platforms/pseries/hotplug-cpu.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * pseries CPU Hotplug infrastructure.
   4  *
   5  * Split out from arch/powerpc/platforms/pseries/setup.c
   6  *  arch/powerpc/kernel/rtas.c, and arch/powerpc/platforms/pseries/smp.c
   7  *
   8  * Peter Bergner, IBM   March 2001.
   9  * Copyright (C) 2001 IBM.
  10  * Dave Engebretsen, Peter Bergner, and
  11  * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com
  12  * Plus various changes from other IBM teams...
  13  *
  14  * Copyright (C) 2006 Michael Ellerman, IBM Corporation
  15  */
  16
  17 #define pr_fmt(fmt)     "pseries-hotplug-cpu: " fmt
  18
  19 #include <linux/kernel.h>
  20 #include <linux/interrupt.h>
  21 #include <linux/delay.h>
  22 #include <linux/sched.h>        /* for idle_task_exit */
  23 #include <linux/sched/hotplug.h>
  24 #include <linux/cpu.h>
  25 #include <linux/of.h>
  26 #include <linux/slab.h>
  27 #include <asm/prom.h>
  28 #include <asm/rtas.h>
  29 #include <asm/firmware.h>
  30 #include <asm/machdep.h>
  31 #include <asm/vdso_datapage.h>
  32 #include <asm/xics.h>
  33 #include <asm/xive.h>
  34 #include <asm/plpar_wrappers.h>
  35 #include <asm/topology.h>
  36
  37 #include "pseries.h"
  38
  39 /* This version can't take the spinlock, because it never returns */
  40 static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE;
  41
  42 /*
  43  * Record the CPU ids used on each nodes.
  44  * Protected by cpu_add_remove_lock.
  45  */
  46 static cpumask_var_t node_recorded_ids_map[MAX_NUMNODES];
  47
  48 static void rtas_stop_self(void)
  49 {
  50         static struct rtas_args args;
  51
  52         local_irq_disable();
  53
  54         BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE);
  55
  56         rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL);
  57
  58         panic("Alas, I survived.\n");
  59 }
  60
  61 static void pseries_cpu_offline_self(void)
  62 {
  63         unsigned int hwcpu = hard_smp_processor_id();
  64
  65         local_irq_disable();
  66         idle_task_exit();
  67         if (xive_enabled())
  68                 xive_teardown_cpu();
  69         else
  70                 xics_teardown_cpu();
  71
  72         unregister_slb_shadow(hwcpu);
  73         unregister_vpa(hwcpu);
  74         rtas_stop_self();
  75
  76         /* Should never get here... */
  77         BUG();
  78         for(;;);
  79 }
  80
  81 static int pseries_cpu_disable(void)
  82 {
  83         int cpu = smp_processor_id();
  84
  85         set_cpu_online(cpu, false);
  86         vdso_data->processorCount--;
  87
  88         /*fix boot_cpuid here*/
  89         if (cpu == boot_cpuid)
  90                 boot_cpuid = cpumask_any(cpu_online_mask);
  91
  92         /* FIXME: abstract this to not be platform specific later on */
  93         if (xive_enabled())
  94                 xive_smp_disable_cpu();
  95         else
  96                 xics_migrate_irqs_away();
  97
  98         cleanup_cpu_mmu_context();
  99
 100         return 0;
 101 }
 102
 103 /*
 104  * pseries_cpu_die: Wait for the cpu to die.
 105  * @cpu: logical processor id of the CPU whose death we're awaiting.
 106  *
 107  * This function is called from the context of the thread which is performing
 108  * the cpu-offline. Here we wait for long enough to allow the cpu in question
 109  * to self-destroy so that the cpu-offline thread can send the CPU_DEAD
 110  * notifications.
 111  *
 112  * OTOH, pseries_cpu_offline_self() is called by the @cpu when it wants to
 113  * self-destruct.
