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 static int dlpar_online_cpu(struct device_node *dn)
 344 {
 345         int rc = 0;
 346         unsigned int cpu;
 347         int len, nthreads, i;
 348         const __be32 *intserv;
 349         u32 thread;
 350
 351         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 352         if (!intserv)
 353                 return -EINVAL;
 354
 355         nthreads = len / sizeof(u32);
 356
 357         cpu_maps_update_begin();
 358         for (i = 0; i < nthreads; i++) {
 359                 thread = be32_to_cpu(intserv[i]);
 360                 for_each_present_cpu(cpu) {
 361                         if (get_hard_smp_processor_id(cpu) != thread)
 362                                 continue;
 363                         BUG_ON(get_cpu_current_state(cpu)
 364                                         != CPU_STATE_OFFLINE);
 365                         cpu_maps_update_done();
 366                         rc = device_online(get_cpu_device(cpu));
 367                         if (rc)
 368                                 goto out;
 369                         cpu_maps_update_begin();
 370
 371                         break;
 372                 }
 373                 if (cpu == num_possible_cpus())
 374                         printk(KERN_WARNING "Could not find cpu to online "
 375                                "with physical id 0x%x\n", thread);
 376         }
 377         cpu_maps_update_done();
 378
 379 out:
 380         return rc;
 381
 382 }
 383
 384 static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index)
 385 {
 386         struct device_node *child = NULL;
 387         u32 my_drc_index;
 388         bool found;
 389         int rc;
 390
 391         /* Assume cpu doesn't exist */
 392         found = false;
 393
 394         for_each_child_of_node(parent, child) {
 395                 rc = of_property_read_u32(child, "ibm,my-drc-index",
 396                                           &my_drc_index);
 397                 if (rc)
 398                         continue;
 399
 400                 if (my_drc_index == drc_index) {
 401                         of_node_put(child);
 402                         found = true;
 403                         break;
 404                 }
 405         }
 406
 407         return found;
 408 }
 409
 410 static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index)
 411 {
 412         bool found = false;
 413         int rc, index;
 414
 415         index = 0;
 416         while (!found) {
 417                 u32 drc;
 418
 419                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
 420                                                 index++, &drc);
 421                 if (rc)
 422                         break;
 423
 424                 if (drc == drc_index)
 425                         found = true;
 426         }
 427
 428         return found;
 429 }
 430
 431 static ssize_t dlpar_cpu_add(u32 drc_index)
 432 {
 433         struct device_node *dn, *parent;
 434         int rc, saved_rc;
 435
 436         pr_debug("Attempting to add CPU, drc index: %x\n", drc_index);
 437
 438         parent = of_find_node_by_path("/cpus");
 439         if (!parent) {
 440                 pr_warn("Failed to find CPU root node \"/cpus\"\n");
 441                 return -ENODEV;
 442         }
 443
 444         if (dlpar_cpu_exists(parent, drc_index)) {
 445                 of_node_put(parent);
 446                 pr_warn("CPU with drc index %x already exists\n", drc_index);
 447                 return -EINVAL;
 448         }
 449
 450         if (!valid_cpu_drc_index(parent, drc_index)) {
 451                 of_node_put(parent);
 452                 pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index);
 453                 return -EINVAL;
 454         }
 455
 456         rc = dlpar_acquire_drc(drc_index);
 457         if (rc) {
 458                 pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n",
 459                         rc, drc_index);
 460                 of_node_put(parent);
 461                 return -EINVAL;
 462         }
 463
 464         dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent);
 465         if (!dn) {
 466                 pr_warn("Failed call to configure-connector, drc index: %x\n",
 467                         drc_index);
 468                 dlpar_release_drc(drc_index);
 469                 of_node_put(parent);
 470                 return -EINVAL;
 471         }
 472
 473         rc = dlpar_attach_node(dn, parent);
 474
 475         /* Regardless we are done with parent now */
 476         of_node_put(parent);
 477
 478         if (rc) {
 479                 saved_rc = rc;
 480                 pr_warn("Failed to attach node %s, rc: %d, drc index: %x\n",
 481                         dn->name, rc, drc_index);
 482
 483                 rc = dlpar_release_drc(drc_index);
 484                 if (!rc)
 485                         dlpar_free_cc_nodes(dn);
 486
 487                 return saved_rc;
 488         }
 489
 490         rc = dlpar_online_cpu(dn);
 491         if (rc) {
 492                 saved_rc = rc;
 493                 pr_warn("Failed to online cpu %s, rc: %d, drc index: %x\n",
 494                         dn->name, rc, drc_index);
 495
 496                 rc = dlpar_detach_node(dn);
 497                 if (!rc)
 498                         dlpar_release_drc(drc_index);
 499
 500                 return saved_rc;
 501         }
 502
 503         pr_debug("Successfully added CPU %s, drc index: %x\n", dn->name,
 504                  drc_index);
 505         return rc;
 506 }
 507
 508 static int dlpar_offline_cpu(struct device_node *dn)
 509 {
 510         int rc = 0;
 511         unsigned int cpu;
 512         int len, nthreads, i;
 513         const __be32 *intserv;
 514         u32 thread;
 515
 516         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
 517         if (!intserv)
 518                 return -EINVAL;
 519
 520         nthreads = len / sizeof(u32);
 521
 522         cpu_maps_update_begin();
 523         for (i = 0; i < nthreads; i++) {
 524                 thread = be32_to_cpu(intserv[i]);
 525                 for_each_present_cpu(cpu) {
 526                         if (get_hard_smp_processor_id(cpu) != thread)
 527                                 continue;
 528
 529                         if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE)
 530                                 break;
 531
 532                         if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
 533                                 set_preferred_offline_state(cpu,
 534                                                             CPU_STATE_OFFLINE);
 535                                 cpu_maps_update_done();
 536                                 rc = device_offline(get_cpu_device(cpu));
 537                                 if (rc)
 538                                         goto out;
 539                                 cpu_maps_update_begin();
 540                                 break;
 541
 542                         }
 543
 544                         /*
 545                          * The cpu is in CPU_STATE_INACTIVE.
