arch/powerpc/platforms/pseries/hotplug-cpu.c

   1 /*
   2  * pseries CPU Hotplug infrastructure.
   3  *
   4  * Split out from arch/powerpc/platforms/pseries/setup.c
   5  *  arch/powerpc/kernel/rtas.c, and arch/powerpc/platforms/pseries/smp.c
   6  *
   7  * Peter Bergner, IBM   March 2001.
   8  * Copyright (C) 2001 IBM.
   9  * Dave Engebretsen, Peter Bergner, and
  10  * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com
  11  * Plus various changes from other IBM teams...
  12  *
  13  * Copyright (C) 2006 Michael Ellerman, IBM Corporation
  14  *
  15  *      This program is free software; you can redistribute it and/or
  16  *      modify it under the terms of the GNU General Public License
  17  *      as published by the Free Software Foundation; either version
  18  *      2 of the License, or (at your option) any later version.
  19  */
  20
  21 #define pr_fmt(fmt)     "pseries-hotplug-cpu: " fmt
  22
  23 #include <linux/kernel.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/delay.h>
  26 #include <linux/sched.h>        /* for idle_task_exit */
  27 #include <linux/sched/hotplug.h>
  28 #include <linux/cpu.h>
  29 #include <linux/of.h>
  30 #include <linux/slab.h>
  31 #include <asm/prom.h>
  32 #include <asm/rtas.h>
  33 #include <asm/firmware.h>
  34 #include <asm/machdep.h>
  35 #include <asm/vdso_datapage.h>
  36 #include <asm/xics.h>
  37 #include <asm/xive.h>
  38 #include <asm/plpar_wrappers.h>
  39
  40 #include "pseries.h"
  41 #include "offline_states.h"
  42
  43 /* This version can't take the spinlock, because it never returns */
  44 static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE;
  45
  46 static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) =
  47                                                         CPU_STATE_OFFLINE;
  48 static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE;
  49
  50 static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE;
  51
  52 static bool cede_offline_enabled __read_mostly = true;
  53
  54 /*
  55  * Enable/disable cede_offline when available.
  56  */
  57 static int __init setup_cede_offline(char *str)
  58 {
  59         return (kstrtobool(str, &cede_offline_enabled) == 0);
  60 }
  61
  62 __setup("cede_offline=", setup_cede_offline);
  63
  64 enum cpu_state_vals get_cpu_current_state(int cpu)
  65 {
  66         return per_cpu(current_state, cpu);
  67 }
  68
  69 void set_cpu_current_state(int cpu, enum cpu_state_vals state)
  70 {
  71         per_cpu(current_state, cpu) = state;
  72 }
  73
  74 enum cpu_state_vals get_preferred_offline_state(int cpu)
  75 {
  76         return per_cpu(preferred_offline_state, cpu);
  77 }
  78
  79 void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
  80 {
  81         per_cpu(preferred_offline_state, cpu) = state;
  82 }
  83
  84 void set_default_offline_state(int cpu)
  85 {
  86         per_cpu(preferred_offline_state, cpu) = default_offline_state;
  87 }
  88
  89 static void rtas_stop_self(void)
  90 {
  91         static struct rtas_args args;
  92
  93         local_irq_disable();
  94
  95         BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE);
  96
  97         printk("cpu %u (hwid %u) Ready to die...\n",
  98                smp_processor_id(), hard_smp_processor_id());
  99
 100         rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL);
 101
 102         panic("Alas, I survived.\n");
 103 }
 104
 105 static void pseries_mach_cpu_die(void)
 106 {
 107         unsigned int cpu = smp_processor_id();
 108         unsigned int hwcpu = hard_smp_processor_id();
 109         u8 cede_latency_hint = 0;
 110
 111         local_irq_disable();
 112         idle_task_exit();
 113         if (xive_enabled())
 114                 xive_teardown_cpu();
 115         else
 116                 xics_teardown_cpu();
 117
 118         if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
 119                 set_cpu_current_state(cpu, CPU_STATE_INACTIVE);
 120                 if (ppc_md.suspend_disable_cpu)
 121                         ppc_md.suspend_disable_cpu();
 122
 123                 cede_latency_hint = 2;
 124
 125                 get_lppaca()->idle = 1;
 126                 if (!lppaca_shared_proc(get_lppaca()))
 127                         get_lppaca()->donate_dedicated_cpu = 1;
 128
 129                 while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
 130                         while (!prep_irq_for_idle()) {
 131                                 local_irq_enable();
 132                                 local_irq_disable();
 133                         }
 134
 135                         extended_cede_processor(cede_latency_hint);
 136                 }
 137
 138                 local_irq_disable();
 139
 140                 if (!lppaca_shared_proc(get_lppaca()))
 141                         get_lppaca()->donate_dedicated_cpu = 0;
 142                 get_lppaca()->idle = 0;
 143
 144                 if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) {
 145                         unregister_slb_shadow(hwcpu);
 146
 147                         hard_irq_disable();
 148                         /*
 149                          * Call to start_secondary_resume() will not return.