 114  */
 115 static void pseries_cpu_die(unsigned int cpu)
 116 {
 117         int cpu_status = 1;
 118         unsigned int pcpu = get_hard_smp_processor_id(cpu);
 119         unsigned long timeout = jiffies + msecs_to_jiffies(120000);
 120
 121         while (true) {
 122                 cpu_status = smp_query_cpu_stopped(pcpu);
 123                 if (cpu_status == QCSS_STOPPED ||
 124                     cpu_status == QCSS_HARDWARE_ERROR)
 125                         break;
 126
 127                 if (time_after(jiffies, timeout)) {
 128                         pr_warn("CPU %i (hwid %i) didn't die after 120 seconds\n",
 129                                 cpu, pcpu);
 130                         timeout = jiffies + msecs_to_jiffies(120000);
 131                 }
 132
 133                 cond_resched();
 134         }
 135
 136         if (cpu_status == QCSS_HARDWARE_ERROR) {
 137                 pr_warn("CPU %i (hwid %i) reported error while dying\n",
 138                         cpu, pcpu);
 139         }
 140
 141         paca_ptrs[cpu]->cpu_start = 0;
 142 }
 143
 144 /**
 145  * find_cpu_id_range - found a linear ranger of @nthreads free CPU ids.
 146  * @nthreads : the number of threads (cpu ids)
 147  * @assigned_node : the node it belongs to or NUMA_NO_NODE if free ids from any
 148  *                  node can be peek.
 149  * @cpu_mask: the returned CPU mask.
 150  *
 151  * Returns 0 on success.
 152  */
 153 static int find_cpu_id_range(unsigned int nthreads, int assigned_node,
 154                              cpumask_var_t *cpu_mask)
 155 {
 156         cpumask_var_t candidate_mask;
 157         unsigned int cpu, node;
 158         int rc = -ENOSPC;
 159
 160         if (!zalloc_cpumask_var(&candidate_mask, GFP_KERNEL))
 161                 return -ENOMEM;
 162
 163         cpumask_clear(*cpu_mask);
 164         for (cpu = 0; cpu < nthreads; cpu++)
 165                 cpumask_set_cpu(cpu, *cpu_mask);
 166
 167         BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask));
 168
 169         /* Get a bitmap of unoccupied slots. */
 170         cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask);
 171
 172         if (assigned_node != NUMA_NO_NODE) {
 173                 /*
 174                  * Remove free ids previously assigned on the other nodes. We
 175                  * can walk only online nodes because once a node became online
 176                  * it is not turned offlined back.
 177                  */
 178                 for_each_online_node(node) {
 179                         if (node == assigned_node)
 180                                 continue;
 181                         cpumask_andnot(candidate_mask, candidate_mask,
 182                                        node_recorded_ids_map[node]);
 183                 }
 184         }
 185
 186         if (cpumask_empty(candidate_mask))
 187                 goto out;
 188
 189         while (!cpumask_empty(*cpu_mask)) {
 190                 if (cpumask_subset(*cpu_mask, candidate_mask))
 191                         /* Found a range where we can insert the new cpu(s) */
 192                         break;
 193                 cpumask_shift_left(*cpu_mask, *cpu_mask, nthreads);
 194         }
 195
 196         if (!cpumask_empty(*cpu_mask))
 197                 rc = 0;
 198
 199 out:
 200         free_cpumask_var(candidate_mask);
 201         return rc;
 202 }
 203
 204 /*
 205  * Update cpu_present_mask and paca(s) for a new cpu node.  The wrinkle
 206  * here is that a cpu device node may represent multiple logical cpus
 207  * in the SMT case.  We must honor the assumption in other code that
 208  * the logical ids for sibling SMT threads x and y are adjacent, such
 209  * that x^1 == y and y^1 == x.
 210  */
 211 static int pseries_add_processor(struct device_node *np)
 212 {
 213         int len, nthreads, node, cpu, assigned_node;
 214         int rc = 0;
 215         cpumask_var_t cpu_mask;
 216         const __be32 *intserv;
 217
 218         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
 219         if (!intserv)
 220                 return 0;
 221
 222         nthreads = len / sizeof(u32);
 223
 224         if (!alloc_cpumask_var(&cpu_mask, GFP_KERNEL))
 225                 return -ENOMEM;
 226
 227         /*
 228          * Fetch from the DT nodes read by dlpar_configure_connector() the NUMA
 229          * node id the added CPU belongs to.
