127         /* Isolation and deallocation are definatly done by
 128          * drslot_chrp_cpu.  If they were not they would be
 129          * done here.  Change isolate state to Isolate and
 130          * change allocation-state to Unusable.
 131          */
 132         paca[cpu].cpu_start = 0;
 133 }
 134
 135 /*
 136  * Update cpu_present_map and paca(s) for a new cpu node.  The wrinkle
 137  * here is that a cpu device node may represent up to two logical cpus
 138  * in the SMT case.  We must honor the assumption in other code that
 139  * the logical ids for sibling SMT threads x and y are adjacent, such
 140  * that x^1 == y and y^1 == x.
 141  */
 142 static int pSeries_add_processor(struct device_node *np)
 143 {
 144         unsigned int cpu;
 145         cpumask_t candidate_map, tmp = CPU_MASK_NONE;
 146         int err = -ENOSPC, len, nthreads, i;
 147         u32 *intserv;
 148
 149         intserv = (u32 *)get_property(np, "ibm,ppc-interrupt-server#s", &len);
 150         if (!intserv)
 151                 return 0;
 152
 153         nthreads = len / sizeof(u32);
 154         for (i = 0; i < nthreads; i++)
 155                 cpu_set(i, tmp);
 156
 157         lock_cpu_hotplug();
 158
 159         BUG_ON(!cpus_subset(cpu_present_map, cpu_possible_map));
 160
 161         /* Get a bitmap of unoccupied slots. */
 162         cpus_xor(candidate_map, cpu_possible_map, cpu_present_map);
 163         if (cpus_empty(candidate_map)) {
 164                 /* If we get here, it most likely means that NR_CPUS is
 165                  * less than the partition's max processors setting.
 166                  */
 167                 printk(KERN_ERR "Cannot add cpu %s; this system configuration"
 168                        " supports %d logical cpus.\n", np->full_name,
 169                        cpus_weight(cpu_possible_map));
 170                 goto out_unlock;
 171         }
 172
 173         while (!cpus_empty(tmp))
 174                 if (cpus_subset(tmp, candidate_map))
 175                         /* Found a range where we can insert the new cpu(s) */
 176                         break;
 177                 else
 178                         cpus_shift_left(tmp, tmp, nthreads);
 179
 180         if (cpus_empty(tmp)) {
 181                 printk(KERN_ERR "Unable to find space in cpu_present_map for"
 182                        " processor %s with %d thread(s)\n", np->name,
 183                        nthreads);
 184                 goto out_unlock;
 185         }
 186
 187         for_each_cpu_mask(cpu, tmp) {
 188                 BUG_ON(cpu_isset(cpu, cpu_present_map));
 189                 cpu_set(cpu, cpu_present_map);
 190                 set_hard_smp_processor_id(cpu, *intserv++);
 191         }
 192         err = 0;
 193 out_unlock:
 194         unlock_cpu_hotplug();
 195         return err;
 196 }
 197
 198 /*
 199  * Update the present map for a cpu node which is going away, and set
 200  * the hard id in the paca(s) to -1 to be consistent with boot time
 201  * convention for non-present cpus.
 202  */
 203 static void pSeries_remove_processor(struct device_node *np)
 204 {
 205         unsigned int cpu;
 206         int len, nthreads, i;
 207         u32 *intserv;
 208
 209         intserv = (u32 *)get_property(np, "ibm,ppc-interrupt-server#s", &len);
 210         if (!intserv)
 211                 return;
 212
 213         nthreads = len / sizeof(u32);
 214
 215         lock_cpu_hotplug();
 216         for (i = 0; i < nthreads; i++) {
 217                 for_each_present_cpu(cpu) {
 218                         if (get_hard_smp_processor_id(cpu) != intserv[i])
 219                                 continue;
 220                         BUG_ON(cpu_online(cpu));
 221                         cpu_clear(cpu, cpu_present_map);
 222                         set_hard_smp_processor_id(cpu, -1);
 223                         break;
 224                 }
 225                 if (cpu == NR_CPUS)
 226                         printk(KERN_WARNING "Could not find cpu to remove "
 227                                "with physical id 0x%x\n", intserv[i]);
 228         }
 229         unlock_cpu_hotplug();
 230 }
 231
 232 static int pSeries_smp_notifier(struct notifier_block *nb, unsigned long action, void *node)
 233 {
 234         int err = NOTIFY_OK;
 235
 236         switch (action) {
 237         case PSERIES_RECONFIG_ADD:
 238                 if (pSeries_add_processor(node))
 239                         err = NOTIFY_BAD;
 240                 break;
 241         case PSERIES_RECONFIG_REMOVE:
 242                 pSeries_remove_processor(node);
 243                 break;
 244         default:
 245                 err = NOTIFY_DONE;
 246                 break;
 247         }
 248         return err;
 249 }
 250
 251 static struct notifier_block pSeries_smp_nb = {
 252         .notifier_call = pSeries_smp_notifier,
 253 };
 254
 255 #endif /* CONFIG_HOTPLUG_CPU */
 256
 257 /**
 258  * smp_startup_cpu() - start the given cpu
 259  *
 260  * At boot time, there is nothing to do for primary threads which were
 261  * started from Open Firmware.  For anything else, call RTAS with the
 262  * appropriate start location.
