platforms/pseries/lpar.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * pSeries_lpar.c
   4  * Copyright (C) 2001 Todd Inglett, IBM Corporation
   5  *
   6  * pSeries LPAR support.
   7  */
   8
   9 /* Enables debugging of low-level hash table routines - careful! */
  10 #undef DEBUG
  11 #define pr_fmt(fmt) "lpar: " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/dma-mapping.h>
  15 #include <linux/console.h>
  16 #include <linux/export.h>
  17 #include <linux/jump_label.h>
  18 #include <linux/delay.h>
  19 #include <linux/stop_machine.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/cpuhotplug.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/proc_fs.h>
  24 #include <linux/pgtable.h>
  25 #include <linux/debugfs.h>
  26
  27 #include <asm/processor.h>
  28 #include <asm/mmu.h>
  29 #include <asm/page.h>
  30 #include <asm/setup.h>
  31 #include <asm/mmu_context.h>
  32 #include <asm/iommu.h>
  33 #include <asm/tlb.h>
  34 #include <asm/cputable.h>
  35 #include <asm/papr-sysparm.h>
  36 #include <asm/udbg.h>
  37 #include <asm/smp.h>
  38 #include <asm/trace.h>
  39 #include <asm/firmware.h>
  40 #include <asm/plpar_wrappers.h>
  41 #include <asm/kexec.h>
  42 #include <asm/fadump.h>
  43 #include <asm/dtl.h>
  44
  45 #include "pseries.h"
  46
  47 /* Flag bits for H_BULK_REMOVE */
  48 #define HBR_REQUEST     0x4000000000000000UL
  49 #define HBR_RESPONSE    0x8000000000000000UL
  50 #define HBR_END         0xc000000000000000UL
  51 #define HBR_AVPN        0x0200000000000000UL
  52 #define HBR_ANDCOND     0x0100000000000000UL
  53
  54
  55 /* in hvCall.S */
  56 EXPORT_SYMBOL(plpar_hcall);
  57 EXPORT_SYMBOL(plpar_hcall9);
  58 EXPORT_SYMBOL(plpar_hcall_norets);
  59
  60 #ifdef CONFIG_PPC_64S_HASH_MMU
  61 /*
  62  * H_BLOCK_REMOVE supported block size for this page size in segment who's base
  63  * page size is that page size.
  64  *
  65  * The first index is the segment base page size, the second one is the actual
  66  * page size.
  67  */
  68 static int hblkrm_size[MMU_PAGE_COUNT][MMU_PAGE_COUNT] __ro_after_init;
  69 #endif
  70
  71 /*
  72  * Due to the involved complexity, and that the current hypervisor is only
  73  * returning this value or 0, we are limiting the support of the H_BLOCK_REMOVE
  74  * buffer size to 8 size block.
  75  */
  76 #define HBLKRM_SUPPORTED_BLOCK_SIZE 8
  77
  78 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
  79 static u8 dtl_mask = DTL_LOG_PREEMPT;
  80 #else
  81 static u8 dtl_mask;
  82 #endif
  83
  84 void alloc_dtl_buffers(unsigned long *time_limit)
  85 {
  86         int cpu;
  87         struct paca_struct *pp;
  88         struct dtl_entry *dtl;
  89
  90         for_each_possible_cpu(cpu) {
  91                 pp = paca_ptrs[cpu];
  92                 if (pp->dispatch_log)
  93                         continue;
  94                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
  95                 if (!dtl) {
  96                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
  97                                 cpu);
  98 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
  99                         pr_warn("Stolen time statistics will be unreliable\n");
 100 #endif
 101                         break;
 102                 }
 103
 104                 pp->dtl_ridx = 0;
 105                 pp->dispatch_log = dtl;
 106                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
 107                 pp->dtl_curr = dtl;
 108
 109                 if (time_limit && time_after(jiffies, *time_limit)) {
 110                         cond_resched();
 111                         *time_limit = jiffies + HZ;
 112                 }
 113         }
 114 }
 115
 116 void register_dtl_buffer(int cpu)
 117 {
 118         long ret;
 119         struct paca_struct *pp;
 120         struct dtl_entry *dtl;
 121         int hwcpu = get_hard_smp_processor_id(cpu);
 122
 123         pp = paca_ptrs[cpu];
 124         dtl = pp->dispatch_log;
 125         if (dtl && dtl_mask) {
 126                 pp->dtl_ridx = 0;
 127                 pp->dtl_curr = dtl;
 128                 lppaca_of(cpu).dtl_idx = 0;
 129
 130                 /* hypervisor reads buffer length from this field */
 131                 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 132                 ret = register_dtl(hwcpu, __pa(dtl));
 133                 if (ret)
 134                         pr_err("WARNING: DTL registration of cpu %d (hw %d) failed with %ld\n",
 135                                cpu, hwcpu, ret);
 136
 137                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 138         }
 139 }
 140
 141 #ifdef CONFIG_PPC_SPLPAR
 142 struct dtl_worker {
 143         struct delayed_work work;
 144         int cpu;
 145 };
 146
 147 struct vcpu_dispatch_data {
 148         int last_disp_cpu;
 149
 150         int total_disp;
 151
 152         int same_cpu_disp;
 153         int same_chip_disp;
 154         int diff_chip_disp;
 155         int far_chip_disp;
 156
 157         int numa_home_disp;
 158         int numa_remote_disp;
 159         int numa_far_disp;
 160 };
 161
 162 /*
 163  * This represents the number of cpus in the hypervisor. Since there is no
 164  * architected way to discover the number of processors in the host, we
 165  * provision for dealing with NR_CPUS. This is currently 2048 by default, and
 166  * is sufficient for our purposes. This will need to be tweaked if
 167  * CONFIG_NR_CPUS is changed.