 150                          * Kernel stack will be reset and start_secondary()
 151                          * will be called to continue the online operation.
 152                          */
 153                         start_secondary_resume();
 154                 }
 155         }
 156
 157         /* Requested state is CPU_STATE_OFFLINE at this point */
 158         WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE);
 159
 160         set_cpu_current_state(cpu, CPU_STATE_OFFLINE);
 161         unregister_slb_shadow(hwcpu);
 162         rtas_stop_self();
 163
 164         /* Should never get here... */
 165         BUG();
 166         for(;;);
 167 }
 168
 169 static int pseries_cpu_disable(void)
 170 {
 171         int cpu = smp_processor_id();
 172
 173         set_cpu_online(cpu, false);
 174         vdso_data->processorCount--;
 175
 176         /*fix boot_cpuid here*/
 177         if (cpu == boot_cpuid)
 178                 boot_cpuid = cpumask_any(cpu_online_mask);
 179
 180         /* FIXME: abstract this to not be platform specific later on */
 181         if (xive_enabled())
 182                 xive_smp_disable_cpu();
 183         else
 184                 xics_migrate_irqs_away();
 185         return 0;
 186 }
 187
 188 /*
 189  * pseries_cpu_die: Wait for the cpu to die.
 190  * @cpu: logical processor id of the CPU whose death we're awaiting.
 191  *
 192  * This function is called from the context of the thread which is performing
 193  * the cpu-offline. Here we wait for long enough to allow the cpu in question
 194  * to self-destroy so that the cpu-offline thread can send the CPU_DEAD
 195  * notifications.
 196  *
 197  * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to
 198  * self-destruct.
 199  */
 200 static void pseries_cpu_die(unsigned int cpu)
 201 {
 202         int tries;
 203         int cpu_status = 1;
 204         unsigned int pcpu = get_hard_smp_processor_id(cpu);
 205
 206         if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
 207                 cpu_status = 1;
 208                 for (tries = 0; tries < 5000; tries++) {
 209                         if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) {
 210                                 cpu_status = 0;
 211                                 break;
 212                         }
 213                         msleep(1);
 214                 }
 215         } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
 216
 217                 for (tries = 0; tries < 25; tries++) {
 218                         cpu_status = smp_query_cpu_stopped(pcpu);
 219                         if (cpu_status == QCSS_STOPPED ||
 220                             cpu_status == QCSS_HARDWARE_ERROR)
 221                                 break;
 222                         cpu_relax();
 223                 }
 224         }
 225
 226         if (cpu_status != 0) {
 227                 printk("Querying DEAD? cpu %i (%i) shows %i\n",
 228                        cpu, pcpu, cpu_status);
 229         }
 230
 231         /* Isolation and deallocation are definitely done by
 232          * drslot_chrp_cpu.  If they were not they would be
 233          * done here.  Change isolate state to Isolate and
 234          * change allocation-state to Unusable.