 263  *
 264  * Returns:
 265  *      0       - failure
 266  *      1       - success
 267  */
 268 static inline int __devinit smp_startup_cpu(unsigned int lcpu)
 269 {
 270         int status;
 271         unsigned long start_here = __pa((u32)*((unsigned long *)
 272                                                pSeries_secondary_smp_init));
 273         unsigned int pcpu;
 274
 275         if (cpu_isset(lcpu, of_spin_map))
 276                 /* Already started by OF and sitting in spin loop */
 277                 return 1;
 278
 279         pcpu = get_hard_smp_processor_id(lcpu);
 280
 281         /* Fixup atomic count: it exited inside IRQ handler. */
 282         paca[lcpu].__current->thread_info->preempt_count        = 0;
 283
 284         status = rtas_call(rtas_token("start-cpu"), 3, 1, NULL,
 285                            pcpu, start_here, lcpu);
 286         if (status != 0) {
 287                 printk(KERN_ERR "start-cpu failed: %i\n", status);
 288                 return 0;
 289         }
 290         return 1;
 291 }
 292
 293 #ifdef CONFIG_XICS
 294 static inline void smp_xics_do_message(int cpu, int msg)
 295 {
 296         set_bit(msg, &xics_ipi_message[cpu].value);
 297         mb();
 298         xics_cause_IPI(cpu);
 299 }
 300
 301 static void smp_xics_message_pass(int target, int msg)
 302 {
 303         unsigned int i;
 304
 305         if (target < NR_CPUS) {
 306                 smp_xics_do_message(target, msg);
 307         } else {
 308                 for_each_online_cpu(i) {
 309                         if (target == MSG_ALL_BUT_SELF
 310                             && i == smp_processor_id())
 311                                 continue;
 312                         smp_xics_do_message(i, msg);
 313                 }
 314         }
 315 }
 316
 317 static int __init smp_xics_probe(void)
 318 {
 319         xics_request_IPIs();
 320
 321         return cpus_weight(cpu_possible_map);
 322 }
 323
 324 static void __devinit smp_xics_setup_cpu(int cpu)
 325 {
 326         if (cpu != boot_cpuid)
 327                 xics_setup_cpu();
 328
 329         if (cur_cpu_spec->firmware_features & FW_FEATURE_SPLPAR)
 330                 vpa_init(cpu);
 331
 332         cpu_clear(cpu, of_spin_map);
 333
 334 }
 335 #endif /* CONFIG_XICS */
 336 #ifdef CONFIG_BPA_IIC
 337 static void smp_iic_message_pass(int target, int msg)
 338 {
 339         unsigned int i;
 340
 341         if (target < NR_CPUS) {
 342                 iic_cause_IPI(target, msg);
 343         } else {
 344                 for_each_online_cpu(i) {
 345                         if (target == MSG_ALL_BUT_SELF
 346                             && i == smp_processor_id())
 347                                 continue;
 348                         iic_cause_IPI(i, msg);
 349                 }
 350         }
 351 }
 352
 353 static int __init smp_iic_probe(void)
 354 {
 355         iic_request_IPIs();
 356
 357         return cpus_weight(cpu_possible_map);
 358 }
 359
 360 static void __devinit smp_iic_setup_cpu(int cpu)
 361 {
 362         if (cpu != boot_cpuid)
 363                 iic_setup_cpu();
 364 }
 365 #endif /* CONFIG_BPA_IIC */
 366
 367 static DEFINE_SPINLOCK(timebase_lock);
 368 static unsigned long timebase = 0;
 369
 370 static void __devinit pSeries_give_timebase(void)
 371 {
 372         spin_lock(&timebase_lock);
 373         rtas_call(rtas_token("freeze-time-base"), 0, 1, NULL);
 374         timebase = get_tb();
 375         spin_unlock(&timebase_lock);
 376
 377         while (timebase)
 378                 barrier();
 379         rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
 380 }
 381
 382 static void __devinit pSeries_take_timebase(void)
 383 {
 384         while (!timebase)
 385                 barrier();
 386         spin_lock(&timebase_lock);
 387         set_tb(timebase >> 32, timebase & 0xffffffff);
 388         timebase = 0;
 389         spin_unlock(&timebase_lock);
 390 }
 391
 392 static void __devinit smp_pSeries_kick_cpu(int nr)
 393 {
 394         BUG_ON(nr < 0 || nr >= NR_CPUS);
 395
 396         if (!smp_startup_cpu(nr))
 397                 return;
 398
 399         /*
 400          * The processor is currently spinning, waiting for the
 401          * cpu_start field to become non-zero After we set cpu_start,
 402          * the processor will continue on to secondary_start
 403          */
 404         paca[nr].cpu_start = 1;
 405 }
 406
 407 static int smp_pSeries_cpu_bootable(unsigned int nr)
 408 {
 409         /* Special case - we inhibit secondary thread startup
 410          * during boot if the user requests it.  Odd-numbered
 411          * cpus are assumed to be secondary threads.