 168  */
 169 #define NR_CPUS_H       NR_CPUS
 170
 171 DEFINE_RWLOCK(dtl_access_lock);
 172 static DEFINE_PER_CPU(struct vcpu_dispatch_data, vcpu_disp_data);
 173 static DEFINE_PER_CPU(u64, dtl_entry_ridx);
 174 static DEFINE_PER_CPU(struct dtl_worker, dtl_workers);
 175 static enum cpuhp_state dtl_worker_state;
 176 static DEFINE_MUTEX(dtl_enable_mutex);
 177 static int vcpudispatch_stats_on __read_mostly;
 178 static int vcpudispatch_stats_freq = 50;
 179 static __be32 *vcpu_associativity, *pcpu_associativity;
 180
 181
 182 static void free_dtl_buffers(unsigned long *time_limit)
 183 {
 184 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 185         int cpu;
 186         struct paca_struct *pp;
 187
 188         for_each_possible_cpu(cpu) {
 189                 pp = paca_ptrs[cpu];
 190                 if (!pp->dispatch_log)
 191                         continue;
 192                 kmem_cache_free(dtl_cache, pp->dispatch_log);
 193                 pp->dtl_ridx = 0;
 194                 pp->dispatch_log = 0;
 195                 pp->dispatch_log_end = 0;
 196                 pp->dtl_curr = 0;
 197
 198                 if (time_limit && time_after(jiffies, *time_limit)) {
 199                         cond_resched();
 200                         *time_limit = jiffies + HZ;
 201                 }
 202         }
 203 #endif
 204 }
 205
 206 static int init_cpu_associativity(void)
 207 {
 208         vcpu_associativity = kcalloc(num_possible_cpus() / threads_per_core,
 209                         VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
 210         pcpu_associativity = kcalloc(NR_CPUS_H / threads_per_core,
 211                         VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
 212
 213         if (!vcpu_associativity || !pcpu_associativity) {
 214                 pr_err("error allocating memory for associativity information\n");
 215                 return -ENOMEM;
 216         }
 217
 218         return 0;
 219 }
 220
 221 static void destroy_cpu_associativity(void)
 222 {
 223         kfree(vcpu_associativity);
 224         kfree(pcpu_associativity);
 225         vcpu_associativity = pcpu_associativity = 0;
 226 }
 227
 228 static __be32 *__get_cpu_associativity(int cpu, __be32 *cpu_assoc, int flag)
 229 {
 230         __be32 *assoc;
 231         int rc = 0;
 232
 233         assoc = &cpu_assoc[(int)(cpu / threads_per_core) * VPHN_ASSOC_BUFSIZE];
 234         if (!assoc[0]) {
 235                 rc = hcall_vphn(cpu, flag, &assoc[0]);
 236                 if (rc)
 237                         return NULL;
 238         }
 239
 240         return assoc;
 241 }
 242
 243 static __be32 *get_pcpu_associativity(int cpu)
 244 {
 245         return __get_cpu_associativity(cpu, pcpu_associativity, VPHN_FLAG_PCPU);
 246 }
 247
 248 static __be32 *get_vcpu_associativity(int cpu)
 249 {
 250         return __get_cpu_associativity(cpu, vcpu_associativity, VPHN_FLAG_VCPU);
 251 }
 252
 253 static int cpu_relative_dispatch_distance(int last_disp_cpu, int cur_disp_cpu)
 254 {
 255         __be32 *last_disp_cpu_assoc, *cur_disp_cpu_assoc;
 256
 257         if (last_disp_cpu >= NR_CPUS_H || cur_disp_cpu >= NR_CPUS_H)
 258                 return -EINVAL;
 259
 260         last_disp_cpu_assoc = get_pcpu_associativity(last_disp_cpu);
 261         cur_disp_cpu_assoc = get_pcpu_associativity(cur_disp_cpu);
 262
 263         if (!last_disp_cpu_assoc || !cur_disp_cpu_assoc)
 264                 return -EIO;
 265
 266         return cpu_relative_distance(last_disp_cpu_assoc, cur_disp_cpu_assoc);
 267 }
 268
 269 static int cpu_home_node_dispatch_distance(int disp_cpu)
 270 {
 271         __be32 *disp_cpu_assoc, *vcpu_assoc;
 272         int vcpu_id = smp_processor_id();
 273
 274         if (disp_cpu >= NR_CPUS_H) {
 275                 pr_debug_ratelimited("vcpu dispatch cpu %d > %d\n",
 276                                                 disp_cpu, NR_CPUS_H);
 277                 return -EINVAL;
 278         }
 279
 280         disp_cpu_assoc = get_pcpu_associativity(disp_cpu);
 281         vcpu_assoc = get_vcpu_associativity(vcpu_id);
 282
 283         if (!disp_cpu_assoc || !vcpu_assoc)
 284                 return -EIO;
 285
 286         return cpu_relative_distance(disp_cpu_assoc, vcpu_assoc);
 287 }
 288
 289 static void update_vcpu_disp_stat(int disp_cpu)
 290 {
 291         struct vcpu_dispatch_data *disp;
 292         int distance;
 293
 294         disp = this_cpu_ptr(&vcpu_disp_data);
 295         if (disp->last_disp_cpu == -1) {
 296                 disp->last_disp_cpu = disp_cpu;
 297                 return;
 298         }
 299
 300         disp->total_disp++;
 301
 302         if (disp->last_disp_cpu == disp_cpu ||
 303                 (cpu_first_thread_sibling(disp->last_disp_cpu) ==
 304                                         cpu_first_thread_sibling(disp_cpu)))
 305                 disp->same_cpu_disp++;
 306         else {
 307                 distance = cpu_relative_dispatch_distance(disp->last_disp_cpu,
 308                                                                 disp_cpu);
 309                 if (distance < 0)
 310                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
 311                                         smp_processor_id());
 312                 else {
 313                         switch (distance) {
 314                         case 0:
 315                                 disp->same_chip_disp++;
 316                                 break;
 317                         case 1:
 318                                 disp->diff_chip_disp++;
 319                                 break;
 320                         case 2:
 321                                 disp->far_chip_disp++;
 322                                 break;
 323                         default:
 324                                 pr_debug_ratelimited("vcpudispatch_stats: cpu %d (%d -> %d): unexpected relative dispatch distance %d\n",
 325                                                  smp_processor_id(),
 326                                                  disp->last_disp_cpu,
 327                                                  disp_cpu,
 328                                                  distance);
 329                         }
 330                 }
 331         }
 332
 333         distance = cpu_home_node_dispatch_distance(disp_cpu);
 334         if (distance < 0)
 335                 pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
 336                                 smp_processor_id());
 337         else {