 235          */
 236         paca[cpu].cpu_start = 0;
 237 }
 238
 239 /*
 240  * Update cpu_present_mask and paca(s) for a new cpu node.  The wrinkle
 241  * here is that a cpu device node may represent up to two logical cpus
 242  * in the SMT case.  We must honor the assumption in other code that
 243  * the logical ids for sibling SMT threads x and y are adjacent, such
 244  * that x^1 == y and y^1 == x.
 245  */
 246 static int pseries_add_processor(struct device_node *np)
 247 {
 248         unsigned int cpu;
 249         cpumask_var_t candidate_mask, tmp;
 250         int err = -ENOSPC, len, nthreads, i;
 251         const __be32 *intserv;
 252
 253         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
 254         if (!intserv)
 255                 return 0;
 256
 257         zalloc_cpumask_var(&candidate_mask, GFP_KERNEL);
 258         zalloc_cpumask_var(&tmp, GFP_KERNEL);
 259
 260         nthreads = len / sizeof(u32);
 261         for (i = 0; i < nthreads; i++)
 262                 cpumask_set_cpu(i, tmp);
 263
 264         cpu_maps_update_begin();
 265
 266         BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask));
 267
 268         /* Get a bitmap of unoccupied slots. */
 269         cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask);
 270         if (cpumask_empty(candidate_mask)) {
 271                 /* If we get here, it most likely means that NR_CPUS is
 272                  * less than the partition's max processors setting.
 273                  */
 274                 printk(KERN_ERR "Cannot add cpu %pOF; this system configuration"
 275                        " supports %d logical cpus.\n", np,
 276                        num_possible_cpus());
 277                 goto out_unlock;
 278         }
 279
 280         while (!cpumask_empty(tmp))
 281                 if (cpumask_subset(tmp, candidate_mask))
 282                         /* Found a range where we can insert the new cpu(s) */
 283                         break;
 284                 else
 285                         cpumask_shift_left(tmp, tmp, nthreads);
 286
 287         if (cpumask_empty(tmp)) {
 288                 printk(KERN_ERR "Unable to find space in cpu_present_mask for"
 289                        " processor %s with %d thread(s)\n", np->name,
 290                        nthreads);
 291                 goto out_unlock;
 292         }
 293
 294         for_each_cpu(cpu, tmp) {
 295                 BUG_ON(cpu_present(cpu));
 296                 set_cpu_present(cpu, true);
 297                 set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++));
 298         }
 299         err = 0;
 300 out_unlock:
 301         cpu_maps_update_done();
 302         free_cpumask_var(candidate_mask);
 303         free_cpumask_var(tmp);
 304         return err;
 305 }
 306
 307 /*
 308  * Update the present map for a cpu node which is going away, and set
 309  * the hard id in the paca(s) to -1 to be consistent with boot time
 310  * convention for non-present cpus.
