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