arch/powerpc/platforms/powernv/idle.c

   1 /*
   2  * PowerNV cpuidle code
   3  *
   4  * Copyright 2015 IBM Corp.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/types.h>
  13 #include <linux/mm.h>
  14 #include <linux/slab.h>
  15 #include <linux/of.h>
  16 #include <linux/device.h>
  17 #include <linux/cpu.h>
  18
  19 #include <asm/firmware.h>
  20 #include <asm/machdep.h>
  21 #include <asm/opal.h>
  22 #include <asm/cputhreads.h>
  23 #include <asm/cpuidle.h>
  24 #include <asm/code-patching.h>
  25 #include <asm/smp.h>
  26 #include <asm/runlatch.h>
  27
  28 #include "powernv.h"
  29 #include "subcore.h"
  30
  31 /* Power ISA 3.0 allows for stop states 0x0 - 0xF */
  32 #define MAX_STOP_STATE  0xF
  33
  34 #define P9_STOP_SPR_MSR 2000
  35 #define P9_STOP_SPR_PSSCR      855
  36
  37 static u32 supported_cpuidle_states;
  38
  39 /*
  40  * The default stop state that will be used by ppc_md.power_save
  41  * function on platforms that support stop instruction.
  42  */
  43 static u64 pnv_default_stop_val;
  44 static u64 pnv_default_stop_mask;
  45 static bool default_stop_found;
  46
  47 /*
  48  * First deep stop state. Used to figure out when to save/restore
  49  * hypervisor context.
  50  */
  51 u64 pnv_first_deep_stop_state = MAX_STOP_STATE;
  52
  53 /*
  54  * psscr value and mask of the deepest stop idle state.
  55  * Used when a cpu is offlined.
  56  */
  57 static u64 pnv_deepest_stop_psscr_val;
  58 static u64 pnv_deepest_stop_psscr_mask;
  59 static bool deepest_stop_found;
  60
  61 static int pnv_save_sprs_for_deep_states(void)
  62 {
  63         int cpu;
  64         int rc;
  65
  66         /*
  67          * hid0, hid1, hid4, hid5, hmeer and lpcr values are symmetric across
  68          * all cpus at boot. Get these reg values of current cpu and use the
  69          * same across all cpus.
  70          */
  71         uint64_t lpcr_val = mfspr(SPRN_LPCR) & ~(u64)LPCR_PECE1;
  72         uint64_t hid0_val = mfspr(SPRN_HID0);
  73         uint64_t hid1_val = mfspr(SPRN_HID1);
  74         uint64_t hid4_val = mfspr(SPRN_HID4);
  75         uint64_t hid5_val = mfspr(SPRN_HID5);
  76         uint64_t hmeer_val = mfspr(SPRN_HMEER);
  77         uint64_t msr_val = MSR_IDLE;
  78         uint64_t psscr_val = pnv_deepest_stop_psscr_val;
  79
  80         for_each_possible_cpu(cpu) {
  81                 uint64_t pir = get_hard_smp_processor_id(cpu);
  82                 uint64_t hsprg0_val = (uint64_t)&paca[cpu];
  83
  84                 rc = opal_slw_set_reg(pir, SPRN_HSPRG0, hsprg0_val);
  85                 if (rc != 0)
  86                         return rc;
  87
  88                 rc = opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
  89                 if (rc != 0)
  90                         return rc;
  91
  92                 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
  93                         rc = opal_slw_set_reg(pir, P9_STOP_SPR_MSR, msr_val);
  94                         if (rc)
  95                                 return rc;
  96
  97                         rc = opal_slw_set_reg(pir,
  98                                               P9_STOP_SPR_PSSCR, psscr_val);
  99
 100                         if (rc)
 101                                 return rc;
 102                 }
 103
 104                 /* HIDs are per core registers */
 105                 if (cpu_thread_in_core(cpu) == 0) {
 106
 107                         rc = opal_slw_set_reg(pir, SPRN_HMEER, hmeer_val);
 108                         if (rc != 0)
 109                                 return rc;
 110
 111                         rc = opal_slw_set_reg(pir, SPRN_HID0, hid0_val);
 112                         if (rc != 0)
 113                                 return rc;
 114
 115                         /* Only p8 needs to set extra HID regiters */
 116                         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
 117
 118                                 rc = opal_slw_set_reg(pir, SPRN_HID1, hid1_val);
 119                                 if (rc != 0)
 120                                         return rc;
 121
 122                                 rc = opal_slw_set_reg(pir, SPRN_HID4, hid4_val);
 123                                 if (rc != 0)
 124                                         return rc;
 125
 126                                 rc = opal_slw_set_reg(pir, SPRN_HID5, hid5_val);
 127                                 if (rc != 0)
 128                                         return rc;
 129                         }
 130                 }
 131         }
 132
 133         return 0;
 134 }
 135
 136 static void pnv_alloc_idle_core_states(void)
 137 {
 138         int i, j;
 139         int nr_cores = cpu_nr_cores();
 140         u32 *core_idle_state;
 141
 142         /*
 143          * core_idle_state - The lower 8 bits track the idle state of
 144          * each thread of the core.