 338                 switch (distance) {
 339                 case 0:
 340                         disp->numa_home_disp++;
 341                         break;
 342                 case 1:
 343                         disp->numa_remote_disp++;
 344                         break;
 345                 case 2:
 346                         disp->numa_far_disp++;
 347                         break;
 348                 default:
 349                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d on %d: unexpected numa dispatch distance %d\n",
 350                                                  smp_processor_id(),
 351                                                  disp_cpu,
 352                                                  distance);
 353                 }
 354         }
 355
 356         disp->last_disp_cpu = disp_cpu;
 357 }
 358
 359 static void process_dtl_buffer(struct work_struct *work)
 360 {
 361         struct dtl_entry dtle;
 362         u64 i = __this_cpu_read(dtl_entry_ridx);
 363         struct dtl_entry *dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
 364         struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
 365         struct lppaca *vpa = local_paca->lppaca_ptr;
 366         struct dtl_worker *d = container_of(work, struct dtl_worker, work.work);
 367
 368         if (!local_paca->dispatch_log)
 369                 return;
 370
 371         /* if we have been migrated away, we cancel ourself */
 372         if (d->cpu != smp_processor_id()) {
 373                 pr_debug("vcpudispatch_stats: cpu %d worker migrated -- canceling worker\n",
 374                                                 smp_processor_id());
 375                 return;
 376         }
 377
 378         if (i == be64_to_cpu(vpa->dtl_idx))
 379                 goto out;
 380
 381         while (i < be64_to_cpu(vpa->dtl_idx)) {
 382                 dtle = *dtl;
 383                 barrier();
 384                 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
 385                         /* buffer has overflowed */
 386                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d lost %lld DTL samples\n",
 387                                 d->cpu,
 388                                 be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG - i);
 389                         i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
 390                         dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
 391                         continue;
 392                 }
 393                 update_vcpu_disp_stat(be16_to_cpu(dtle.processor_id));
 394                 ++i;
 395                 ++dtl;
 396                 if (dtl == dtl_end)
 397                         dtl = local_paca->dispatch_log;
 398         }
 399
 400         __this_cpu_write(dtl_entry_ridx, i);
 401
 402 out:
 403         schedule_delayed_work_on(d->cpu, to_delayed_work(work),
 404                                         HZ / vcpudispatch_stats_freq);
 405 }
 406
 407 static int dtl_worker_online(unsigned int cpu)
 408 {
 409         struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
 410
 411         memset(d, 0, sizeof(*d));
 412         INIT_DELAYED_WORK(&d->work, process_dtl_buffer);
 413         d->cpu = cpu;
 414
 415 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 416         per_cpu(dtl_entry_ridx, cpu) = 0;
 417         register_dtl_buffer(cpu);
 418 #else
 419         per_cpu(dtl_entry_ridx, cpu) = be64_to_cpu(lppaca_of(cpu).dtl_idx);
 420 #endif
 421
 422         schedule_delayed_work_on(cpu, &d->work, HZ / vcpudispatch_stats_freq);
 423         return 0;
 424 }
 425
 426 static int dtl_worker_offline(unsigned int cpu)
 427 {
 428         struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
 429
 430         cancel_delayed_work_sync(&d->work);
 431
 432 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 433         unregister_dtl(get_hard_smp_processor_id(cpu));
 434 #endif
 435
 436         return 0;
 437 }
 438
 439 static void set_global_dtl_mask(u8 mask)
 440 {
 441         int cpu;
 442
 443         dtl_mask = mask;
 444         for_each_present_cpu(cpu)
 445                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 446 }
 447
 448 static void reset_global_dtl_mask(void)
 449 {
 450         int cpu;
 451
 452 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 453         dtl_mask = DTL_LOG_PREEMPT;
 454 #else
 455         dtl_mask = 0;
 456 #endif
 457         for_each_present_cpu(cpu)
 458                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 459 }
 460
 461 static int dtl_worker_enable(unsigned long *time_limit)
 462 {
 463         int rc = 0, state;
 464
 465         if (!write_trylock(&dtl_access_lock)) {
 466                 rc = -EBUSY;
 467                 goto out;
 468         }
 469
 470         set_global_dtl_mask(DTL_LOG_ALL);
 471
 472         /* Setup dtl buffers and register those */
 473         alloc_dtl_buffers(time_limit);
 474
 475         state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powerpc/dtl:online",
 476                                         dtl_worker_online, dtl_worker_offline);
 477         if (state < 0) {
 478                 pr_err("vcpudispatch_stats: unable to setup workqueue for DTL processing\n");
 479                 free_dtl_buffers(time_limit);
 480                 reset_global_dtl_mask();
 481                 write_unlock(&dtl_access_lock);
 482                 rc = -EINVAL;
 483                 goto out;
 484         }
 485         dtl_worker_state = state;
 486
 487 out:
 488         return rc;
 489 }
 490
 491 static void dtl_worker_disable(unsigned long *time_limit)
 492 {
 493         cpuhp_remove_state(dtl_worker_state);
 494         free_dtl_buffers(time_limit);
 495         reset_global_dtl_mask();
 496         write_unlock(&dtl_access_lock);
 497 }
 498
 499 static ssize_t vcpudispatch_stats_write(struct file *file, const char __user *p,
 500                 size_t count, loff_t *ppos)
 501 {
 502         unsigned long time_limit = jiffies + HZ;
 503         struct vcpu_dispatch_data *disp;
 504         int rc, cmd, cpu;
 505         char buf[16];
 506
 507         if (count > 15)
 508                 return -EINVAL;
 509
 510         if (copy_from_user(buf, p, count))
 511                 return -EFAULT;
 512
 513         buf[count] = 0;
 514         rc = kstrtoint(buf, 0, &cmd);
 515         if (rc || cmd < 0 || cmd > 1) {
 516                 pr_err("vcpudispatch_stats: please use 0 to disable or 1 to enable dispatch statistics\n");
