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/cpu.h>
  28 #include <linux/of.h>
  29 #include <linux/slab.h>
  30 #include <asm/prom.h>
  31 #include <asm/rtas.h>
  32 #include <asm/firmware.h>
  33 #include <asm/machdep.h>
  34 #include <asm/vdso_datapage.h>
  35 #include <asm/xics.h>
  36 #include <asm/plpar_wrappers.h>
  37
  38 #include "pseries.h"
  39 #include "offline_states.h"
  40
  41 /* This version can't take the spinlock, because it never returns */
  42 static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE;
  43
  44 static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) =
  45                                                         CPU_STATE_OFFLINE;
  46 static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE;
  47
  48 static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE;
  49
  50 static int cede_offline_enabled __read_mostly = 1;
  51
  52 /*
  53  * Enable/disable cede_offline when available.
  54  */
  55 static int __init setup_cede_offline(char *str)
  56 {
  57         if (!strcmp(str, "off"))
  58                 cede_offline_enabled = 0;
  59         else if (!strcmp(str, "on"))
  60                 cede_offline_enabled = 1;
  61         else
  62                 return 0;
  63         return 1;
  64 }
  65
  66 __setup("cede_offline=", setup_cede_offline);
  67
  68 enum cpu_state_vals get_cpu_current_state(int cpu)
  69 {
  70         return per_cpu(current_state, cpu);
  71 }
  72
  73 void set_cpu_current_state(int cpu, enum cpu_state_vals state)
  74 {
  75         per_cpu(current_state, cpu) = state;
  76 }
  77
  78 enum cpu_state_vals get_preferred_offline_state(int cpu)
  79 {
  80         return per_cpu(preferred_offline_state, cpu);
  81 }
  82
  83 void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
  84 {
  85         per_cpu(preferred_offline_state, cpu) = state;
  86 }
  87
  88 void set_default_offline_state(int cpu)
  89 {
  90         per_cpu(preferred_offline_state, cpu) = default_offline_state;
  91 }
  92
  93 static void rtas_stop_self(void)
  94 {
  95         static struct rtas_args args;
  96
  97         local_irq_disable();
  98
  99         BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE);
 100
 101         printk("cpu %u (hwid %u) Ready to die...\n",
 102                smp_processor_id(), hard_smp_processor_id());
 103
 104         rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL);
 105
 106         panic("Alas, I survived.\n");
 107 }
 108
 109 static void pseries_mach_cpu_die(void)
 110 {
 111         unsigned int cpu = smp_processor_id();
 112         unsigned int hwcpu = hard_smp_processor_id();
 113         u8 cede_latency_hint = 0;
 114
 115         local_irq_disable();
 116         idle_task_exit();
 117         xics_teardown_cpu();
 118
 119         if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
 120                 set_cpu_current_state(cpu, CPU_STATE_INACTIVE);
 121                 if (ppc_md.suspend_disable_cpu)
 122                         ppc_md.suspend_disable_cpu();
 123
 124                 cede_latency_hint = 2;
 125
 126                 get_lppaca()->idle = 1;
 127                 if (!lppaca_shared_proc(get_lppaca()))
 128                         get_lppaca()->donate_dedicated_cpu = 1;
 129
 130                 while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
 131                         while (!prep_irq_for_idle()) {
 132                                 local_irq_enable();
 133                                 local_irq_disable();
 134                         }
 135
 136                         extended_cede_processor(cede_latency_hint);
 137                 }
 138
 139                 local_irq_disable();
 140
 141                 if (!lppaca_shared_proc(get_lppaca()))
 142                         get_lppaca()->donate_dedicated_cpu = 0;
 143                 get_lppaca()->idle = 0;
 144
 145                 if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) {
 146                         unregister_slb_shadow(hwcpu);
 147
 148                         hard_irq_disable();
 149                         /*
 150                          * Call to start_secondary_resume() will not return.
 151                          * Kernel stack will be reset and start_secondary()
 152                          * will be called to continue the online operation.