 145          *
 146          * The most significant bit is the lock bit.
 147          *
 148          * Initially all the bits corresponding to threads_per_core
 149          * are set. They are cleared when the thread enters deep idle
 150          * state like sleep and winkle/stop.
 151          *
 152          * Initially the lock bit is cleared.  The lock bit has 2
 153          * purposes:
 154          *      a. While the first thread in the core waking up from
 155          *         idle is restoring core state, it prevents other
 156          *         threads in the core from switching to process
 157          *         context.
 158          *      b. While the last thread in the core is saving the
 159          *         core state, it prevents a different thread from
 160          *         waking up.
 161          */
 162         for (i = 0; i < nr_cores; i++) {
 163                 int first_cpu = i * threads_per_core;
 164                 int node = cpu_to_node(first_cpu);
 165                 size_t paca_ptr_array_size;
 166
 167                 core_idle_state = kmalloc_node(sizeof(u32), GFP_KERNEL, node);
 168                 *core_idle_state = (1 << threads_per_core) - 1;
 169                 paca_ptr_array_size = (threads_per_core *
 170                                        sizeof(struct paca_struct *));
 171
 172                 for (j = 0; j < threads_per_core; j++) {
 173                         int cpu = first_cpu + j;
 174
 175                         paca[cpu].core_idle_state_ptr = core_idle_state;
 176                         paca[cpu].thread_idle_state = PNV_THREAD_RUNNING;
 177                         paca[cpu].thread_mask = 1 << j;
 178                         if (!cpu_has_feature(CPU_FTR_POWER9_DD1))
 179                                 continue;
 180                         paca[cpu].thread_sibling_pacas =
 181                                 kmalloc_node(paca_ptr_array_size,
 182                                              GFP_KERNEL, node);
 183                 }
 184         }
 185
 186         update_subcore_sibling_mask();
 187
 188         if (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT)
 189                 pnv_save_sprs_for_deep_states();
 190 }
 191
 192 u32 pnv_get_supported_cpuidle_states(void)
 193 {
 194         return supported_cpuidle_states;
 195 }
 196 EXPORT_SYMBOL_GPL(pnv_get_supported_cpuidle_states);
 197
 198 static void pnv_fastsleep_workaround_apply(void *info)
 199
 200 {
 201         int rc;
 202         int *err = info;
 203
 204         rc = opal_config_cpu_idle_state(OPAL_CONFIG_IDLE_FASTSLEEP,
 205                                         OPAL_CONFIG_IDLE_APPLY);
 206         if (rc)
 207                 *err = 1;
 208 }
 209
 210 /*
 211  * Used to store fastsleep workaround state
 212  * 0 - Workaround applied/undone at fastsleep entry/exit path (Default)
 213  * 1 - Workaround applied once, never undone.
 214  */
 215 static u8 fastsleep_workaround_applyonce;
 216
 217 static ssize_t show_fastsleep_workaround_applyonce(struct device *dev,
 218                 struct device_attribute *attr, char *buf)
 219 {
 220         return sprintf(buf, "%u\n", fastsleep_workaround_applyonce);
 221 }
 222
 223 static ssize_t store_fastsleep_workaround_applyonce(struct device *dev,
 224                 struct device_attribute *attr, const char *buf,
 225                 size_t count)
 226 {
 227         cpumask_t primary_thread_mask;
 228         int err;
 229         u8 val;
 230
 231         if (kstrtou8(buf, 0, &val) || val != 1)
 232                 return -EINVAL;
 233
 234         if (fastsleep_workaround_applyonce == 1)
 235                 return count;
 236
 237         /*
 238          * fastsleep_workaround_applyonce = 1 implies
 239          * fastsleep workaround needs to be left in 'applied' state on all
 240          * the cores. Do this by-
 241          * 1. Patching out the call to 'undo' workaround in fastsleep exit path
 242          * 2. Sending ipi to all the cores which have at least one online thread
 243          * 3. Patching out the call to 'apply' workaround in fastsleep entry
 244          * path
 245          * There is no need to send ipi to cores which have all threads
 246          * offlined, as last thread of the core entering fastsleep or deeper
 247          * state would have applied workaround.