 311  */
 312 static void pseries_remove_processor(struct device_node *np)
 313 {
 314         unsigned int cpu;
 315         int len, nthreads, i;
 316         const __be32 *intserv;
 317         u32 thread;
 318
 319         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
 320         if (!intserv)
 321                 return;
 322
 323         nthreads = len / sizeof(u32);
 324
 325         cpu_maps_update_begin();
 326         for (i = 0; i < nthreads; i++) {
 327                 thread = be32_to_cpu(intserv[i]);
 328                 for_each_present_cpu(cpu) {
 329                         if (get_hard_smp_processor_id(cpu) != thread)
 330                                 continue;
 331                         BUG_ON(cpu_online(cpu));
 332                         set_cpu_present(cpu, false);
 333                         set_hard_smp_processor_id(cpu, -1);
 334                         break;
 335                 }
 336                 if (cpu >= nr_cpu_ids)
 337                         printk(KERN_WARNING "Could not find cpu to remove "
 338                                "with physical id 0x%x\n", thread);
 339         }
 340         cpu_maps_update_done();
 341 }
 342
 343 extern int find_and_online_cpu_nid(int cpu);
 344
 345 static int dlpar_online_cpu(struct device_node *dn)
 346 {
 347         int rc = 0;
 348         unsigned int cpu;
 349         int len, nthreads, i;
 350         const __be32 *intserv;
 351         u32 thread;
 352
 353         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 354         if (!intserv)
 355                 return -EINVAL;
 356
 357         nthreads = len / sizeof(u32);
 358
 359         cpu_maps_update_begin();
 360         for (i = 0; i < nthreads; i++) {
 361                 thread = be32_to_cpu(intserv[i]);
 362                 for_each_present_cpu(cpu) {
 363                         if (get_hard_smp_processor_id(cpu) != thread)
 364                                 continue;
 365                         BUG_ON(get_cpu_current_state(cpu)
 366                                         != CPU_STATE_OFFLINE);
 367                         cpu_maps_update_done();
 368                         timed_topology_update(1);
 369                         find_and_online_cpu_nid(cpu);
 370                         rc = device_online(get_cpu_device(cpu));
 371                         if (rc)
 372                                 goto out;
 373                         cpu_maps_update_begin();
 374
 375                         break;
 376                 }
 377                 if (cpu == num_possible_cpus())
 378                         printk(KERN_WARNING "Could not find cpu to online "
 379                                "with physical id 0x%x\n", thread);
 380         }
 381         cpu_maps_update_done();
 382
 383 out:
 384         return rc;
 385
 386 }
 387
 388 static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index)
 389 {
 390         struct device_node *child = NULL;
 391         u32 my_drc_index;
 392         bool found;
 393         int rc;
 394
 395         /* Assume cpu doesn't exist */
 396         found = false;
 397
 398         for_each_child_of_node(parent, child) {
 399                 rc = of_property_read_u32(child, "ibm,my-drc-index",
 400                                           &my_drc_index);
 401                 if (rc)
 402                         continue;
 403
 404                 if (my_drc_index == drc_index) {
 405                         of_node_put(child);
 406                         found = true;
 407                         break;
 408                 }
 409         }
 410
 411         return found;
 412 }
 413
 414 static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index)
 415 {
 416         bool found = false;
 417         int rc, index;
 418
 419         index = 0;
 420         while (!found) {
 421                 u32 drc;
 422
 423                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
 424                                                 index++, &drc);
 425                 if (rc)
 426                         break;
 427
 428                 if (drc == drc_index)
 429                         found = true;
 430         }
 431
 432         return found;
 433 }
 434
 435 static ssize_t dlpar_cpu_add(u32 drc_index)
 436 {
 437         struct device_node *dn, *parent;
 438         int rc, saved_rc;
 439
 440         pr_debug("Attempting to add CPU, drc index: %x\n", drc_index);
 441
 442         parent = of_find_node_by_path("/cpus");
 443         if (!parent) {
 444                 pr_warn("Failed to find CPU root node \"/cpus\"\n");
 445                 return -ENODEV;
 446         }
 447
 448         if (dlpar_cpu_exists(parent, drc_index)) {
 449                 of_node_put(parent);
 450                 pr_warn("CPU with drc index %x already exists\n", drc_index);
 451                 return -EINVAL;
 452         }
 453
 454         if (!valid_cpu_drc_index(parent, drc_index)) {
 455                 of_node_put(parent);