 153                          */
 154                         start_secondary_resume();
 155                 }
 156         }
 157
 158         /* Requested state is CPU_STATE_OFFLINE at this point */
 159         WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE);
 160
 161         set_cpu_current_state(cpu, CPU_STATE_OFFLINE);
 162         unregister_slb_shadow(hwcpu);
 163         rtas_stop_self();
 164
 165         /* Should never get here... */
 166         BUG();
 167         for(;;);
 168 }
 169
 170 static int pseries_cpu_disable(void)
 171 {
 172         int cpu = smp_processor_id();
 173
 174         set_cpu_online(cpu, false);
 175         vdso_data->processorCount--;
 176
 177         /*fix boot_cpuid here*/
 178         if (cpu == boot_cpuid)
 179                 boot_cpuid = cpumask_any(cpu_online_mask);
 180
 181         /* FIXME: abstract this to not be platform specific later on */
 182         xics_migrate_irqs_away();
 183         return 0;
 184 }
 185
 186 /*
 187  * pseries_cpu_die: Wait for the cpu to die.
 188  * @cpu: logical processor id of the CPU whose death we're awaiting.
 189  *
 190  * This function is called from the context of the thread which is performing
 191  * the cpu-offline. Here we wait for long enough to allow the cpu in question
 192  * to self-destroy so that the cpu-offline thread can send the CPU_DEAD
 193  * notifications.
 194  *
 195  * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to
 196  * self-destruct.
 197  */
 198 static void pseries_cpu_die(unsigned int cpu)
 199 {
 200         int tries;
 201         int cpu_status = 1;
 202         unsigned int pcpu = get_hard_smp_processor_id(cpu);
 203
 204         if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
 205                 cpu_status = 1;
 206                 for (tries = 0; tries < 5000; tries++) {
 207                         if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) {
 208                                 cpu_status = 0;
 209                                 break;
 210                         }
 211                         msleep(1);
 212                 }
 213         } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
 214
 215                 for (tries = 0; tries < 25; tries++) {
 216                         cpu_status = smp_query_cpu_stopped(pcpu);
 217                         if (cpu_status == QCSS_STOPPED ||
 218                             cpu_status == QCSS_HARDWARE_ERROR)
 219                                 break;
 220                         cpu_relax();
 221                 }
 222         }
 223
 224         if (cpu_status != 0) {
 225                 printk("Querying DEAD? cpu %i (%i) shows %i\n",
 226                        cpu, pcpu, cpu_status);
 227         }
 228
 229         /* Isolation and deallocation are definitely done by
 230          * drslot_chrp_cpu.  If they were not they would be
 231          * done here.  Change isolate state to Isolate and
 232          * change allocation-state to Unusable.
 233          */
 234         paca[cpu].cpu_start = 0;
 235 }
 236
 237 /*
 238  * Update cpu_present_mask and paca(s) for a new cpu node.  The wrinkle
 239  * here is that a cpu device node may represent up to two logical cpus
 240  * in the SMT case.  We must honor the assumption in other code that
 241  * the logical ids for sibling SMT threads x and y are adjacent, such
 242  * that x^1 == y and y^1 == x.
 243  */
 244 static int pseries_add_processor(struct device_node *np)
 245 {
 246         unsigned int cpu;
 247         cpumask_var_t candidate_mask, tmp;
 248         int err = -ENOSPC, len, nthreads, i;
 249         const __be32 *intserv;
 250
 251         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
 252         if (!intserv)
 253                 return 0;
 254
 255         zalloc_cpumask_var(&candidate_mask, GFP_KERNEL);
 256         zalloc_cpumask_var(&tmp, GFP_KERNEL);
 257
 258         nthreads = len / sizeof(u32);
 259         for (i = 0; i < nthreads; i++)
 260                 cpumask_set_cpu(i, tmp);
 261
 262         cpu_maps_update_begin();
 263
 264         BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask));
 265
 266         /* Get a bitmap of unoccupied slots. */
 267         cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask);
 268         if (cpumask_empty(candidate_mask)) {
 269                 /* If we get here, it most likely means that NR_CPUS is
 270                  * less than the partition's max processors setting.