 248          */
 249         err = patch_instruction(
 250                 (unsigned int *)pnv_fastsleep_workaround_at_exit,
 251                 PPC_INST_NOP);
 252         if (err) {
 253                 pr_err("fastsleep_workaround_applyonce change failed while patching pnv_fastsleep_workaround_at_exit");
 254                 goto fail;
 255         }
 256
 257         get_online_cpus();
 258         primary_thread_mask = cpu_online_cores_map();
 259         on_each_cpu_mask(&primary_thread_mask,
 260                                 pnv_fastsleep_workaround_apply,
 261                                 &err, 1);
 262         put_online_cpus();
 263         if (err) {
 264                 pr_err("fastsleep_workaround_applyonce change failed while running pnv_fastsleep_workaround_apply");
 265                 goto fail;
 266         }
 267
 268         err = patch_instruction(
 269                 (unsigned int *)pnv_fastsleep_workaround_at_entry,
 270                 PPC_INST_NOP);
 271         if (err) {
 272                 pr_err("fastsleep_workaround_applyonce change failed while patching pnv_fastsleep_workaround_at_entry");
 273                 goto fail;
 274         }
 275
 276         fastsleep_workaround_applyonce = 1;
 277
 278         return count;
 279 fail:
 280         return -EIO;
 281 }
 282
 283 static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600,
 284                         show_fastsleep_workaround_applyonce,
 285                         store_fastsleep_workaround_applyonce);
 286
 287 static unsigned long __power7_idle_type(unsigned long type)
 288 {
 289         unsigned long srr1;
 290
 291         if (!prep_irq_for_idle_irqsoff())
 292                 return 0;
 293
 294         __ppc64_runlatch_off();
 295         srr1 = power7_idle_insn(type);
 296         __ppc64_runlatch_on();
 297
 298         fini_irq_for_idle_irqsoff();
 299
 300         return srr1;
 301 }
 302
 303 void power7_idle_type(unsigned long type)
 304 {
 305         unsigned long srr1;
 306
 307         srr1 = __power7_idle_type(type);
 308         irq_set_pending_from_srr1(srr1);
 309 }
 310
 311 void power7_idle(void)
 312 {
 313         if (!powersave_nap)
 314                 return;
 315
 316         power7_idle_type(PNV_THREAD_NAP);
 317 }
 318
 319 static unsigned long __power9_idle_type(unsigned long stop_psscr_val,
 320                                       unsigned long stop_psscr_mask)
 321 {
 322         unsigned long psscr;
 323         unsigned long srr1;
 324
 325         if (!prep_irq_for_idle_irqsoff())
 326                 return 0;
 327
 328         psscr = mfspr(SPRN_PSSCR);
 329         psscr = (psscr & ~stop_psscr_mask) | stop_psscr_val;
 330
 331         __ppc64_runlatch_off();
 332         srr1 = power9_idle_stop(psscr);
 333         __ppc64_runlatch_on();
 334
 335         fini_irq_for_idle_irqsoff();
 336
 337         return srr1;
 338 }
 339
 340 void power9_idle_type(unsigned long stop_psscr_val,
 341                                       unsigned long stop_psscr_mask)
 342 {
 343         unsigned long srr1;
 344
 345         srr1 = __power9_idle_type(stop_psscr_val, stop_psscr_mask);
 346         irq_set_pending_from_srr1(srr1);
 347 }
 348
 349 /*
 350  * Used for ppc_md.power_save which needs a function with no parameters
 351  */
 352 void power9_idle(void)
 353 {
 354         power9_idle_type(pnv_default_stop_val, pnv_default_stop_mask);
 355 }
 356
 357 #ifdef CONFIG_HOTPLUG_CPU
 358 /*
 359  * pnv_cpu_offline: A function that puts the CPU into the deepest
 360  * available platform idle state on a CPU-Offline.
 361  * interrupts hard disabled and no lazy irq pending.