 546                          * Upgrade it's state to CPU_STATE_OFFLINE.
 547                          */
 548                         set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
 549                         BUG_ON(plpar_hcall_norets(H_PROD, thread)
 550                                                                 != H_SUCCESS);
 551                         __cpu_die(cpu);
 552                         break;
 553                 }
 554                 if (cpu == num_possible_cpus())
 555                         printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread);
 556         }
 557         cpu_maps_update_done();
 558
 559 out:
 560         return rc;
 561
 562 }
 563
 564 static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index)
 565 {
 566         int rc;
 567
 568         pr_debug("Attempting to remove CPU %s, drc index: %x\n",
 569                  dn->name, drc_index);
 570
 571         rc = dlpar_offline_cpu(dn);
 572         if (rc) {
 573                 pr_warn("Failed to offline CPU %s, rc: %d\n", dn->name, rc);
 574                 return -EINVAL;
 575         }
 576
 577         rc = dlpar_release_drc(drc_index);
 578         if (rc) {
 579                 pr_warn("Failed to release drc (%x) for CPU %s, rc: %d\n",
 580                         drc_index, dn->name, rc);
 581                 dlpar_online_cpu(dn);
 582                 return rc;
 583         }
 584
 585         rc = dlpar_detach_node(dn);
 586         if (rc) {
 587                 int saved_rc = rc;
 588
 589                 pr_warn("Failed to detach CPU %s, rc: %d", dn->name, rc);
 590
 591                 rc = dlpar_acquire_drc(drc_index);
 592                 if (!rc)
 593                         dlpar_online_cpu(dn);
 594
 595                 return saved_rc;
 596         }
 597
 598         pr_debug("Successfully removed CPU, drc index: %x\n", drc_index);
 599         return 0;
 600 }
 601
 602 static struct device_node *cpu_drc_index_to_dn(u32 drc_index)
 603 {
 604         struct device_node *dn;
 605         u32 my_index;
 606         int rc;
 607
 608         for_each_node_by_type(dn, "cpu") {
 609                 rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index);
 610                 if (rc)
 611                         continue;
 612
 613                 if (my_index == drc_index)
 614                         break;
 615         }
 616
 617         return dn;
 618 }
 619
 620 static int dlpar_cpu_remove_by_index(u32 drc_index)
 621 {
 622         struct device_node *dn;
 623         int rc;
 624
 625         dn = cpu_drc_index_to_dn(drc_index);
 626         if (!dn) {
 627                 pr_warn("Cannot find CPU (drc index %x) to remove\n",
 628                         drc_index);
 629                 return -ENODEV;
 630         }
 631
 632         rc = dlpar_cpu_remove(dn, drc_index);
 633         of_node_put(dn);
 634         return rc;
 635 }
 636
 637 static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove)
 638 {
 639         struct device_node *dn;
 640         int cpus_found = 0;
 641         int rc;
 642
 643         /* We want to find cpus_to_remove + 1 CPUs to ensure we do not
 644          * remove the last CPU.