 456                 pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index);
 457                 return -EINVAL;
 458         }
 459
 460         rc = dlpar_acquire_drc(drc_index);
 461         if (rc) {
 462                 pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n",
 463                         rc, drc_index);
 464                 of_node_put(parent);
 465                 return -EINVAL;
 466         }
 467
 468         dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent);
 469         if (!dn) {
 470                 pr_warn("Failed call to configure-connector, drc index: %x\n",
 471                         drc_index);
 472                 dlpar_release_drc(drc_index);
 473                 of_node_put(parent);
 474                 return -EINVAL;
 475         }
 476
 477         rc = dlpar_attach_node(dn, parent);
 478
 479         /* Regardless we are done with parent now */
 480         of_node_put(parent);
 481
 482         if (rc) {
 483                 saved_rc = rc;
 484                 pr_warn("Failed to attach node %s, rc: %d, drc index: %x\n",
 485                         dn->name, rc, drc_index);
 486
 487                 rc = dlpar_release_drc(drc_index);
 488                 if (!rc)
 489                         dlpar_free_cc_nodes(dn);
 490
 491                 return saved_rc;
 492         }
 493
 494         rc = dlpar_online_cpu(dn);
 495         if (rc) {
 496                 saved_rc = rc;
 497                 pr_warn("Failed to online cpu %s, rc: %d, drc index: %x\n",
 498                         dn->name, rc, drc_index);
 499
 500                 rc = dlpar_detach_node(dn);
 501                 if (!rc)
 502                         dlpar_release_drc(drc_index);
 503
 504                 return saved_rc;
 505         }
 506
 507         pr_debug("Successfully added CPU %s, drc index: %x\n", dn->name,
 508                  drc_index);
 509         return rc;
 510 }
 511
 512 static int dlpar_offline_cpu(struct device_node *dn)
 513 {
 514         int rc = 0;
 515         unsigned int cpu;
 516         int len, nthreads, i;
 517         const __be32 *intserv;
 518         u32 thread;
 519
 520         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 521         if (!intserv)
 522                 return -EINVAL;
 523
 524         nthreads = len / sizeof(u32);
 525
 526         cpu_maps_update_begin();
 527         for (i = 0; i < nthreads; i++) {
 528                 thread = be32_to_cpu(intserv[i]);
 529                 for_each_present_cpu(cpu) {
 530                         if (get_hard_smp_processor_id(cpu) != thread)
 531                                 continue;
 532
 533                         if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE)
 534                                 break;
 535
 536                         if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
 537                                 set_preferred_offline_state(cpu,
 538                                                             CPU_STATE_OFFLINE);
 539                                 cpu_maps_update_done();
 540                                 timed_topology_update(1);
 541                                 rc = device_offline(get_cpu_device(cpu));
 542                                 if (rc)
 543                                         goto out;
 544                                 cpu_maps_update_begin();
 545                                 break;
 546
 547                         }
 548
 549                         /*
 550                          * The cpu is in CPU_STATE_INACTIVE.
 551                          * Upgrade it's state to CPU_STATE_OFFLINE.
 552                          */
 553                         set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
 554                         BUG_ON(plpar_hcall_norets(H_PROD, thread)
 555                                                                 != H_SUCCESS);
 556                         __cpu_die(cpu);
 557                         break;
 558                 }
 559                 if (cpu == num_possible_cpus())
 560                         printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread);
 561         }
 562         cpu_maps_update_done();
 563
 564 out:
 565         return rc;
 566
 567 }
 568
 569 static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index)
 570 {
 571         int rc;
 572
 573         pr_debug("Attempting to remove CPU %s, drc index: %x\n",
 574                  dn->name, drc_index);
 575
 576         rc = dlpar_offline_cpu(dn);
 577         if (rc) {
 578                 pr_warn("Failed to offline CPU %s, rc: %d\n", dn->name, rc);
 579                 return -EINVAL;
 580         }
 581
 582         rc = dlpar_release_drc(drc_index);
 583         if (rc) {
 584                 pr_warn("Failed to release drc (%x) for CPU %s, rc: %d\n",
 585                         drc_index, dn->name, rc);
 586                 dlpar_online_cpu(dn);
 587                 return rc;
 588         }
 589
 590         rc = dlpar_detach_node(dn);