 271                  */
 272                 printk(KERN_ERR "Cannot add cpu %s; this system configuration"
 273                        " supports %d logical cpus.\n", np->full_name,
 274                        num_possible_cpus());
 275                 goto out_unlock;
 276         }
 277
 278         while (!cpumask_empty(tmp))
 279                 if (cpumask_subset(tmp, candidate_mask))
 280                         /* Found a range where we can insert the new cpu(s) */
 281                         break;
 282                 else
 283                         cpumask_shift_left(tmp, tmp, nthreads);
 284
 285         if (cpumask_empty(tmp)) {
 286                 printk(KERN_ERR "Unable to find space in cpu_present_mask for"
 287                        " processor %s with %d thread(s)\n", np->name,
 288                        nthreads);
 289                 goto out_unlock;
 290         }
 291
 292         for_each_cpu(cpu, tmp) {
 293                 BUG_ON(cpu_present(cpu));
 294                 set_cpu_present(cpu, true);
 295                 set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++));
 296         }
 297         err = 0;
 298 out_unlock:
 299         cpu_maps_update_done();
 300         free_cpumask_var(candidate_mask);
 301         free_cpumask_var(tmp);
 302         return err;
 303 }
 304
 305 /*
 306  * Update the present map for a cpu node which is going away, and set
 307  * the hard id in the paca(s) to -1 to be consistent with boot time
 308  * convention for non-present cpus.
 309  */
 310 static void pseries_remove_processor(struct device_node *np)
 311 {
 312         unsigned int cpu;
 313         int len, nthreads, i;
 314         const __be32 *intserv;
 315         u32 thread;
 316
 317         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
 318         if (!intserv)
 319                 return;
 320
 321         nthreads = len / sizeof(u32);
 322
 323         cpu_maps_update_begin();
 324         for (i = 0; i < nthreads; i++) {
 325                 thread = be32_to_cpu(intserv[i]);
 326                 for_each_present_cpu(cpu) {
 327                         if (get_hard_smp_processor_id(cpu) != thread)
 328                                 continue;
 329                         BUG_ON(cpu_online(cpu));
 330                         set_cpu_present(cpu, false);
 331                         set_hard_smp_processor_id(cpu, -1);
 332                         break;
 333                 }
 334                 if (cpu >= nr_cpu_ids)
 335                         printk(KERN_WARNING "Could not find cpu to remove "
 336                                "with physical id 0x%x\n", thread);
 337         }
 338         cpu_maps_update_done();
 339 }
 340
 341 static int dlpar_online_cpu(struct device_node *dn)
 342 {
 343         int rc = 0;
 344         unsigned int cpu;
 345         int len, nthreads, i;
 346         const __be32 *intserv;
 347         u32 thread;
 348
 349         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 350         if (!intserv)
 351                 return -EINVAL;
 352
 353         nthreads = len / sizeof(u32);
 354
 355         cpu_maps_update_begin();
 356         for (i = 0; i < nthreads; i++) {
 357                 thread = be32_to_cpu(intserv[i]);
 358                 for_each_present_cpu(cpu) {
 359                         if (get_hard_smp_processor_id(cpu) != thread)
 360                                 continue;
 361                         BUG_ON(get_cpu_current_state(cpu)
 362                                         != CPU_STATE_OFFLINE);
 363                         cpu_maps_update_done();
 364                         rc = device_online(get_cpu_device(cpu));
 365                         if (rc)
 366                                 goto out;
 367                         cpu_maps_update_begin();
 368
 369                         break;
 370                 }
 371                 if (cpu == num_possible_cpus())
 372                         printk(KERN_WARNING "Could not find cpu to online "
 373                                "with physical id 0x%x\n", thread);
 374         }
 375         cpu_maps_update_done();
 376
 377 out:
 378         return rc;
 379
 380 }
 381
 382 static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index)
 383 {
 384         struct device_node *child = NULL;
 385         u32 my_drc_index;
 386         bool found;
 387         int rc;
 388
 389         /* Assume cpu doesn't exist */
 390         found = false;
 391
 392         for_each_child_of_node(parent, child) {
 393                 rc = of_property_read_u32(child, "ibm,my-drc-index",
 394                                           &my_drc_index);
 395                 if (rc)
 396                         continue;
 397
 398                 if (my_drc_index == drc_index) {
 399                         of_node_put(child);
 400                         found = true;
 401                         break;