 230          */
 231         node = of_node_to_nid(np);
 232         if (node < 0 || !node_possible(node))
 233                 node = first_online_node;
 234
 235         BUG_ON(node == NUMA_NO_NODE);
 236         assigned_node = node;
 237
 238         cpu_maps_update_begin();
 239
 240         rc = find_cpu_id_range(nthreads, node, &cpu_mask);
 241         if (rc && nr_node_ids > 1) {
 242                 /*
 243                  * Try again, considering the free CPU ids from the other node.
 244                  */
 245                 node = NUMA_NO_NODE;
 246                 rc = find_cpu_id_range(nthreads, NUMA_NO_NODE, &cpu_mask);
 247         }
 248
 249         if (rc) {
 250                 pr_err("Cannot add cpu %pOF; this system configuration"
 251                        " supports %d logical cpus.\n", np, num_possible_cpus());
 252                 goto out;
 253         }
 254
 255         for_each_cpu(cpu, cpu_mask) {
 256                 BUG_ON(cpu_present(cpu));
 257                 set_cpu_present(cpu, true);
 258                 set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++));
 259         }
 260
 261         /* Record the newly used CPU ids for the associate node. */
 262         cpumask_or(node_recorded_ids_map[assigned_node],
 263                    node_recorded_ids_map[assigned_node], cpu_mask);
 264
 265         /*
 266          * If node is set to NUMA_NO_NODE, CPU ids have be reused from
 267          * another node, remove them from its mask.
 268          */
 269         if (node == NUMA_NO_NODE) {
 270                 cpu = cpumask_first(cpu_mask);
 271                 pr_warn("Reusing free CPU ids %d-%d from another node\n",
 272                         cpu, cpu + nthreads - 1);
 273                 for_each_online_node(node) {
 274                         if (node == assigned_node)
 275                                 continue;
 276                         cpumask_andnot(node_recorded_ids_map[node],
 277                                        node_recorded_ids_map[node],
 278                                        cpu_mask);
 279                 }
 280         }
 281
 282 out:
 283         cpu_maps_update_done();
 284         free_cpumask_var(cpu_mask);
 285         return rc;
 286 }
 287
 288 /*
 289  * Update the present map for a cpu node which is going away, and set
 290  * the hard id in the paca(s) to -1 to be consistent with boot time
 291  * convention for non-present cpus.
 292  */
 293 static void pseries_remove_processor(struct device_node *np)
 294 {
 295         unsigned int cpu;
 296         int len, nthreads, i;
 297         const __be32 *intserv;
 298         u32 thread;
 299
 300         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
 301         if (!intserv)
 302                 return;
 303
 304         nthreads = len / sizeof(u32);
 305
 306         cpu_maps_update_begin();
 307         for (i = 0; i < nthreads; i++) {
 308                 thread = be32_to_cpu(intserv[i]);
 309                 for_each_present_cpu(cpu) {
 310                         if (get_hard_smp_processor_id(cpu) != thread)
 311                                 continue;
 312                         BUG_ON(cpu_online(cpu));
 313                         set_cpu_present(cpu, false);
 314                         set_hard_smp_processor_id(cpu, -1);
 315                         update_numa_cpu_lookup_table(cpu, -1);
 316                         break;
 317                 }
 318                 if (cpu >= nr_cpu_ids)
 319                         printk(KERN_WARNING "Could not find cpu to remove "
 320                                "with physical id 0x%x\n", thread);
 321         }
 322         cpu_maps_update_done();
 323 }
 324
 325 static int dlpar_offline_cpu(struct device_node *dn)
 326 {
 327         int rc = 0;
 328         unsigned int cpu;
 329         int len, nthreads, i;
 330         const __be32 *intserv;
 331         u32 thread;
 332
 333         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 334         if (!intserv)
 335                 return -EINVAL;
 336
 337         nthreads = len / sizeof(u32);
 338
 339         cpu_maps_update_begin();
 340         for (i = 0; i < nthreads; i++) {
 341                 thread = be32_to_cpu(intserv[i]);
 342                 for_each_present_cpu(cpu) {
 343                         if (get_hard_smp_processor_id(cpu) != thread)
 344                                 continue;
 345
 346                         if (!cpu_online(cpu))
 347                                 break;
 348
 349                         /*
 350                          * device_offline() will return -EBUSY (via cpu_down()) if there
 351                          * is only one CPU left. Check it here to fail earlier and with a
 352                          * more informative error message, while also retaining the
 353                          * cpu_add_remove_lock to be sure that no CPUs are being
 354                          * online/offlined during this check.