 591         if (rc) {
 592                 int saved_rc = rc;
 593
 594                 pr_warn("Failed to detach CPU %s, rc: %d", dn->name, rc);
 595
 596                 rc = dlpar_acquire_drc(drc_index);
 597                 if (!rc)
 598                         dlpar_online_cpu(dn);
 599
 600                 return saved_rc;
 601         }
 602
 603         pr_debug("Successfully removed CPU, drc index: %x\n", drc_index);
 604         return 0;
 605 }
 606
 607 static struct device_node *cpu_drc_index_to_dn(u32 drc_index)
 608 {
 609         struct device_node *dn;
 610         u32 my_index;
 611         int rc;
 612
 613         for_each_node_by_type(dn, "cpu") {
 614                 rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index);
 615                 if (rc)
 616                         continue;
 617
 618                 if (my_index == drc_index)
 619                         break;
 620         }
 621
 622         return dn;
 623 }
 624
 625 static int dlpar_cpu_remove_by_index(u32 drc_index)
 626 {
 627         struct device_node *dn;
 628         int rc;
 629
 630         dn = cpu_drc_index_to_dn(drc_index);
 631         if (!dn) {
 632                 pr_warn("Cannot find CPU (drc index %x) to remove\n",
 633                         drc_index);
 634                 return -ENODEV;
 635         }
 636
 637         rc = dlpar_cpu_remove(dn, drc_index);
 638         of_node_put(dn);
 639         return rc;
 640 }
 641
 642 static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove)
 643 {
 644         struct device_node *dn;
 645         int cpus_found = 0;
 646         int rc;
 647
 648         /* We want to find cpus_to_remove + 1 CPUs to ensure we do not
 649          * remove the last CPU.
 650          */
 651         for_each_node_by_type(dn, "cpu") {
 652                 cpus_found++;
 653
 654                 if (cpus_found > cpus_to_remove) {
 655                         of_node_put(dn);
 656                         break;
 657                 }
 658
 659                 /* Note that cpus_found is always 1 ahead of the index
 660                  * into the cpu_drcs array, so we use cpus_found - 1
 661                  */
 662                 rc = of_property_read_u32(dn, "ibm,my-drc-index",
 663                                           &cpu_drcs[cpus_found - 1]);
 664                 if (rc) {
 665                         pr_warn("Error occurred getting drc-index for %s\n",
 666                                 dn->name);
 667                         of_node_put(dn);
 668                         return -1;
 669                 }
 670         }
 671
 672         if (cpus_found < cpus_to_remove) {
 673                 pr_warn("Failed to find enough CPUs (%d of %d) to remove\n",
 674                         cpus_found, cpus_to_remove);
 675         } else if (cpus_found == cpus_to_remove) {
 676                 pr_warn("Cannot remove all CPUs\n");
 677         }
 678
 679         return cpus_found;
 680 }
 681
 682 static int dlpar_cpu_remove_by_count(u32 cpus_to_remove)
 683 {
 684         u32 *cpu_drcs;
 685         int cpus_found;
 686         int cpus_removed = 0;
 687         int i, rc;
 688
 689         pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove);
 690
 691         cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL);
 692         if (!cpu_drcs)
 693                 return -EINVAL;
 694
 695         cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove);
 696         if (cpus_found <= cpus_to_remove) {
 697                 kfree(cpu_drcs);
 698                 return -EINVAL;
 699         }
 700
 701         for (i = 0; i < cpus_to_remove; i++) {
 702                 rc = dlpar_cpu_remove_by_index(cpu_drcs[i]);
 703                 if (rc)
 704                         break;
 705
 706                 cpus_removed++;
 707         }
 708
 709         if (cpus_removed != cpus_to_remove) {
 710                 pr_warn("CPU hot-remove failed, adding back removed CPUs\n");
 711
 712                 for (i = 0; i < cpus_removed; i++)
 713                         dlpar_cpu_add(cpu_drcs[i]);
 714
 715                 rc = -EINVAL;
 716         } else {
 717                 rc = 0;
 718         }
 719
 720         kfree(cpu_drcs);
 721         return rc;
 722 }
 723
 724 static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add)
 725 {
 726         struct device_node *parent;
 727         int cpus_found = 0;
 728         int index, rc;
 729
 730         parent = of_find_node_by_path("/cpus");
 731         if (!parent) {
 732                 pr_warn("Could not find CPU root node in device tree\n");
 733                 kfree(cpu_drcs);
 734                 return -1;
 735         }
 736
 737         /* Search the ibm,drc-indexes array for possible CPU drcs to
 738          * add. Note that the format of the ibm,drc-indexes array is
 739          * the number of entries in the array followed by the array
 740          * of drc values so we start looking at index = 1.