 362  */
 363 unsigned long pnv_cpu_offline(unsigned int cpu)
 364 {
 365         unsigned long srr1;
 366         u32 idle_states = pnv_get_supported_cpuidle_states();
 367
 368         __ppc64_runlatch_off();
 369
 370         if (cpu_has_feature(CPU_FTR_ARCH_300) && deepest_stop_found) {
 371                 unsigned long psscr;
 372
 373                 psscr = mfspr(SPRN_PSSCR);
 374                 psscr = (psscr & ~pnv_deepest_stop_psscr_mask) |
 375                                                 pnv_deepest_stop_psscr_val;
 376                 srr1 = power9_idle_stop(psscr);
 377
 378         } else if (idle_states & OPAL_PM_WINKLE_ENABLED) {
 379                 srr1 = power7_idle_insn(PNV_THREAD_WINKLE);
 380         } else if ((idle_states & OPAL_PM_SLEEP_ENABLED) ||
 381                    (idle_states & OPAL_PM_SLEEP_ENABLED_ER1)) {
 382                 srr1 = power7_idle_insn(PNV_THREAD_SLEEP);
 383         } else if (idle_states & OPAL_PM_NAP_ENABLED) {
 384                 srr1 = power7_idle_insn(PNV_THREAD_NAP);
 385         } else {
 386                 /* This is the fallback method. We emulate snooze */
 387                 while (!generic_check_cpu_restart(cpu)) {
 388                         HMT_low();
 389                         HMT_very_low();
 390                 }
 391                 srr1 = 0;
 392                 HMT_medium();
 393         }
 394
 395         __ppc64_runlatch_on();
 396
 397         return srr1;
 398 }
 399 #endif
 400
 401 /*
 402  * Power ISA 3.0 idle initialization.
 403  *
 404  * POWER ISA 3.0 defines a new SPR Processor stop Status and Control
 405  * Register (PSSCR) to control idle behavior.
 406  *
 407  * PSSCR layout:
 408  * ----------------------------------------------------------
 409  * | PLS | /// | SD | ESL | EC | PSLL | /// | TR | MTL | RL |
 410  * ----------------------------------------------------------
 411  * 0      4     41   42    43   44     48    54   56    60
 412  *
 413  * PSSCR key fields:
 414  *      Bits 0:3  - Power-Saving Level Status (PLS). This field indicates the
 415  *      lowest power-saving state the thread entered since stop instruction was
 416  *      last executed.
 417  *
 418  *      Bit 41 - Status Disable(SD)
 419  *      0 - Shows PLS entries
 420  *      1 - PLS entries are all 0
 421  *
 422  *      Bit 42 - Enable State Loss
 423  *      0 - No state is lost irrespective of other fields
 424  *      1 - Allows state loss
 425  *
 426  *      Bit 43 - Exit Criterion
 427  *      0 - Exit from power-save mode on any interrupt
 428  *      1 - Exit from power-save mode controlled by LPCR's PECE bits
 429  *
 430  *      Bits 44:47 - Power-Saving Level Limit
 431  *      This limits the power-saving level that can be entered into.
 432  *
 433  *      Bits 60:63 - Requested Level
 434  *      Used to specify which power-saving level must be entered on executing
 435  *      stop instruction
 436  */
 437
 438 int validate_psscr_val_mask(u64 *psscr_val, u64 *psscr_mask, u32 flags)
 439 {
 440         int err = 0;
 441
 442         /*
 443          * psscr_mask == 0xf indicates an older firmware.
 444          * Set remaining fields of psscr to the default values.
 445          * See NOTE above definition of PSSCR_HV_DEFAULT_VAL
 446          */
 447         if (*psscr_mask == 0xf) {
 448                 *psscr_val = *psscr_val | PSSCR_HV_DEFAULT_VAL;
 449                 *psscr_mask = PSSCR_HV_DEFAULT_MASK;
 450                 return err;
 451         }
 452
 453         /*
 454          * New firmware is expected to set the psscr_val bits correctly.
 455          * Validate that the following invariants are correctly maintained by
 456          * the new firmware.
 457          * - ESL bit value matches the EC bit value.
 458          * - ESL bit is set for all the deep stop states.
 459          */
 460         if (GET_PSSCR_ESL(*psscr_val) != GET_PSSCR_EC(*psscr_val)) {
 461                 err = ERR_EC_ESL_MISMATCH;
 462         } else if ((flags & OPAL_PM_LOSE_FULL_CONTEXT) &&
 463                 GET_PSSCR_ESL(*psscr_val) == 0) {
 464                 err = ERR_DEEP_STATE_ESL_MISMATCH;
 465         }
 466
 467         return err;
 468 }
 469
 470 /*
 471  * pnv_arch300_idle_init: Initializes the default idle state, first
 472  *                        deep idle state and deepest idle state on
 473  *                        ISA 3.0 CPUs.