 355                          */
 356                         if (num_online_cpus() == 1) {
 357                                 pr_warn("Unable to remove last online CPU %pOFn\n", dn);
 358                                 rc = -EBUSY;
 359                                 goto out_unlock;
 360                         }
 361
 362                         cpu_maps_update_done();
 363                         rc = device_offline(get_cpu_device(cpu));
 364                         if (rc)
 365                                 goto out;
 366                         cpu_maps_update_begin();
 367                         break;
 368                 }
 369                 if (cpu == num_possible_cpus()) {
 370                         pr_warn("Could not find cpu to offline with physical id 0x%x\n",
 371                                 thread);
 372                 }
 373         }
 374 out_unlock:
 375         cpu_maps_update_done();
 376
 377 out:
 378         return rc;
 379 }
 380
 381 static int dlpar_online_cpu(struct device_node *dn)
 382 {
 383         int rc = 0;
 384         unsigned int cpu;
 385         int len, nthreads, i;
 386         const __be32 *intserv;
 387         u32 thread;
 388
 389         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 390         if (!intserv)
 391                 return -EINVAL;
 392
 393         nthreads = len / sizeof(u32);
 394
 395         cpu_maps_update_begin();
 396         for (i = 0; i < nthreads; i++) {
 397                 thread = be32_to_cpu(intserv[i]);
 398                 for_each_present_cpu(cpu) {
 399                         if (get_hard_smp_processor_id(cpu) != thread)
 400                                 continue;
 401                         cpu_maps_update_done();
 402                         find_and_update_cpu_nid(cpu);
 403                         rc = device_online(get_cpu_device(cpu));
 404                         if (rc) {
 405                                 dlpar_offline_cpu(dn);
 406                                 goto out;
 407                         }
 408                         cpu_maps_update_begin();
 409
 410                         break;
 411                 }
 412                 if (cpu == num_possible_cpus())
 413                         printk(KERN_WARNING "Could not find cpu to online "
 414                                "with physical id 0x%x\n", thread);
 415         }
 416         cpu_maps_update_done();
 417
 418 out:
 419         return rc;
 420
 421 }
 422
 423 static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index)
 424 {
 425         struct device_node *child = NULL;
 426         u32 my_drc_index;
 427         bool found;
 428         int rc;
 429
 430         /* Assume cpu doesn't exist */
 431         found = false;
 432
 433         for_each_child_of_node(parent, child) {
 434                 rc = of_property_read_u32(child, "ibm,my-drc-index",
 435                                           &my_drc_index);
 436                 if (rc)
 437                         continue;
 438
 439                 if (my_drc_index == drc_index) {
 440                         of_node_put(child);
 441                         found = true;
 442                         break;
 443                 }
 444         }
 445
 446         return found;
 447 }
 448
 449 static bool drc_info_valid_index(struct device_node *parent, u32 drc_index)
 450 {
 451         struct property *info;
 452         struct of_drc_info drc;
 453         const __be32 *value;
 454         u32 index;
 455         int count, i, j;
 456
 457         info = of_find_property(parent, "ibm,drc-info", NULL);
 458         if (!info)
 459                 return false;
 460
 461         value = of_prop_next_u32(info, NULL, &count);
 462
 463         /* First value of ibm,drc-info is number of drc-info records */
 464         if (value)
 465                 value++;
 466         else
 467                 return false;
 468
 469         for (i = 0; i < count; i++) {
 470                 if (of_read_drc_info_cell(&info, &value, &drc))
 471                         return false;
 472
 473                 if (strncmp(drc.drc_type, "CPU", 3))
 474                         break;
 475
 476                 if (drc_index > drc.last_drc_index)
 477                         continue;
 478
 479                 index = drc.drc_index_start;
 480                 for (j = 0; j < drc.num_sequential_elems; j++) {
 481                         if (drc_index == index)
 482                                 return true;
 483
 484                         index += drc.sequential_inc;
 485                 }
 486         }
 487
 488         return false;
 489 }
 490
 491 static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index)
 492 {
 493         bool found = false;
 494         int rc, index;
 495
 496         if (of_property_present(parent, "ibm,drc-info"))
 497                 return drc_info_valid_index(parent, drc_index);
 498
 499         /* Note that the format of the ibm,drc-indexes array is
 500          * the number of entries in the array followed by the array
 501          * of drc values so we start looking at index = 1.