 741          */
 742         index = 1;
 743         while (cpus_found < cpus_to_add) {
 744                 u32 drc;
 745
 746                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
 747                                                 index++, &drc);
 748                 if (rc)
 749                         break;
 750
 751                 if (dlpar_cpu_exists(parent, drc))
 752                         continue;
 753
 754                 cpu_drcs[cpus_found++] = drc;
 755         }
 756
 757         of_node_put(parent);
 758         return cpus_found;
 759 }
 760
 761 static int dlpar_cpu_add_by_count(u32 cpus_to_add)
 762 {
 763         u32 *cpu_drcs;
 764         int cpus_added = 0;
 765         int cpus_found;
 766         int i, rc;
 767
 768         pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add);
 769
 770         cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL);
 771         if (!cpu_drcs)
 772                 return -EINVAL;
 773
 774         cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add);
 775         if (cpus_found < cpus_to_add) {
 776                 pr_warn("Failed to find enough CPUs (%d of %d) to add\n",
 777                         cpus_found, cpus_to_add);
 778                 kfree(cpu_drcs);
 779                 return -EINVAL;
 780         }
 781
 782         for (i = 0; i < cpus_to_add; i++) {
 783                 rc = dlpar_cpu_add(cpu_drcs[i]);
 784                 if (rc)
 785                         break;
 786
 787                 cpus_added++;
 788         }
 789
 790         if (cpus_added < cpus_to_add) {
 791                 pr_warn("CPU hot-add failed, removing any added CPUs\n");
 792
 793                 for (i = 0; i < cpus_added; i++)
 794                         dlpar_cpu_remove_by_index(cpu_drcs[i]);
 795
 796                 rc = -EINVAL;
 797         } else {
 798                 rc = 0;
 799         }
 800
 801         kfree(cpu_drcs);
 802         return rc;
 803 }
 804
 805 int dlpar_cpu(struct pseries_hp_errorlog *hp_elog)
 806 {
 807         u32 count, drc_index;
 808         int rc;
 809
 810         count = hp_elog->_drc_u.drc_count;
 811         drc_index = hp_elog->_drc_u.drc_index;
 812
 813         lock_device_hotplug();
 814
 815         switch (hp_elog->action) {
 816         case PSERIES_HP_ELOG_ACTION_REMOVE:
 817                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
 818                         rc = dlpar_cpu_remove_by_count(count);
 819                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
 820                         rc = dlpar_cpu_remove_by_index(drc_index);
 821                 else
 822                         rc = -EINVAL;
 823                 break;
 824         case PSERIES_HP_ELOG_ACTION_ADD:
 825                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
 826                         rc = dlpar_cpu_add_by_count(count);
 827                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
 828                         rc = dlpar_cpu_add(drc_index);
 829                 else
 830                         rc = -EINVAL;
 831                 break;
 832         default:
 833                 pr_err("Invalid action (%d) specified\n", hp_elog->action);
 834                 rc = -EINVAL;
 835                 break;
 836         }
 837
 838         unlock_device_hotplug();
 839         return rc;
 840 }
 841
 842 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 843
 844 static ssize_t dlpar_cpu_probe(const char *buf, size_t count)
 845 {
 846         u32 drc_index;
 847         int rc;
 848
 849         rc = kstrtou32(buf, 0, &drc_index);
 850         if (rc)
 851                 return -EINVAL;
 852
 853         rc = dlpar_cpu_add(drc_index);
 854
 855         return rc ? rc : count;