 517                 return rc ? rc : -EINVAL;
 518         }
 519
 520         mutex_lock(&dtl_enable_mutex);
 521
 522         if ((cmd == 0 && !vcpudispatch_stats_on) ||
 523                         (cmd == 1 && vcpudispatch_stats_on))
 524                 goto out;
 525
 526         if (cmd) {
 527                 rc = init_cpu_associativity();
 528                 if (rc)
 529                         goto out;
 530
 531                 for_each_possible_cpu(cpu) {
 532                         disp = per_cpu_ptr(&vcpu_disp_data, cpu);
 533                         memset(disp, 0, sizeof(*disp));
 534                         disp->last_disp_cpu = -1;
 535                 }
 536
 537                 rc = dtl_worker_enable(&time_limit);
 538                 if (rc) {
 539                         destroy_cpu_associativity();
 540                         goto out;
 541                 }
 542         } else {
 543                 dtl_worker_disable(&time_limit);
 544                 destroy_cpu_associativity();
 545         }
 546
 547         vcpudispatch_stats_on = cmd;
 548
 549 out:
 550         mutex_unlock(&dtl_enable_mutex);
 551         if (rc)
 552                 return rc;
 553         return count;
 554 }
 555
 556 static int vcpudispatch_stats_display(struct seq_file *p, void *v)
 557 {
 558         int cpu;
 559         struct vcpu_dispatch_data *disp;
 560
 561         if (!vcpudispatch_stats_on) {
 562                 seq_puts(p, "off\n");
 563                 return 0;
 564         }
 565
 566         for_each_online_cpu(cpu) {
 567                 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
 568                 seq_printf(p, "cpu%d", cpu);
 569                 seq_put_decimal_ull(p, " ", disp->total_disp);
 570                 seq_put_decimal_ull(p, " ", disp->same_cpu_disp);
 571                 seq_put_decimal_ull(p, " ", disp->same_chip_disp);
 572                 seq_put_decimal_ull(p, " ", disp->diff_chip_disp);
 573                 seq_put_decimal_ull(p, " ", disp->far_chip_disp);
 574                 seq_put_decimal_ull(p, " ", disp->numa_home_disp);
 575                 seq_put_decimal_ull(p, " ", disp->numa_remote_disp);
 576                 seq_put_decimal_ull(p, " ", disp->numa_far_disp);
 577                 seq_puts(p, "\n");
 578         }
 579
 580         return 0;
 581 }
 582
 583 static int vcpudispatch_stats_open(struct inode *inode, struct file *file)
 584 {
 585         return single_open(file, vcpudispatch_stats_display, NULL);
 586 }
 587
 588 static const struct proc_ops vcpudispatch_stats_proc_ops = {
 589         .proc_open      = vcpudispatch_stats_open,
 590         .proc_read      = seq_read,
 591         .proc_write     = vcpudispatch_stats_write,
 592         .proc_lseek     = seq_lseek,
 593         .proc_release   = single_release,
 594 };
 595
 596 static ssize_t vcpudispatch_stats_freq_write(struct file *file,
 597                 const char __user *p, size_t count, loff_t *ppos)
 598 {
 599         int rc, freq;
 600         char buf[16];
 601
 602         if (count > 15)
 603                 return -EINVAL;
 604
 605         if (copy_from_user(buf, p, count))
 606                 return -EFAULT;
 607
 608         buf[count] = 0;
 609         rc = kstrtoint(buf, 0, &freq);
 610         if (rc || freq < 1 || freq > HZ) {
 611                 pr_err("vcpudispatch_stats_freq: please specify a frequency between 1 and %d\n",
 612                                 HZ);
 613                 return rc ? rc : -EINVAL;
 614         }
 615
 616         vcpudispatch_stats_freq = freq;
 617
 618         return count;
 619 }
 620
 621 static int vcpudispatch_stats_freq_display(struct seq_file *p, void *v)
 622 {
 623         seq_printf(p, "%d\n", vcpudispatch_stats_freq);
 624         return 0;
 625 }
 626
 627 static int vcpudispatch_stats_freq_open(struct inode *inode, struct file *file)
 628 {
 629         return single_open(file, vcpudispatch_stats_freq_display, NULL);
 630 }
 631
 632 static const struct proc_ops vcpudispatch_stats_freq_proc_ops = {
 633         .proc_open      = vcpudispatch_stats_freq_open,
 634         .proc_read      = seq_read,
 635         .proc_write     = vcpudispatch_stats_freq_write,
 636         .proc_lseek     = seq_lseek,
 637         .proc_release   = single_release,
 638 };
 639
 640 static int __init vcpudispatch_stats_procfs_init(void)
 641 {
 642         if (!lppaca_shared_proc())
 643                 return 0;
 644
 645         if (!proc_create("powerpc/vcpudispatch_stats", 0600, NULL,
 646                                         &vcpudispatch_stats_proc_ops))
 647                 pr_err("vcpudispatch_stats: error creating procfs file\n");
 648         else if (!proc_create("powerpc/vcpudispatch_stats_freq", 0600, NULL,
 649                                         &vcpudispatch_stats_freq_proc_ops))
 650                 pr_err("vcpudispatch_stats_freq: error creating procfs file\n");
 651
 652         return 0;
 653 }
 654
 655 machine_device_initcall(pseries, vcpudispatch_stats_procfs_init);
 656
 657 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
 658 u64 pseries_paravirt_steal_clock(int cpu)
 659 {
 660         struct lppaca *lppaca = &lppaca_of(cpu);
 661
 662         return be64_to_cpu(READ_ONCE(lppaca->enqueue_dispatch_tb)) +
 663                 be64_to_cpu(READ_ONCE(lppaca->ready_enqueue_tb));
 664 }
 665 #endif
 666
 667 #endif /* CONFIG_PPC_SPLPAR */
 668
 669 void vpa_init(int cpu)
 670 {
 671         int hwcpu = get_hard_smp_processor_id(cpu);
 672         unsigned long addr;
 673         long ret;
 674
 675         /*
 676          * The spec says it "may be problematic" if CPU x registers the VPA of
 677          * CPU y. We should never do that, but wail if we ever do.
 678          */
 679         WARN_ON(cpu != smp_processor_id());
 680
 681         if (cpu_has_feature(CPU_FTR_ALTIVEC))
 682                 lppaca_of(cpu).vmxregs_in_use = 1;
 683
 684         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 685                 lppaca_of(cpu).ebb_regs_in_use = 1;
 686
 687         addr = __pa(&lppaca_of(cpu));
 688         ret = register_vpa(hwcpu, addr);
 689
 690         if (ret) {
 691                 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
 692                        "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
 693                 return;
 694         }
 695
 696 #ifdef CONFIG_PPC_64S_HASH_MMU
 697         /*
 698          * PAPR says this feature is SLB-Buffer but firmware never
 699          * reports that.  All SPLPAR support SLB shadow buffer.