 402                 }
 403         }
 404
 405         return found;
 406 }
 407
 408 static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index)
 409 {
 410         bool found = false;
 411         int rc, index;
 412
 413         index = 0;
 414         while (!found) {
 415                 u32 drc;
 416
 417                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
 418                                                 index++, &drc);
 419                 if (rc)
 420                         break;
 421
 422                 if (drc == drc_index)
 423                         found = true;
 424         }
 425
 426         return found;
 427 }
 428
 429 static ssize_t dlpar_cpu_add(u32 drc_index)
 430 {
 431         struct device_node *dn, *parent;
 432         int rc, saved_rc;
 433
 434         pr_debug("Attempting to add CPU, drc index: %x\n", drc_index);
 435
 436         parent = of_find_node_by_path("/cpus");
 437         if (!parent) {
 438                 pr_warn("Failed to find CPU root node \"/cpus\"\n");
 439                 return -ENODEV;
 440         }
 441
 442         if (dlpar_cpu_exists(parent, drc_index)) {
 443                 of_node_put(parent);
 444                 pr_warn("CPU with drc index %x already exists\n", drc_index);
 445                 return -EINVAL;
 446         }
 447
 448         if (!valid_cpu_drc_index(parent, drc_index)) {
 449                 of_node_put(parent);
 450                 pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index);
 451                 return -EINVAL;
 452         }
 453
 454         rc = dlpar_acquire_drc(drc_index);
 455         if (rc) {
 456                 pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n",
 457                         rc, drc_index);
 458                 of_node_put(parent);
 459                 return -EINVAL;
 460         }
 461
 462         dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent);
 463         of_node_put(parent);
 464         if (!dn) {
 465                 pr_warn("Failed call to configure-connector, drc index: %x\n",
 466                         drc_index);
 467                 dlpar_release_drc(drc_index);
 468                 return -EINVAL;
 469         }
 470
 471         rc = dlpar_attach_node(dn);
 472         if (rc) {
 473                 saved_rc = rc;
 474                 pr_warn("Failed to attach node %s, rc: %d, drc index: %x\n",
 475                         dn->name, rc, drc_index);
 476
 477                 rc = dlpar_release_drc(drc_index);
 478                 if (!rc)
 479                         dlpar_free_cc_nodes(dn);
 480
 481                 return saved_rc;
 482         }
 483
 484         rc = dlpar_online_cpu(dn);
 485         if (rc) {
 486                 saved_rc = rc;
 487                 pr_warn("Failed to online cpu %s, rc: %d, drc index: %x\n",
 488                         dn->name, rc, drc_index);
 489
 490                 rc = dlpar_detach_node(dn);
 491                 if (!rc)
 492                         dlpar_release_drc(drc_index);
 493
 494                 return saved_rc;
 495         }
 496
 497         pr_debug("Successfully added CPU %s, drc index: %x\n", dn->name,
 498                  drc_index);
 499         return rc;
 500 }
 501
 502 static int dlpar_offline_cpu(struct device_node *dn)
 503 {
 504         int rc = 0;
 505         unsigned int cpu;
 506         int len, nthreads, i;
 507         const __be32 *intserv;
 508         u32 thread;
 509
 510         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 511         if (!intserv)
 512                 return -EINVAL;
 513
 514         nthreads = len / sizeof(u32);
 515
 516         cpu_maps_update_begin();
 517         for (i = 0; i < nthreads; i++) {
 518                 thread = be32_to_cpu(intserv[i]);
 519                 for_each_present_cpu(cpu) {
 520                         if (get_hard_smp_processor_id(cpu) != thread)
 521                                 continue;
 522
 523                         if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE)
 524                                 break;
 525
 526                         if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
 527                                 set_preferred_offline_state(cpu,
 528                                                             CPU_STATE_OFFLINE);
 529                                 cpu_maps_update_done();
 530                                 rc = device_offline(get_cpu_device(cpu));
 531                                 if (rc)
 532                                         goto out;
 533                                 cpu_maps_update_begin();
 534                                 break;
 535
 536                         }
 537
 538                         /*
 539                          * The cpu is in CPU_STATE_INACTIVE.
 540                          * Upgrade it's state to CPU_STATE_OFFLINE.