 856 }
 857
 858 static ssize_t dlpar_cpu_release(const char *buf, size_t count)
 859 {
 860         struct device_node *dn;
 861         u32 drc_index;
 862         int rc;
 863
 864         dn = of_find_node_by_path(buf);
 865         if (!dn)
 866                 return -EINVAL;
 867
 868         rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index);
 869         if (rc) {
 870                 of_node_put(dn);
 871                 return -EINVAL;
 872         }
 873
 874         rc = dlpar_cpu_remove(dn, drc_index);
 875         of_node_put(dn);
 876
 877         return rc ? rc : count;
 878 }
 879
 880 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 881
 882 static int pseries_smp_notifier(struct notifier_block *nb,
 883                                 unsigned long action, void *data)
 884 {
 885         struct of_reconfig_data *rd = data;
 886         int err = 0;
 887
 888         switch (action) {
 889         case OF_RECONFIG_ATTACH_NODE:
 890                 err = pseries_add_processor(rd->dn);
 891                 break;
 892         case OF_RECONFIG_DETACH_NODE:
 893                 pseries_remove_processor(rd->dn);
 894                 break;
 895         }
 896         return notifier_from_errno(err);
 897 }
 898
 899 static struct notifier_block pseries_smp_nb = {
 900         .notifier_call = pseries_smp_notifier,
 901 };
 902
 903 #define MAX_CEDE_LATENCY_LEVELS         4
 904 #define CEDE_LATENCY_PARAM_LENGTH       10
 905 #define CEDE_LATENCY_PARAM_MAX_LENGTH   \
 906         (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
 907 #define CEDE_LATENCY_TOKEN              45
 908
 909 static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
 910
 911 static int parse_cede_parameters(void)
 912 {
 913         memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
 914         return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 915                          NULL,
 916                          CEDE_LATENCY_TOKEN,
 917                          __pa(cede_parameters),
 918                          CEDE_LATENCY_PARAM_MAX_LENGTH);
 919 }
 920
 921 static int __init pseries_cpu_hotplug_init(void)
 922 {
 923         int cpu;
 924         int qcss_tok;
 925
 926 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 927         ppc_md.cpu_probe = dlpar_cpu_probe;
 928         ppc_md.cpu_release = dlpar_cpu_release;
 929 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 930
 931         rtas_stop_self_token = rtas_token("stop-self");
 932         qcss_tok = rtas_token("query-cpu-stopped-state");
 933
 934         if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE ||
 935                         qcss_tok == RTAS_UNKNOWN_SERVICE) {
 936                 printk(KERN_INFO "CPU Hotplug not supported by firmware "
 937                                 "- disabling.\n");
 938                 return 0;
 939         }
 940
 941         ppc_md.cpu_die = pseries_mach_cpu_die;
 942         smp_ops->cpu_disable = pseries_cpu_disable;
 943         smp_ops->cpu_die = pseries_cpu_die;
 944
 945         /* Processors can be added/removed only on LPAR */
 946         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 947                 of_reconfig_notifier_register(&pseries_smp_nb);
 948                 cpu_maps_update_begin();
 949                 if (cede_offline_enabled && parse_cede_parameters() == 0) {
 950                         default_offline_state = CPU_STATE_INACTIVE;
 951                         for_each_online_cpu(cpu)
 952                                 set_default_offline_state(cpu);
 953                 }
 954                 cpu_maps_update_done();
 955         }
 956
 957         return 0;
 958 }
 959 machine_arch_initcall(pseries, pseries_cpu_hotplug_init);