 700          */
 701         if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
 702                 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
 703                 ret = register_slb_shadow(hwcpu, addr);
 704                 if (ret)
 705                         pr_err("WARNING: SLB shadow buffer registration for "
 706                                "cpu %d (hw %d) of area %lx failed with %ld\n",
 707                                cpu, hwcpu, addr, ret);
 708         }
 709 #endif /* CONFIG_PPC_64S_HASH_MMU */
 710
 711         /*
 712          * Register dispatch trace log, if one has been allocated.
 713          */
 714         register_dtl_buffer(cpu);
 715 }
 716
 717 #ifdef CONFIG_PPC_BOOK3S_64
 718
 719 static int __init pseries_lpar_register_process_table(unsigned long base,
 720                         unsigned long page_size, unsigned long table_size)
 721 {
 722         long rc;
 723         unsigned long flags = 0;
 724
 725         if (table_size)
 726                 flags |= PROC_TABLE_NEW;
 727         if (radix_enabled()) {
 728                 flags |= PROC_TABLE_RADIX;
 729                 if (mmu_has_feature(MMU_FTR_GTSE))
 730                         flags |= PROC_TABLE_GTSE;
 731         } else
 732                 flags |= PROC_TABLE_HPT_SLB;
 733         for (;;) {
 734                 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
 735                                         page_size, table_size);
 736                 if (!H_IS_LONG_BUSY(rc))
 737                         break;
 738                 mdelay(get_longbusy_msecs(rc));
 739         }
 740         if (rc != H_SUCCESS) {
 741                 pr_err("Failed to register process table (rc=%ld)\n", rc);
 742                 BUG();
 743         }
 744         return rc;
 745 }
 746
 747 #ifdef CONFIG_PPC_64S_HASH_MMU
 748
 749 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 750                                      unsigned long vpn, unsigned long pa,
 751                                      unsigned long rflags, unsigned long vflags,
 752                                      int psize, int apsize, int ssize)
 753 {
 754         unsigned long lpar_rc;
 755         unsigned long flags;
 756         unsigned long slot;
 757         unsigned long hpte_v, hpte_r;
 758
 759         if (!(vflags & HPTE_V_BOLTED))
 760                 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
 761                          "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
 762                          hpte_group, vpn,  pa, rflags, vflags, psize);
 763
 764         hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
 765         hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
 766
 767         if (!(vflags & HPTE_V_BOLTED))
 768                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 769
 770         /* Now fill in the actual HPTE */
 771         /* Set CEC cookie to 0         */
 772         /* Zero page = 0               */
 773         /* I-cache Invalidate = 0      */
 774         /* I-cache synchronize = 0     */
 775         /* Exact = 0                   */
 776         flags = 0;
 777
 778         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 779                 flags |= H_COALESCE_CAND;
 780
 781         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 782         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 783                 pr_devel("Hash table group is full\n");
 784                 return -1;
 785         }
 786
 787         /*
 788          * Since we try and ioremap PHBs we don't own, the pte insert
 789          * will fail. However we must catch the failure in hash_page
 790          * or we will loop forever, so return -2 in this case.
 791          */
 792         if (unlikely(lpar_rc != H_SUCCESS)) {
 793                 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
 794                 return -2;
 795         }
 796         if (!(vflags & HPTE_V_BOLTED))
 797                 pr_devel(" -> slot: %lu\n", slot & 7);
 798
 799         /* Because of iSeries, we have to pass down the secondary
 800          * bucket bit here as well
 801          */
 802         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 803 }
 804
 805 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 806
 807 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 808 {
 809         unsigned long slot_offset;
 810         unsigned long lpar_rc;
 811         int i;
 812         unsigned long dummy1, dummy2;
 813
 814         /* pick a random slot to start at */
 815         slot_offset = mftb() & 0x7;
 816
 817         for (i = 0; i < HPTES_PER_GROUP; i++) {
 818
 819                 /* don't remove a bolted entry */
 820                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 821                                            HPTE_V_BOLTED, &dummy1, &dummy2);
 822                 if (lpar_rc == H_SUCCESS)
 823                         return i;
 824
 825                 /*
 826                  * The test for adjunct partition is performed before the
 827                  * ANDCOND test.  H_RESOURCE may be returned, so we need to
 828                  * check for that as well.
 829                  */
 830                 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
 831
 832                 slot_offset++;
 833                 slot_offset &= 0x7;
 834         }
 835
 836         return -1;
 837 }
 838
 839 /* Called during kexec sequence with MMU off */
 840 static notrace void manual_hpte_clear_all(void)
 841 {
 842         unsigned long size_bytes = 1UL << ppc64_pft_size;
 843         unsigned long hpte_count = size_bytes >> 4;
 844         struct {
 845                 unsigned long pteh;
 846                 unsigned long ptel;
 847         } ptes[4];
 848         long lpar_rc;
 849         unsigned long i, j;
 850
 851         /* Read in batches of 4,
 852          * invalidate only valid entries not in the VRMA
 853          * hpte_count will be a multiple of 4
 854          */
 855         for (i = 0; i < hpte_count; i += 4) {
 856                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 857                 if (lpar_rc != H_SUCCESS) {
 858                         pr_info("Failed to read hash page table at %ld err %ld\n",
 859                                 i, lpar_rc);
 860                         continue;
 861                 }
 862                 for (j = 0; j < 4; j++){
 863                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 864                                 HPTE_V_VRMA_MASK)
 865                                 continue;
 866                         if (ptes[j].pteh & HPTE_V_VALID)
 867                                 plpar_pte_remove_raw(0, i + j, 0,
 868                                         &(ptes[j].pteh), &(ptes[j].ptel));
 869                 }
 870         }
 871 }
 872
 873 /* Called during kexec sequence with MMU off */
 874 static notrace int hcall_hpte_clear_all(void)
 875 {
 876         int rc;
 877
 878         do {
 879                 rc = plpar_hcall_norets(H_CLEAR_HPT);
 880         } while (rc == H_CONTINUE);
 881
 882         return rc;
 883 }
 884
 885 /* Called during kexec sequence with MMU off */
 886 static notrace void pseries_hpte_clear_all(void)
 887 {
 888         int rc;
 889
 890         rc = hcall_hpte_clear_all();
 891         if (rc != H_SUCCESS)
 892                 manual_hpte_clear_all();
 893
 894 #ifdef __LITTLE_ENDIAN__
 895         /*
 896          * Reset exceptions to big endian.