 474  *
 475  * @np: /ibm,opal/power-mgt device node
 476  * @flags: cpu-idle-state-flags array
 477  * @dt_idle_states: Number of idle state entries
 478  * Returns 0 on success
 479  */
 480 static int __init pnv_power9_idle_init(struct device_node *np, u32 *flags,
 481                                         int dt_idle_states)
 482 {
 483         u64 *psscr_val = NULL;
 484         u64 *psscr_mask = NULL;
 485         u32 *residency_ns = NULL;
 486         u64 max_residency_ns = 0;
 487         int rc = 0, i;
 488
 489         psscr_val = kcalloc(dt_idle_states, sizeof(*psscr_val), GFP_KERNEL);
 490         psscr_mask = kcalloc(dt_idle_states, sizeof(*psscr_mask), GFP_KERNEL);
 491         residency_ns = kcalloc(dt_idle_states, sizeof(*residency_ns),
 492                                GFP_KERNEL);
 493
 494         if (!psscr_val || !psscr_mask || !residency_ns) {
 495                 rc = -1;
 496                 goto out;
 497         }
 498
 499         if (of_property_read_u64_array(np,
 500                 "ibm,cpu-idle-state-psscr",
 501                 psscr_val, dt_idle_states)) {
 502                 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n");
 503                 rc = -1;
 504                 goto out;
 505         }
 506
 507         if (of_property_read_u64_array(np,
 508                                        "ibm,cpu-idle-state-psscr-mask",
 509                                        psscr_mask, dt_idle_states)) {
 510                 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr-mask in DT\n");
 511                 rc = -1;
 512                 goto out;
 513         }
 514
 515         if (of_property_read_u32_array(np,
 516                                        "ibm,cpu-idle-state-residency-ns",
 517                                         residency_ns, dt_idle_states)) {
 518                 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-residency-ns in DT\n");
 519                 rc = -1;
 520                 goto out;
 521         }
 522
 523         /*
 524          * Set pnv_first_deep_stop_state, pnv_deepest_stop_psscr_{val,mask},
 525          * and the pnv_default_stop_{val,mask}.
 526          *
 527          * pnv_first_deep_stop_state should be set to the first stop
 528          * level to cause hypervisor state loss.
 529          *
 530          * pnv_deepest_stop_{val,mask} should be set to values corresponding to
 531          * the deepest stop state.
 532          *
 533          * pnv_default_stop_{val,mask} should be set to values corresponding to
 534          * the shallowest (OPAL_PM_STOP_INST_FAST) loss-less stop state.
 535          */
 536         pnv_first_deep_stop_state = MAX_STOP_STATE;
 537         for (i = 0; i < dt_idle_states; i++) {
 538                 int err;
 539                 u64 psscr_rl = psscr_val[i] & PSSCR_RL_MASK;
 540
 541                 if ((flags[i] & OPAL_PM_LOSE_FULL_CONTEXT) &&
 542                      (pnv_first_deep_stop_state > psscr_rl))
 543                         pnv_first_deep_stop_state = psscr_rl;
 544
 545                 err = validate_psscr_val_mask(&psscr_val[i], &psscr_mask[i],
 546                                               flags[i]);
 547                 if (err) {
 548                         report_invalid_psscr_val(psscr_val[i], err);
 549                         continue;
 550                 }
 551
 552                 if (max_residency_ns < residency_ns[i]) {
 553                         max_residency_ns = residency_ns[i];
 554                         pnv_deepest_stop_psscr_val = psscr_val[i];
 555                         pnv_deepest_stop_psscr_mask = psscr_mask[i];
 556                         deepest_stop_found = true;
 557                 }
 558
 559                 if (!default_stop_found &&
 560                     (flags[i] & OPAL_PM_STOP_INST_FAST)) {
 561                         pnv_default_stop_val = psscr_val[i];
 562                         pnv_default_stop_mask = psscr_mask[i];
 563                         default_stop_found = true;
 564                 }
 565         }
 566
 567         if (unlikely(!default_stop_found)) {
 568                 pr_warn("cpuidle-powernv: No suitable default stop state found. Disabling platform idle.\n");
 569         } else {
 570                 ppc_md.power_save = power9_idle;
 571                 pr_info("cpuidle-powernv: Default stop: psscr = 0x%016llx,mask=0x%016llx\n",
 572                         pnv_default_stop_val, pnv_default_stop_mask);
 573         }