 645          */
 646         for_each_node_by_type(dn, "cpu") {
 647                 cpus_found++;
 648
 649                 if (cpus_found > cpus_to_remove) {
 650                         of_node_put(dn);
 651                         break;
 652                 }
 653
 654                 /* Note that cpus_found is always 1 ahead of the index
 655                  * into the cpu_drcs array, so we use cpus_found - 1
 656                  */
 657                 rc = of_property_read_u32(dn, "ibm,my-drc-index",
 658                                           &cpu_drcs[cpus_found - 1]);
 659                 if (rc) {
 660                         pr_warn("Error occurred getting drc-index for %s\n",
 661                                 dn->name);
 662                         of_node_put(dn);
 663                         return -1;
 664                 }
 665         }
 666
 667         if (cpus_found < cpus_to_remove) {
 668                 pr_warn("Failed to find enough CPUs (%d of %d) to remove\n",
 669                         cpus_found, cpus_to_remove);
 670         } else if (cpus_found == cpus_to_remove) {
 671                 pr_warn("Cannot remove all CPUs\n");
 672         }
 673
 674         return cpus_found;
 675 }
 676
 677 static int dlpar_cpu_remove_by_count(u32 cpus_to_remove)
 678 {
 679         u32 *cpu_drcs;
 680         int cpus_found;
 681         int cpus_removed = 0;
 682         int i, rc;
 683
 684         pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove);
 685
 686         cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL);
 687         if (!cpu_drcs)
 688                 return -EINVAL;
 689
 690         cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove);
 691         if (cpus_found <= cpus_to_remove) {
 692                 kfree(cpu_drcs);
 693                 return -EINVAL;
 694         }
 695
 696         for (i = 0; i < cpus_to_remove; i++) {
 697                 rc = dlpar_cpu_remove_by_index(cpu_drcs[i]);
 698                 if (rc)
 699                         break;
 700
 701                 cpus_removed++;
 702         }
 703
 704         if (cpus_removed != cpus_to_remove) {
 705                 pr_warn("CPU hot-remove failed, adding back removed CPUs\n");
 706
 707                 for (i = 0; i < cpus_removed; i++)
 708                         dlpar_cpu_add(cpu_drcs[i]);
 709
 710                 rc = -EINVAL;
 711         } else {
 712                 rc = 0;
 713         }
 714
 715         kfree(cpu_drcs);
 716         return rc;
 717 }
 718
 719 static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add)
 720 {
 721         struct device_node *parent;
 722         int cpus_found = 0;
 723         int index, rc;
 724
 725         parent = of_find_node_by_path("/cpus");
 726         if (!parent) {
 727                 pr_warn("Could not find CPU root node in device tree\n");
 728                 kfree(cpu_drcs);
 729                 return -1;
 730         }
 731
 732         /* Search the ibm,drc-indexes array for possible CPU drcs to
 733          * add. Note that the format of the ibm,drc-indexes array is
 734          * the number of entries in the array followed by the array
 735          * of drc values so we start looking at index = 1.
 736          */
 737         index = 1;
 738         while (cpus_found < cpus_to_add) {
 739                 u32 drc;
 740
 741                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
 742                                                 index++, &drc);
 743                 if (rc)
 744                         break;
 745
 746                 if (dlpar_cpu_exists(parent, drc))
 747                         continue;
 748
 749                 cpu_drcs[cpus_found++] = drc;
 750         }
 751
 752         of_node_put(parent);
 753         return cpus_found;
 754 }
 755
 756 static int dlpar_cpu_add_by_count(u32 cpus_to_add)
 757 {
 758         u32 *cpu_drcs;
 759         int cpus_added = 0;
 760         int cpus_found;
 761         int i, rc;
 762
 763         pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add);
 764
 765         cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL);
 766         if (!cpu_drcs)
 767                 return -EINVAL;
 768
 769         cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add);
 770         if (cpus_found < cpus_to_add) {
 771                 pr_warn("Failed to find enough CPUs (%d of %d) to add\n",
 772                         cpus_found, cpus_to_add);
 773                 kfree(cpu_drcs);
 774                 return -EINVAL;
 775         }
 776
 777         for (i = 0; i < cpus_to_add; i++) {
 778                 rc = dlpar_cpu_add(cpu_drcs[i]);
 779                 if (rc)
 780                         break;
 781
 782                 cpus_added++;
 783         }
 784
 785         if (cpus_added < cpus_to_add) {
 786                 pr_warn("CPU hot-add failed, removing any added CPUs\n");
 787
 788                 for (i = 0; i < cpus_added; i++)
 789                         dlpar_cpu_remove_by_index(cpu_drcs[i]);
 790
 791                 rc = -EINVAL;
 792         } else {
 793                 rc = 0;
 794         }
 795
 796         kfree(cpu_drcs);
 797         return rc;
 798 }
 799
 800 int dlpar_cpu(struct pseries_hp_errorlog *hp_elog)
 801 {
 802         u32 count, drc_index;
 803         int rc;
 804
 805         count = hp_elog->_drc_u.drc_count;