 897          *
 898          * FIXME this is a hack for kexec, we need to reset the exception
 899          * endian before starting the new kernel and this is a convenient place
 900          * to do it.
 901          *
 902          * This is also called on boot when a fadump happens. In that case we
 903          * must not change the exception endian mode.
 904          */
 905         if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
 906                 pseries_big_endian_exceptions();
 907 #endif
 908 }
 909
 910 /*
 911  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 912  * the low 3 bits of flags happen to line up.  So no transform is needed.
 913  * We can probably optimize here and assume the high bits of newpp are
 914  * already zero.  For now I am paranoid.
 915  */
 916 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 917                                        unsigned long newpp,
 918                                        unsigned long vpn,
 919                                        int psize, int apsize,
 920                                        int ssize, unsigned long inv_flags)
 921 {
 922         unsigned long lpar_rc;
 923         unsigned long flags;
 924         unsigned long want_v;
 925
 926         want_v = hpte_encode_avpn(vpn, psize, ssize);
 927
 928         flags = (newpp & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO)) | H_AVPN;
 929         flags |= (newpp & HPTE_R_KEY_HI) >> 48;
 930         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 931                 /* Move pp0 into bit 8 (IBM 55) */
 932                 flags |= (newpp & HPTE_R_PP0) >> 55;
 933
 934         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 935                  want_v, slot, flags, psize);
 936
 937         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 938
 939         if (lpar_rc == H_NOT_FOUND) {
 940                 pr_devel("not found !\n");
 941                 return -1;
 942         }
 943
 944         pr_devel("ok\n");
 945
 946         BUG_ON(lpar_rc != H_SUCCESS);
 947
 948         return 0;
 949 }
 950
 951 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
 952 {
 953         long lpar_rc;
 954         unsigned long i, j;
 955         struct {
 956                 unsigned long pteh;
 957                 unsigned long ptel;
 958         } ptes[4];
 959
 960         for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
 961
 962                 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
 963                 if (lpar_rc != H_SUCCESS) {
 964                         pr_info("Failed to read hash page table at %ld err %ld\n",
 965                                 hpte_group, lpar_rc);
 966                         continue;
 967                 }
 968
 969                 for (j = 0; j < 4; j++) {
 970                         if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
 971                             (ptes[j].pteh & HPTE_V_VALID))
 972                                 return i + j;
 973                 }
 974         }
 975
 976         return -1;
 977 }
 978
 979 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 980 {
 981         long slot;
 982         unsigned long hash;
 983         unsigned long want_v;
 984         unsigned long hpte_group;
 985
 986         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 987         want_v = hpte_encode_avpn(vpn, psize, ssize);
 988
 989         /*
 990          * We try to keep bolted entries always in primary hash
 991          * But in some case we can find them in secondary too.
 992          */
 993         hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 994         slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 995         if (slot < 0) {
 996                 /* Try in secondary */
 997                 hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
 998                 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 999                 if (slot < 0)
1000                         return -1;
1001         }
1002         return hpte_group + slot;
1003 }
1004
1005 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
1006                                              unsigned long ea,
1007                                              int psize, int ssize)
1008 {
1009         unsigned long vpn;
1010         unsigned long lpar_rc, slot, vsid, flags;
1011
1012         vsid = get_kernel_vsid(ea, ssize);
1013         vpn = hpt_vpn(ea, vsid, ssize);
1014
1015         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1016         BUG_ON(slot == -1);
1017
1018         flags = newpp & (HPTE_R_PP | HPTE_R_N);
1019         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
1020                 /* Move pp0 into bit 8 (IBM 55) */
1021                 flags |= (newpp & HPTE_R_PP0) >> 55;
1022
1023         flags |= ((newpp & HPTE_R_KEY_HI) >> 48) | (newpp & HPTE_R_KEY_LO);
1024
1025         lpar_rc = plpar_pte_protect(flags, slot, 0);
1026
1027         BUG_ON(lpar_rc != H_SUCCESS);
1028 }
1029
1030 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
1031                                          int psize, int apsize,
1032                                          int ssize, int local)
1033 {
1034         unsigned long want_v;
1035         unsigned long lpar_rc;
1036         unsigned long dummy1, dummy2;
1037
1038         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
1039                  slot, vpn, psize, local);
1040
1041         want_v = hpte_encode_avpn(vpn, psize, ssize);
1042         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
1043         if (lpar_rc == H_NOT_FOUND)
1044                 return;
1045
1046         BUG_ON(lpar_rc != H_SUCCESS);
1047 }
1048
1049
1050 /*
1051  * As defined in the PAPR's section 14.5.4.1.8
1052  * The control mask doesn't include the returned reference and change bit from
1053  * the processed PTE.
1054  */
1055 #define HBLKR_AVPN              0x0100000000000000UL
1056 #define HBLKR_CTRL_MASK         0xf800000000000000UL
1057 #define HBLKR_CTRL_SUCCESS      0x8000000000000000UL
1058 #define HBLKR_CTRL_ERRNOTFOUND  0x8800000000000000UL
1059 #define HBLKR_CTRL_ERRBUSY      0xa000000000000000UL
1060
1061 /*
1062  * Returned true if we are supporting this block size for the specified segment
1063  * base page size and actual page size.
1064  *
1065  * Currently, we only support 8 size block.
1066  */
1067 static inline bool is_supported_hlbkrm(int bpsize, int psize)
1068 {
1069         return (hblkrm_size[bpsize][psize] == HBLKRM_SUPPORTED_BLOCK_SIZE);
1070 }
1071
1072 /**
1073  * H_BLOCK_REMOVE caller.
1074  * @idx should point to the latest @param entry set with a PTEX.
1075  * If PTE cannot be processed because another CPUs has already locked that
1076  * group, those entries are put back in @param starting at index 1.
1077  * If entries has to be retried and @retry_busy is set to true, these entries
1078  * are retried until success. If @retry_busy is set to false, the returned
1079  * is the number of entries yet to process.