 502          */
 503         index = 1;
 504         while (!found) {
 505                 u32 drc;
 506
 507                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
 508                                                 index++, &drc);
 509
 510                 if (rc)
 511                         break;
 512
 513                 if (drc == drc_index)
 514                         found = true;
 515         }
 516
 517         return found;
 518 }
 519
 520 static int pseries_cpuhp_attach_nodes(struct device_node *dn)
 521 {
 522         struct of_changeset cs;
 523         int ret;
 524
 525         /*
 526          * This device node is unattached but may have siblings; open-code the
 527          * traversal.
 528          */
 529         for (of_changeset_init(&cs); dn != NULL; dn = dn->sibling) {
 530                 ret = of_changeset_attach_node(&cs, dn);
 531                 if (ret)
 532                         goto out;
 533         }
 534
 535         ret = of_changeset_apply(&cs);
 536 out:
 537         of_changeset_destroy(&cs);
 538         return ret;
 539 }
 540
 541 static ssize_t dlpar_cpu_add(u32 drc_index)
 542 {
 543         struct device_node *dn, *parent;
 544         int rc, saved_rc;
 545
 546         pr_debug("Attempting to add CPU, drc index: %x\n", drc_index);
 547
 548         parent = of_find_node_by_path("/cpus");
 549         if (!parent) {
 550                 pr_warn("Failed to find CPU root node \"/cpus\"\n");
 551                 return -ENODEV;
 552         }
 553
 554         if (dlpar_cpu_exists(parent, drc_index)) {
 555                 of_node_put(parent);
 556                 pr_warn("CPU with drc index %x already exists\n", drc_index);
 557                 return -EINVAL;
 558         }
 559
 560         if (!valid_cpu_drc_index(parent, drc_index)) {
 561                 of_node_put(parent);
 562                 pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index);
 563                 return -EINVAL;
 564         }
 565
 566         rc = dlpar_acquire_drc(drc_index);
 567         if (rc) {
 568                 pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n",
 569                         rc, drc_index);
 570                 of_node_put(parent);
 571                 return -EINVAL;
 572         }
 573
 574         dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent);
 575         if (!dn) {
 576                 pr_warn("Failed call to configure-connector, drc index: %x\n",
 577                         drc_index);
 578                 dlpar_release_drc(drc_index);
 579                 of_node_put(parent);
 580                 return -EINVAL;
 581         }
 582
 583         rc = pseries_cpuhp_attach_nodes(dn);
 584
 585         /* Regardless we are done with parent now */
 586         of_node_put(parent);
 587
 588         if (rc) {
 589                 saved_rc = rc;
 590                 pr_warn("Failed to attach node %pOFn, rc: %d, drc index: %x\n",
 591                         dn, rc, drc_index);
 592
 593                 rc = dlpar_release_drc(drc_index);
 594                 if (!rc)
 595                         dlpar_free_cc_nodes(dn);
 596
 597                 return saved_rc;
 598         }
 599
 600         update_numa_distance(dn);
 601
 602         rc = dlpar_online_cpu(dn);
 603         if (rc) {
 604                 saved_rc = rc;
 605                 pr_warn("Failed to online cpu %pOFn, rc: %d, drc index: %x\n",
 606                         dn, rc, drc_index);
 607
 608                 rc = dlpar_detach_node(dn);
 609                 if (!rc)
 610                         dlpar_release_drc(drc_index);
 611
 612                 return saved_rc;
 613         }
 614
 615         pr_debug("Successfully added CPU %pOFn, drc index: %x\n", dn,
 616                  drc_index);
 617         return rc;
 618 }
 619
 620 static unsigned int pseries_cpuhp_cache_use_count(const struct device_node *cachedn)
 621 {
 622         unsigned int use_count = 0;
 623         struct device_node *dn, *tn;
 624
 625         WARN_ON(!of_node_is_type(cachedn, "cache"));
 626
 627         for_each_of_cpu_node(dn) {
 628                 tn = of_find_next_cache_node(dn);
 629                 of_node_put(tn);