 412          */
 413         if (system_state < SYSTEM_RUNNING &&
 414             cpu_has_feature(CPU_FTR_SMT) &&
 415             !smt_enabled_at_boot && nr % 2 != 0)
 416                 return 0;
 417
 418         return 1;
 419 }
 420 #ifdef CONFIG_MPIC
 421 static struct smp_ops_t pSeries_mpic_smp_ops = {
 422         .message_pass   = smp_mpic_message_pass,
 423         .probe          = smp_mpic_probe,
 424         .kick_cpu       = smp_pSeries_kick_cpu,
 425         .setup_cpu      = smp_mpic_setup_cpu,
 426 };
 427 #endif
 428 #ifdef CONFIG_XICS
 429 static struct smp_ops_t pSeries_xics_smp_ops = {
 430         .message_pass   = smp_xics_message_pass,
 431         .probe          = smp_xics_probe,
 432         .kick_cpu       = smp_pSeries_kick_cpu,
 433         .setup_cpu      = smp_xics_setup_cpu,
 434         .cpu_bootable   = smp_pSeries_cpu_bootable,
 435 };
 436 #endif
 437 #ifdef CONFIG_BPA_IIC
 438 static struct smp_ops_t bpa_iic_smp_ops = {
 439         .message_pass   = smp_iic_message_pass,
 440         .probe          = smp_iic_probe,
 441         .kick_cpu       = smp_pSeries_kick_cpu,
 442         .setup_cpu      = smp_iic_setup_cpu,
 443         .cpu_bootable   = smp_pSeries_cpu_bootable,
 444 };
 445 #endif
 446
 447 /* This is called very early */
 448 void __init smp_init_pSeries(void)
 449 {
 450         int i;
 451
 452         DBG(" -> smp_init_pSeries()\n");
 453
 454         switch (ppc64_interrupt_controller) {
 455 #ifdef CONFIG_MPIC
 456         case IC_OPEN_PIC:
 457                 smp_ops = &pSeries_mpic_smp_ops;
 458                 break;
 459 #endif
 460 #ifdef CONFIG_XICS
 461         case IC_PPC_XIC:
 462                 smp_ops = &pSeries_xics_smp_ops;
 463                 break;
 464 #endif
 465 #ifdef CONFIG_BPA_IIC
 466         case IC_BPA_IIC:
 467                 smp_ops = &bpa_iic_smp_ops;
 468                 break;
 469 #endif
 470         default:
 471                 panic("Invalid interrupt controller");
 472         }
 473
 474 #ifdef CONFIG_HOTPLUG_CPU
 475         smp_ops->cpu_disable = pSeries_cpu_disable;
 476         smp_ops->cpu_die = pSeries_cpu_die;
 477
 478         /* Processors can be added/removed only on LPAR */
 479         if (systemcfg->platform == PLATFORM_PSERIES_LPAR)
 480                 pSeries_reconfig_notifier_register(&pSeries_smp_nb);
 481 #endif
 482
 483         /* Mark threads which are still spinning in hold loops. */
 484         if (cpu_has_feature(CPU_FTR_SMT)) {
 485                 for_each_present_cpu(i) {
 486                         if (i % 2 == 0)
 487                                 /*
 488                                  * Even-numbered logical cpus correspond to
 489                                  * primary threads.
 490                                  */
 491                                 cpu_set(i, of_spin_map);
 492                 }
 493         } else {
 494                 of_spin_map = cpu_present_map;
 495         }
 496
 497         cpu_clear(boot_cpuid, of_spin_map);
 498
 499         /* Non-lpar has additional take/give timebase */
 500         if (rtas_token("freeze-time-base") != RTAS_UNKNOWN_SERVICE) {
 501                 smp_ops->give_timebase = pSeries_give_timebase;
 502                 smp_ops->take_timebase = pSeries_take_timebase;
 503         }
 504
 505         DBG(" <- smp_init_pSeries()\n");
 506 }
 507