 806         drc_index = hp_elog->_drc_u.drc_index;
 807
 808         lock_device_hotplug();
 809
 810         switch (hp_elog->action) {
 811         case PSERIES_HP_ELOG_ACTION_REMOVE:
 812                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
 813                         rc = dlpar_cpu_remove_by_count(count);
 814                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
 815                         rc = dlpar_cpu_remove_by_index(drc_index);
 816                 else
 817                         rc = -EINVAL;
 818                 break;
 819         case PSERIES_HP_ELOG_ACTION_ADD:
 820                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
 821                         rc = dlpar_cpu_add_by_count(count);
 822                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
 823                         rc = dlpar_cpu_add(drc_index);
 824                 else
 825                         rc = -EINVAL;
 826                 break;
 827         default:
 828                 pr_err("Invalid action (%d) specified\n", hp_elog->action);
 829                 rc = -EINVAL;
 830                 break;
 831         }
 832
 833         unlock_device_hotplug();
 834         return rc;
 835 }
 836
 837 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 838
 839 static ssize_t dlpar_cpu_probe(const char *buf, size_t count)
 840 {
 841         u32 drc_index;
 842         int rc;
 843
 844         rc = kstrtou32(buf, 0, &drc_index);
 845         if (rc)
 846                 return -EINVAL;
 847
 848         rc = dlpar_cpu_add(drc_index);
 849
 850         return rc ? rc : count;
 851 }
 852
 853 static ssize_t dlpar_cpu_release(const char *buf, size_t count)
 854 {
 855         struct device_node *dn;
 856         u32 drc_index;
 857         int rc;
 858
 859         dn = of_find_node_by_path(buf);
 860         if (!dn)
 861                 return -EINVAL;
 862
 863         rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index);
 864         if (rc) {
 865                 of_node_put(dn);
 866                 return -EINVAL;
 867         }
 868
 869         rc = dlpar_cpu_remove(dn, drc_index);
 870         of_node_put(dn);
 871
 872         return rc ? rc : count;
 873 }
 874
 875 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 876
 877 static int pseries_smp_notifier(struct notifier_block *nb,
 878                                 unsigned long action, void *data)
 879 {
 880         struct of_reconfig_data *rd = data;
 881         int err = 0;
 882
 883         switch (action) {
 884         case OF_RECONFIG_ATTACH_NODE:
 885                 err = pseries_add_processor(rd->dn);
 886                 break;
 887         case OF_RECONFIG_DETACH_NODE:
 888                 pseries_remove_processor(rd->dn);
 889                 break;
 890         }
 891         return notifier_from_errno(err);
 892 }
 893
 894 static struct notifier_block pseries_smp_nb = {
 895         .notifier_call = pseries_smp_notifier,
 896 };
 897
 898 #define MAX_CEDE_LATENCY_LEVELS         4
 899 #define CEDE_LATENCY_PARAM_LENGTH       10
 900 #define CEDE_LATENCY_PARAM_MAX_LENGTH   \
 901         (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
 902 #define CEDE_LATENCY_TOKEN              45
 903
 904 static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
 905
 906 static int parse_cede_parameters(void)
 907 {
 908         memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
 909         return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 910                          NULL,
 911                          CEDE_LATENCY_TOKEN,
 912                          __pa(cede_parameters),
 913                          CEDE_LATENCY_PARAM_MAX_LENGTH);
 914 }
 915
 916 static int __init pseries_cpu_hotplug_init(void)
 917 {
 918         int cpu;
 919         int qcss_tok;
 920
 921 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
 922         ppc_md.cpu_probe = dlpar_cpu_probe;
 923         ppc_md.cpu_release = dlpar_cpu_release;
 924 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
 925
 926         rtas_stop_self_token = rtas_token("stop-self");
 927         qcss_tok = rtas_token("query-cpu-stopped-state");
 928
 929         if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE ||
 930                         qcss_tok == RTAS_UNKNOWN_SERVICE) {
 931                 printk(KERN_INFO "CPU Hotplug not supported by firmware "
 932                                 "- disabling.\n");
 933                 return 0;
 934         }
 935
 936         ppc_md.cpu_die = pseries_mach_cpu_die;
 937         smp_ops->cpu_disable = pseries_cpu_disable;
 938         smp_ops->cpu_die = pseries_cpu_die;
 939
 940         /* Processors can be added/removed only on LPAR */
 941         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 942                 of_reconfig_notifier_register(&pseries_smp_nb);
 943                 cpu_maps_update_begin();
 944                 if (cede_offline_enabled && parse_cede_parameters() == 0) {
 945                         default_offline_state = CPU_STATE_INACTIVE;
 946                         for_each_online_cpu(cpu)
 947                                 set_default_offline_state(cpu);
 948                 }
 949                 cpu_maps_update_done();
 950         }
 951
 952         return 0;
 953 }
 954 machine_arch_initcall(pseries, pseries_cpu_hotplug_init);