 630                 if (tn == cachedn)
 631                         use_count++;
 632         }
 633
 634         for_each_node_by_type(dn, "cache") {
 635                 tn = of_find_next_cache_node(dn);
 636                 of_node_put(tn);
 637                 if (tn == cachedn)
 638                         use_count++;
 639         }
 640
 641         return use_count;
 642 }
 643
 644 static int pseries_cpuhp_detach_nodes(struct device_node *cpudn)
 645 {
 646         struct device_node *dn;
 647         struct of_changeset cs;
 648         int ret = 0;
 649
 650         of_changeset_init(&cs);
 651         ret = of_changeset_detach_node(&cs, cpudn);
 652         if (ret)
 653                 goto out;
 654
 655         dn = cpudn;
 656         while ((dn = of_find_next_cache_node(dn))) {
 657                 if (pseries_cpuhp_cache_use_count(dn) > 1) {
 658                         of_node_put(dn);
 659                         break;
 660                 }
 661
 662                 ret = of_changeset_detach_node(&cs, dn);
 663                 of_node_put(dn);
 664                 if (ret)
 665                         goto out;
 666         }
 667
 668         ret = of_changeset_apply(&cs);
 669 out:
 670         of_changeset_destroy(&cs);
 671         return ret;
 672 }
 673
 674 static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index)
 675 {
 676         int rc;
 677
 678         pr_debug("Attempting to remove CPU %pOFn, drc index: %x\n",
 679                  dn, drc_index);
 680
 681         rc = dlpar_offline_cpu(dn);
 682         if (rc) {
 683                 pr_warn("Failed to offline CPU %pOFn, rc: %d\n", dn, rc);
 684                 return -EINVAL;
 685         }
 686
 687         rc = dlpar_release_drc(drc_index);
 688         if (rc) {
 689                 pr_warn("Failed to release drc (%x) for CPU %pOFn, rc: %d\n",
 690                         drc_index, dn, rc);
 691                 dlpar_online_cpu(dn);
 692                 return rc;
 693         }
 694
 695         rc = pseries_cpuhp_detach_nodes(dn);
 696         if (rc) {
 697                 int saved_rc = rc;
 698
 699                 pr_warn("Failed to detach CPU %pOFn, rc: %d", dn, rc);
 700
 701                 rc = dlpar_acquire_drc(drc_index);
 702                 if (!rc)
 703                         dlpar_online_cpu(dn);
 704
 705                 return saved_rc;
 706         }
 707
 708         pr_debug("Successfully removed CPU, drc index: %x\n", drc_index);
 709         return 0;
 710 }
 711
 712 static struct device_node *cpu_drc_index_to_dn(u32 drc_index)
 713 {
 714         struct device_node *dn;
 715         u32 my_index;
 716         int rc;
 717
 718         for_each_node_by_type(dn, "cpu") {
 719                 rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index);
 720                 if (rc)
 721                         continue;
 722
 723                 if (my_index == drc_index)
 724                         break;
 725         }
 726
 727         return dn;
 728 }
 729
 730 static int dlpar_cpu_remove_by_index(u32 drc_index)
 731 {
 732         struct device_node *dn;
 733         int rc;
 734
 735         dn = cpu_drc_index_to_dn(drc_index);
 736         if (!dn) {
 737                 pr_warn("Cannot find CPU (drc index %x) to remove\n",
 738                         drc_index);
 739                 return -ENODEV;
 740         }
 741
 742         rc = dlpar_cpu_remove(dn, drc_index);
 743         of_node_put(dn);
 744         return rc;
 745 }
 746
 747 int dlpar_cpu(struct pseries_hp_errorlog *hp_elog)
 748 {
 749         u32 drc_index;
 750         int rc;
 751
 752         drc_index = hp_elog->_drc_u.drc_index;
 753
 754         lock_device_hotplug();
 755
 756         switch (hp_elog->action) {
 757         case PSERIES_HP_ELOG_ACTION_REMOVE:
 758                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX) {
 759                         rc = dlpar_cpu_remove_by_index(drc_index);
 760                         /*
 761                          * Setting the isolation state of an UNISOLATED/CONFIGURED
 762                          * device to UNISOLATE is a no-op, but the hypervisor can
 763                          * use it as a hint that the CPU removal failed.