1080  */
1081 static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
1082                                        bool retry_busy)
1083 {
1084         unsigned long i, rc, new_idx;
1085         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1086
1087         if (idx < 2) {
1088                 pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
1089                 return 0;
1090         }
1091 again:
1092         new_idx = 0;
1093         if (idx > PLPAR_HCALL9_BUFSIZE) {
1094                 pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
1095                 idx = PLPAR_HCALL9_BUFSIZE;
1096         } else if (idx < PLPAR_HCALL9_BUFSIZE)
1097                 param[idx] = HBR_END;
1098
1099         rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
1100                           param[0], /* AVA */
1101                           param[1],  param[2],  param[3],  param[4], /* TS0-7 */
1102                           param[5],  param[6],  param[7],  param[8]);
1103         if (rc == H_SUCCESS)
1104                 return 0;
1105
1106         BUG_ON(rc != H_PARTIAL);
1107
1108         /* Check that the unprocessed entries were 'not found' or 'busy' */
1109         for (i = 0; i < idx-1; i++) {
1110                 unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
1111
1112                 if (ctrl == HBLKR_CTRL_ERRBUSY) {
1113                         param[++new_idx] = param[i+1];
1114                         continue;
1115                 }
1116
1117                 BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
1118                        && ctrl != HBLKR_CTRL_ERRNOTFOUND);
1119         }
1120
1121         /*
1122          * If there were entries found busy, retry these entries if requested,
1123          * of if all the entries have to be retried.
1124          */
1125         if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
1126                 idx = new_idx + 1;
1127                 goto again;
1128         }
1129
1130         return new_idx;
1131 }
1132
1133 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1134 /*
1135  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
1136  * to make sure that we avoid bouncing the hypervisor tlbie lock.
1137  */
1138 #define PPC64_HUGE_HPTE_BATCH 12
1139
1140 static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
1141                                       int count, int psize, int ssize)
1142 {
1143         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1144         unsigned long shift, current_vpgb, vpgb;
1145         int i, pix = 0;
1146
1147         shift = mmu_psize_defs[psize].shift;
1148
1149         for (i = 0; i < count; i++) {
1150                 /*
1151                  * Shifting 3 bits more on the right to get a
1152                  * 8 pages aligned virtual addresse.
1153                  */
1154                 vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
1155                 if (!pix || vpgb != current_vpgb) {
1156                         /*
1157                          * Need to start a new 8 pages block, flush
1158                          * the current one if needed.
1159                          */
1160                         if (pix)
1161                                 (void)call_block_remove(pix, param, true);
1162                         current_vpgb = vpgb;
1163                         param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
1164                         pix = 1;
1165                 }
1166
1167                 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
1168                 if (pix == PLPAR_HCALL9_BUFSIZE) {
1169                         pix = call_block_remove(pix, param, false);
1170                         /*
1171                          * pix = 0 means that all the entries were
1172                          * removed, we can start a new block.
1173                          * Otherwise, this means that there are entries
1174                          * to retry, and pix points to latest one, so
1175                          * we should increment it and try to continue
1176                          * the same block.
1177                          */
1178                         if (pix)
1179                                 pix++;
1180                 }
1181         }
1182         if (pix)
1183                 (void)call_block_remove(pix, param, true);
1184 }
1185
1186 static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
1187                                      int count, int psize, int ssize)
1188 {
1189         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1190         int i = 0, pix = 0, rc;
1191
1192         for (i = 0; i < count; i++) {
1193
1194                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1195                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
1196                                                      ssize, 0);
1197                 } else {
1198                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
1199                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
1200                         pix += 2;
1201                         if (pix == 8) {
1202                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
1203                                                   param[0], param[1], param[2],
1204                                                   param[3], param[4], param[5],
1205                                                   param[6], param[7]);
1206                                 BUG_ON(rc != H_SUCCESS);
1207                                 pix = 0;
1208                         }
1209                 }
1210         }
1211         if (pix) {
1212                 param[pix] = HBR_END;
1213                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1214                                   param[2], param[3], param[4], param[5],
1215                                   param[6], param[7]);
1216                 BUG_ON(rc != H_SUCCESS);
1217         }
1218 }
1219
1220 static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
1221                                                       unsigned long *vpn,
1222                                                       int count, int psize,
1223                                                       int ssize)
1224 {
1225         unsigned long flags = 0;
1226         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1227
1228         if (lock_tlbie)
1229                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1230
1231         /* Assuming THP size is 16M */
1232         if (is_supported_hlbkrm(psize, MMU_PAGE_16M))
1233                 hugepage_block_invalidate(slot, vpn, count, psize, ssize);
1234         else
1235                 hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
1236
1237         if (lock_tlbie)
1238                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1239 }
1240
1241 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1242                                              unsigned long addr,
1243                                              unsigned char *hpte_slot_array,
1244                                              int psize, int ssize, int local)
1245 {
1246         int i, index = 0;
1247         unsigned long s_addr = addr;
1248         unsigned int max_hpte_count, valid;
1249         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
1250         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
1251         unsigned long shift, hidx, vpn = 0, hash, slot;
1252
1253         shift = mmu_psize_defs[psize].shift;
1254         max_hpte_count = 1U << (PMD_SHIFT - shift);
1255
1256         for (i = 0; i < max_hpte_count; i++) {
1257                 valid = hpte_valid(hpte_slot_array, i);
1258                 if (!valid)
1259                         continue;
1260                 hidx =  hpte_hash_index(hpte_slot_array, i);
1261
1262                 /* get the vpn */
1263                 addr = s_addr + (i * (1ul << shift));
1264                 vpn = hpt_vpn(addr, vsid, ssize);
1265                 hash = hpt_hash(vpn, shift, ssize);
1266                 if (hidx & _PTEIDX_SECONDARY)
1267                         hash = ~hash;
1268
1269                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1270                 slot += hidx & _PTEIDX_GROUP_IX;
1271
1272                 slot_array[index] = slot;
1273                 vpn_array[index] = vpn;
1274                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
1275                         /*
1276                          * Now do a bluk invalidate
1277                          */
1278                         __pSeries_lpar_hugepage_invalidate(slot_array,
1279                                                            vpn_array,
1280                                                            PPC64_HUGE_HPTE_BATCH,
1281                                                            psize, ssize);
1282                         index = 0;
1283                 } else
1284                         index++;
1285         }
1286         if (index)
1287                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
1288                                                    index, psize, ssize);
1289 }
1290 #else
1291 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1292                                              unsigned long addr,
1293                                              unsigned char *hpte_slot_array,
1294                                              int psize, int ssize, int local)
1295 {
1296         WARN(1, "%s called without THP support\n", __func__);
1297 }
1298 #endif
1299
1300 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
1301                                           int psize, int ssize)
1302 {
1303         unsigned long vpn;
1304         unsigned long slot, vsid;
1305
1306         vsid = get_kernel_vsid(ea, ssize);
1307         vpn = hpt_vpn(ea, vsid, ssize);
1308
1309         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1310         if (slot == -1)
1311                 return -ENOENT;
1312
1313         /*
1314          * lpar doesn't use the passed actual page size
1315          */
1316         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
1317         return 0;
1318 }
1319
1320
1321 static inline unsigned long compute_slot(real_pte_t pte,
1322                                          unsigned long vpn,
1323                                          unsigned long index,
1324                                          unsigned long shift,
1325                                          int ssize)
1326 {
1327         unsigned long slot, hash, hidx;
1328
1329         hash = hpt_hash(vpn, shift, ssize);
1330         hidx = __rpte_to_hidx(pte, index);
1331         if (hidx & _PTEIDX_SECONDARY)
1332                 hash = ~hash;
1333         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1334         slot += hidx & _PTEIDX_GROUP_IX;
1335         return slot;
1336 }
1337
1338 /**
1339  * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
1340  * "all within the same naturally aligned 8 page virtual address block".