 541                          */
 542                         set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
 543                         BUG_ON(plpar_hcall_norets(H_PROD, thread)
 544                                                                 != H_SUCCESS);
 545                         __cpu_die(cpu);
 546                         break;
 547                 }
 548                 if (cpu == num_possible_cpus())
 549                         printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread);
 550         }
 551         cpu_maps_update_done();
 552
 553 out:
 554         return rc;
 555
 556 }
 557
 558 static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index)
 559 {
 560         int rc;
 561
 562         pr_debug("Attemping to remove CPU %s, drc index: %x\n",
 563                  dn->name, drc_index);
 564
 565         rc = dlpar_offline_cpu(dn);
 566         if (rc) {
 567                 pr_warn("Failed to offline CPU %s, rc: %d\n", dn->name, rc);
 568                 return -EINVAL;
 569         }
 570
 571         rc = dlpar_release_drc(drc_index);
 572         if (rc) {
 573                 pr_warn("Failed to release drc (%x) for CPU %s, rc: %d\n",
 574                         drc_index, dn->name, rc);
 575                 dlpar_online_cpu(dn);
 576                 return rc;
 577         }
 578
 579         rc = dlpar_detach_node(dn);
 580         if (rc) {
 581                 int saved_rc = rc;
 582
 583                 pr_warn("Failed to detach CPU %s, rc: %d", dn->name, rc);
 584
 585                 rc = dlpar_acquire_drc(drc_index);
 586                 if (!rc)
 587                         dlpar_online_cpu(dn);
 588
 589                 return saved_rc;
 590         }
 591
 592         pr_debug("Successfully removed CPU, drc index: %x\n", drc_index);
 593         return 0;
 594 }
 595
 596 static struct device_node *cpu_drc_index_to_dn(u32 drc_index)
 597 {
 598         struct device_node *dn;
 599         u32 my_index;
 600         int rc;
 601
 602         for_each_node_by_type(dn, "cpu") {
 603                 rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index);
 604                 if (rc)
 605                         continue;
 606
 607                 if (my_index == drc_index)
 608                         break;
 609         }
 610
 611         return dn;
 612 }
 613
 614 static int dlpar_cpu_remove_by_index(u32 drc_index)
 615 {
 616         struct device_node *dn;
 617         int rc;
 618
 619         dn = cpu_drc_index_to_dn(drc_index);
 620         if (!dn) {
 621                 pr_warn("Cannot find CPU (drc index %x) to remove\n",
 622                         drc_index);
 623                 return -ENODEV;
 624         }
 625
 626         rc = dlpar_cpu_remove(dn, drc_index);
 627         of_node_put(dn);
 628         return rc;
 629 }
 630
 631 static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove)
 632 {
 633         struct device_node *dn;
 634         int cpus_found = 0;
 635         int rc;
 636
 637         /* We want to find cpus_to_remove + 1 CPUs to ensure we do not
 638          * remove the last CPU.
 639          */
 640         for_each_node_by_type(dn, "cpu") {
 641                 cpus_found++;
 642
 643                 if (cpus_found > cpus_to_remove) {
 644                         of_node_put(dn);
 645                         break;
 646                 }
 647
 648                 /* Note that cpus_found is always 1 ahead of the index
 649                  * into the cpu_drcs array, so we use cpus_found - 1
 650                  */
 651                 rc = of_property_read_u32(dn, "ibm,my-drc-index",
 652                                           &cpu_drcs[cpus_found - 1]);
 653                 if (rc) {
 654                         pr_warn("Error occurred getting drc-index for %s\n",
 655                                 dn->name);
 656                         of_node_put(dn);
 657                         return -1;
 658                 }
 659         }
 660
 661         if (cpus_found < cpus_to_remove) {
 662                 pr_warn("Failed to find enough CPUs (%d of %d) to remove\n",
 663                         cpus_found, cpus_to_remove);
 664         } else if (cpus_found == cpus_to_remove) {
 665                 pr_warn("Cannot remove all CPUs\n");
 666         }
 667
 668         return cpus_found;
 669 }
 670
 671 static int dlpar_cpu_remove_by_count(u32 cpus_to_remove)
 672 {
 673         u32 *cpu_drcs;
 674         int cpus_found;
 675         int cpus_removed = 0;
 676         int i, rc;
 677
 678         pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove);
 679
 680         cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL);
 681         if (!cpu_drcs)
 682                 return -EINVAL;
 683
 684         cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove);
 685         if (cpus_found <= cpus_to_remove) {
 686                 kfree(cpu_drcs);
 687                 return -EINVAL;
 688         }
 689
 690         for (i = 0; i < cpus_to_remove; i++) {
 691                 rc = dlpar_cpu_remove_by_index(cpu_drcs[i]);
 692                 if (rc)
 693                         break;
 694
 695                 cpus_removed++;
 696         }
 697
 698         if (cpus_removed != cpus_to_remove) {
 699                 pr_warn("CPU hot-remove failed, adding back removed CPUs\n");
 700
 701                 for (i = 0; i < cpus_removed; i++)
 702                         dlpar_cpu_add(cpu_drcs[i]);
 703
 704                 rc = -EINVAL;
 705         } else {
 706                 rc = 0;
 707         }
 708
 709         kfree(cpu_drcs);
 710         return rc;
 711 }
 712
 713 static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add)
 714 {
 715         struct device_node *parent;
 716         int cpus_found = 0;
 717         int index, rc;
 718
 719         parent = of_find_node_by_path("/cpus");
 720         if (!parent) {
 721                 pr_warn("Could not find CPU root node in device tree\n");
 722                 kfree(cpu_drcs);
 723                 return -1;
 724         }
 725
 726         /* Search the ibm,drc-indexes array for possible CPU drcs to
 727          * add. Note that the format of the ibm,drc-indexes array is
 728          * the number of entries in the array followed by the array
 729          * of drc values so we start looking at index = 1.