 764                          */
 765                         if (rc)
 766                                 dlpar_unisolate_drc(drc_index);
 767                 }
 768                 else
 769                         rc = -EINVAL;
 770                 break;
 771         case PSERIES_HP_ELOG_ACTION_ADD:
 772                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
 773                         rc = dlpar_cpu_add(drc_index);
 774                 else
 775                         rc = -EINVAL;
 776                 break;
 777         default:
 778                 pr_err("Invalid action (%d) specified\n", hp_elog->action);
 779                 rc = -EINVAL;
 780                 break;
 781         }
 782
 783         unlock_device_hotplug();
 784         return rc;
 785 }
 786
 787 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 788
 789 static ssize_t dlpar_cpu_probe(const char *buf, size_t count)
 790 {
 791         u32 drc_index;
 792         int rc;
 793
 794         rc = kstrtou32(buf, 0, &drc_index);
 795         if (rc)
 796                 return -EINVAL;
 797
 798         rc = dlpar_cpu_add(drc_index);
 799
 800         return rc ? rc : count;
 801 }
 802
 803 static ssize_t dlpar_cpu_release(const char *buf, size_t count)
 804 {
 805         struct device_node *dn;
 806         u32 drc_index;
 807         int rc;
 808
 809         dn = of_find_node_by_path(buf);
 810         if (!dn)
 811                 return -EINVAL;
 812
 813         rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index);
 814         if (rc) {
 815                 of_node_put(dn);
 816                 return -EINVAL;
 817         }
 818
 819         rc = dlpar_cpu_remove(dn, drc_index);
 820         of_node_put(dn);
 821
 822         return rc ? rc : count;
 823 }
 824
 825 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 826
 827 static int pseries_smp_notifier(struct notifier_block *nb,
 828                                 unsigned long action, void *data)
 829 {
 830         struct of_reconfig_data *rd = data;
 831         int err = 0;
 832
 833         switch (action) {
 834         case OF_RECONFIG_ATTACH_NODE:
 835                 err = pseries_add_processor(rd->dn);
 836                 break;
 837         case OF_RECONFIG_DETACH_NODE:
 838                 pseries_remove_processor(rd->dn);
 839                 break;
 840         }
 841         return notifier_from_errno(err);
 842 }
 843
 844 static struct notifier_block pseries_smp_nb = {
 845         .notifier_call = pseries_smp_notifier,
 846 };
 847
 848 static int __init pseries_cpu_hotplug_init(void)
 849 {
 850         int qcss_tok;
 851         unsigned int node;
 852
 853 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 854         ppc_md.cpu_probe = dlpar_cpu_probe;
 855         ppc_md.cpu_release = dlpar_cpu_release;
 856 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 857
 858         rtas_stop_self_token = rtas_function_token(RTAS_FN_STOP_SELF);
 859         qcss_tok = rtas_function_token(RTAS_FN_QUERY_CPU_STOPPED_STATE);
 860
 861         if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE ||
 862                         qcss_tok == RTAS_UNKNOWN_SERVICE) {
 863                 printk(KERN_INFO "CPU Hotplug not supported by firmware "
 864                                 "- disabling.\n");
 865                 return 0;
 866         }
 867
 868         smp_ops->cpu_offline_self = pseries_cpu_offline_self;
 869         smp_ops->cpu_disable = pseries_cpu_disable;
 870         smp_ops->cpu_die = pseries_cpu_die;
 871
 872         /* Processors can be added/removed only on LPAR */
 873         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 874                 for_each_node(node) {
 875                         if (!alloc_cpumask_var_node(&node_recorded_ids_map[node],
 876                                                     GFP_KERNEL, node))
 877                                 return -ENOMEM;
 878
 879                         /* Record ids of CPU added at boot time */
 880                         cpumask_copy(node_recorded_ids_map[node],
 881                                      cpumask_of_node(node));
 882                 }
 883
 884                 of_reconfig_notifier_register(&pseries_smp_nb);
 885         }
 886
 887         return 0;
 888 }
 889 machine_arch_initcall(pseries, pseries_cpu_hotplug_init);