 574
 575         if (unlikely(!deepest_stop_found)) {
 576                 pr_warn("cpuidle-powernv: No suitable stop state for CPU-Hotplug. Offlined CPUs will busy wait");
 577         } else {
 578                 pr_info("cpuidle-powernv: Deepest stop: psscr = 0x%016llx,mask=0x%016llx\n",
 579                         pnv_deepest_stop_psscr_val,
 580                         pnv_deepest_stop_psscr_mask);
 581         }
 582
 583         pr_info("cpuidle-powernv: Requested Level (RL) value of first deep stop = 0x%llx\n",
 584                 pnv_first_deep_stop_state);
 585 out:
 586         kfree(psscr_val);
 587         kfree(psscr_mask);
 588         kfree(residency_ns);
 589         return rc;
 590 }
 591
 592 /*
 593  * Probe device tree for supported idle states
 594  */
 595 static void __init pnv_probe_idle_states(void)
 596 {
 597         struct device_node *np;
 598         int dt_idle_states;
 599         u32 *flags = NULL;
 600         int i;
 601
 602         np = of_find_node_by_path("/ibm,opal/power-mgt");
 603         if (!np) {
 604                 pr_warn("opal: PowerMgmt Node not found\n");
 605                 goto out;
 606         }
 607         dt_idle_states = of_property_count_u32_elems(np,
 608                         "ibm,cpu-idle-state-flags");
 609         if (dt_idle_states < 0) {
 610                 pr_warn("cpuidle-powernv: no idle states found in the DT\n");
 611                 goto out;
 612         }
 613
 614         flags = kcalloc(dt_idle_states, sizeof(*flags),  GFP_KERNEL);
 615
 616         if (of_property_read_u32_array(np,
 617                         "ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
 618                 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-flags in DT\n");
 619                 goto out;
 620         }
 621
 622         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 623                 if (pnv_power9_idle_init(np, flags, dt_idle_states))
 624                         goto out;
 625         }
 626
 627         for (i = 0; i < dt_idle_states; i++)
 628                 supported_cpuidle_states |= flags[i];
 629
 630 out:
 631         kfree(flags);
 632 }
 633 static int __init pnv_init_idle_states(void)
 634 {
 635
 636         supported_cpuidle_states = 0;
 637
 638         if (cpuidle_disable != IDLE_NO_OVERRIDE)
 639                 goto out;
 640
 641         pnv_probe_idle_states();
 642
 643         if (!(supported_cpuidle_states & OPAL_PM_SLEEP_ENABLED_ER1)) {
 644                 patch_instruction(
 645                         (unsigned int *)pnv_fastsleep_workaround_at_entry,
 646                         PPC_INST_NOP);
 647                 patch_instruction(
 648                         (unsigned int *)pnv_fastsleep_workaround_at_exit,
 649                         PPC_INST_NOP);
 650         } else {
 651                 /*
 652                  * OPAL_PM_SLEEP_ENABLED_ER1 is set. It indicates that
 653                  * workaround is needed to use fastsleep. Provide sysfs
 654                  * control to choose how this workaround has to be applied.
 655                  */
 656                 device_create_file(cpu_subsys.dev_root,
 657                                 &dev_attr_fastsleep_workaround_applyonce);
 658         }
 659
 660         pnv_alloc_idle_core_states();
 661
 662         /*
 663          * For each CPU, record its PACA address in each of it's
 664          * sibling thread's PACA at the slot corresponding to this
 665          * CPU's index in the core.
 666          */
 667         if (cpu_has_feature(CPU_FTR_POWER9_DD1)) {
 668                 int cpu;
 669
 670                 pr_info("powernv: idle: Saving PACA pointers of all CPUs in their thread sibling PACA\n");
 671                 for_each_possible_cpu(cpu) {
 672                         int base_cpu = cpu_first_thread_sibling(cpu);
 673                         int idx = cpu_thread_in_core(cpu);
 674                         int i;
 675
 676                         for (i = 0; i < threads_per_core; i++) {
 677                                 int j = base_cpu + i;
 678
 679                                 paca[j].thread_sibling_pacas[idx] = &paca[cpu];
 680                         }
 681                 }
 682         }
 683
 684         if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED)
 685                 ppc_md.power_save = power7_idle;
 686
 687 out:
 688         return 0;
 689 }
 690 machine_subsys_initcall(powernv, pnv_init_idle_states);