 730          */
 731         index = 1;
 732         while (cpus_found < cpus_to_add) {
 733                 u32 drc;
 734
 735                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
 736                                                 index++, &drc);
 737                 if (rc)
 738                         break;
 739
 740                 if (dlpar_cpu_exists(parent, drc))
 741                         continue;
 742
 743                 cpu_drcs[cpus_found++] = drc;
 744         }
 745
 746         of_node_put(parent);
 747         return cpus_found;
 748 }
 749
 750 static int dlpar_cpu_add_by_count(u32 cpus_to_add)
 751 {
 752         u32 *cpu_drcs;
 753         int cpus_added = 0;
 754         int cpus_found;
 755         int i, rc;
 756
 757         pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add);
 758
 759         cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL);
 760         if (!cpu_drcs)
 761                 return -EINVAL;
 762
 763         cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add);
 764         if (cpus_found < cpus_to_add) {
 765                 pr_warn("Failed to find enough CPUs (%d of %d) to add\n",
 766                         cpus_found, cpus_to_add);
 767                 kfree(cpu_drcs);
 768                 return -EINVAL;
 769         }
 770
 771         for (i = 0; i < cpus_to_add; i++) {
 772                 rc = dlpar_cpu_add(cpu_drcs[i]);
 773                 if (rc)
 774                         break;
 775
 776                 cpus_added++;
 777         }
 778
 779         if (cpus_added < cpus_to_add) {
 780                 pr_warn("CPU hot-add failed, removing any added CPUs\n");
 781
 782                 for (i = 0; i < cpus_added; i++)
 783                         dlpar_cpu_remove_by_index(cpu_drcs[i]);
 784
 785                 rc = -EINVAL;
 786         } else {
 787                 rc = 0;
 788         }
 789
 790         kfree(cpu_drcs);
 791         return rc;
 792 }
 793
 794 int dlpar_cpu(struct pseries_hp_errorlog *hp_elog)
 795 {
 796         u32 count, drc_index;
 797         int rc;
 798
 799         count = hp_elog->_drc_u.drc_count;
 800         drc_index = hp_elog->_drc_u.drc_index;
 801
 802         lock_device_hotplug();
 803
 804         switch (hp_elog->action) {
 805         case PSERIES_HP_ELOG_ACTION_REMOVE:
 806                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
 807                         rc = dlpar_cpu_remove_by_count(count);
 808                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
 809                         rc = dlpar_cpu_remove_by_index(drc_index);
 810                 else
 811                         rc = -EINVAL;
 812                 break;
 813         case PSERIES_HP_ELOG_ACTION_ADD:
 814                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
 815                         rc = dlpar_cpu_add_by_count(count);
 816                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
 817                         rc = dlpar_cpu_add(drc_index);
 818                 else
 819                         rc = -EINVAL;
 820                 break;
 821         default:
 822                 pr_err("Invalid action (%d) specified\n", hp_elog->action);
 823                 rc = -EINVAL;
 824                 break;
 825         }
 826
 827         unlock_device_hotplug();
 828         return rc;
 829 }
 830
 831 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 832
 833 static ssize_t dlpar_cpu_probe(const char *buf, size_t count)
 834 {
 835         u32 drc_index;
 836         int rc;
 837
 838         rc = kstrtou32(buf, 0, &drc_index);
 839         if (rc)
 840                 return -EINVAL;
 841
 842         rc = dlpar_cpu_add(drc_index);
 843
 844         return rc ? rc : count;
 845 }
 846
 847 static ssize_t dlpar_cpu_release(const char *buf, size_t count)
 848 {
 849         struct device_node *dn;
 850         u32 drc_index;
 851         int rc;
 852
 853         dn = of_find_node_by_path(buf);
 854         if (!dn)
 855                 return -EINVAL;
 856
 857         rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index);
 858         if (rc) {
 859                 of_node_put(dn);
 860                 return -EINVAL;
 861         }
 862
 863         rc = dlpar_cpu_remove(dn, drc_index);
 864         of_node_put(dn);
 865
 866         return rc ? rc : count;
 867 }
 868
 869 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 870
 871 static int pseries_smp_notifier(struct notifier_block *nb,
 872                                 unsigned long action, void *data)
 873 {
 874         struct of_reconfig_data *rd = data;
 875         int err = 0;
 876
 877         switch (action) {
 878         case OF_RECONFIG_ATTACH_NODE:
 879                 err = pseries_add_processor(rd->dn);
 880                 break;
 881         case OF_RECONFIG_DETACH_NODE:
 882                 pseries_remove_processor(rd->dn);
 883                 break;
 884         }
 885         return notifier_from_errno(err);
 886 }
 887
 888 static struct notifier_block pseries_smp_nb = {
 889         .notifier_call = pseries_smp_notifier,
 890 };
 891
 892 #define MAX_CEDE_LATENCY_LEVELS         4
 893 #define CEDE_LATENCY_PARAM_LENGTH       10
 894 #define CEDE_LATENCY_PARAM_MAX_LENGTH   \
 895         (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
 896 #define CEDE_LATENCY_TOKEN              45
 897
 898 static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
 899
 900 static int parse_cede_parameters(void)
 901 {
 902         memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
 903         return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 904                          NULL,
 905                          CEDE_LATENCY_TOKEN,
 906                          __pa(cede_parameters),
 907                          CEDE_LATENCY_PARAM_MAX_LENGTH);
 908 }
 909
 910 static int __init pseries_cpu_hotplug_init(void)
 911 {
 912         struct device_node *np;
 913         const char *typep;
 914         int cpu;
 915         int qcss_tok;
 916
 917 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 918         ppc_md.cpu_probe = dlpar_cpu_probe;
 919         ppc_md.cpu_release = dlpar_cpu_release;
 920 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 921
 922         for_each_node_by_name(np, "interrupt-controller") {
 923                 typep = of_get_property(np, "compatible", NULL);
 924                 if (strstr(typep, "open-pic")) {
 925                         of_node_put(np);
 926
 927                         printk(KERN_INFO "CPU Hotplug not supported on "
 928                                 "systems using MPIC\n");
 929                         return 0;
 930                 }
 931         }
 932
 933         rtas_stop_self_token = rtas_token("stop-self");
 934         qcss_tok = rtas_token("query-cpu-stopped-state");
 935
 936         if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE ||
 937                         qcss_tok == RTAS_UNKNOWN_SERVICE) {
 938                 printk(KERN_INFO "CPU Hotplug not supported by firmware "
 939                                 "- disabling.\n");
 940                 return 0;
 941         }
 942
 943         ppc_md.cpu_die = pseries_mach_cpu_die;
 944         smp_ops->cpu_disable = pseries_cpu_disable;
 945         smp_ops->cpu_die = pseries_cpu_die;
 946
 947         /* Processors can be added/removed only on LPAR */
 948         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 949                 of_reconfig_notifier_register(&pseries_smp_nb);
 950                 cpu_maps_update_begin();
 951                 if (cede_offline_enabled && parse_cede_parameters() == 0) {
 952                         default_offline_state = CPU_STATE_INACTIVE;
 953                         for_each_online_cpu(cpu)
 954                                 set_default_offline_state(cpu);
 955                 }
 956                 cpu_maps_update_done();
 957         }
 958
 959         return 0;
 960 }
 961 machine_arch_initcall(pseries, pseries_cpu_hotplug_init);