1341  */
1342 static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
1343                             unsigned long *param)
1344 {
1345         unsigned long vpn;
1346         unsigned long i, pix = 0;
1347         unsigned long index, shift, slot, current_vpgb, vpgb;
1348         real_pte_t pte;
1349         int psize, ssize;
1350
1351         psize = batch->psize;
1352         ssize = batch->ssize;
1353
1354         for (i = 0; i < number; i++) {
1355                 vpn = batch->vpn[i];
1356                 pte = batch->pte[i];
1357                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1358                         /*
1359                          * Shifting 3 bits more on the right to get a
1360                          * 8 pages aligned virtual addresse.
1361                          */
1362                         vpgb = (vpn >> (shift - VPN_SHIFT + 3));
1363                         if (!pix || vpgb != current_vpgb) {
1364                                 /*
1365                                  * Need to start a new 8 pages block, flush
1366                                  * the current one if needed.
1367                                  */
1368                                 if (pix)
1369                                         (void)call_block_remove(pix, param,
1370                                                                 true);
1371                                 current_vpgb = vpgb;
1372                                 param[0] = hpte_encode_avpn(vpn, psize,
1373                                                             ssize);
1374                                 pix = 1;
1375                         }
1376
1377                         slot = compute_slot(pte, vpn, index, shift, ssize);
1378                         param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
1379
1380                         if (pix == PLPAR_HCALL9_BUFSIZE) {
1381                                 pix = call_block_remove(pix, param, false);
1382                                 /*
1383                                  * pix = 0 means that all the entries were
1384                                  * removed, we can start a new block.
1385                                  * Otherwise, this means that there are entries
1386                                  * to retry, and pix points to latest one, so
1387                                  * we should increment it and try to continue
1388                                  * the same block.
1389                                  */
1390                                 if (pix)
1391                                         pix++;
1392                         }
1393                 } pte_iterate_hashed_end();
1394         }
1395
1396         if (pix)
1397                 (void)call_block_remove(pix, param, true);
1398 }
1399
1400 /*
1401  * TLB Block Invalidate Characteristics
1402  *
1403  * These characteristics define the size of the block the hcall H_BLOCK_REMOVE
1404  * is able to process for each couple segment base page size, actual page size.
1405  *
1406  * The ibm,get-system-parameter properties is returning a buffer with the
1407  * following layout:
1408  *
1409  * [ 2 bytes size of the RTAS buffer (excluding these 2 bytes) ]
1410  * -----------------
1411  * TLB Block Invalidate Specifiers:
1412  * [ 1 byte LOG base 2 of the TLB invalidate block size being specified ]
1413  * [ 1 byte Number of page sizes (N) that are supported for the specified
1414  *          TLB invalidate block size ]
1415  * [ 1 byte Encoded segment base page size and actual page size
1416  *          MSB=0 means 4k segment base page size and actual page size
1417  *          MSB=1 the penc value in mmu_psize_def ]
1418  * ...
1419  * -----------------
1420  * Next TLB Block Invalidate Specifiers...
1421  * -----------------
1422  * [ 0 ]
1423  */
1424 static inline void set_hblkrm_bloc_size(int bpsize, int psize,
1425                                         unsigned int block_size)
1426 {
1427         if (block_size > hblkrm_size[bpsize][psize])
1428                 hblkrm_size[bpsize][psize] = block_size;
1429 }
1430
1431 /*
1432  * Decode the Encoded segment base page size and actual page size.
1433  * PAPR specifies:
1434  *   - bit 7 is the L bit
1435  *   - bits 0-5 are the penc value
1436  * If the L bit is 0, this means 4K segment base page size and actual page size
1437  * otherwise the penc value should be read.
1438  */
1439 #define HBLKRM_L_MASK           0x80
1440 #define HBLKRM_PENC_MASK        0x3f
1441 static inline void __init check_lp_set_hblkrm(unsigned int lp,
1442                                               unsigned int block_size)
1443 {
1444         unsigned int bpsize, psize;
1445
1446         /* First, check the L bit, if not set, this means 4K */
1447         if ((lp & HBLKRM_L_MASK) == 0) {
1448                 set_hblkrm_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
1449                 return;
1450         }
1451
1452         lp &= HBLKRM_PENC_MASK;
1453         for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
1454                 struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
1455
1456                 for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1457                         if (def->penc[psize] == lp) {
1458                                 set_hblkrm_bloc_size(bpsize, psize, block_size);
1459                                 return;
1460                         }
1461                 }
1462         }
1463 }
1464
1465 /*
1466  * The size of the TLB Block Invalidate Characteristics is variable. But at the
1467  * maximum it will be the number of possible page sizes *2 + 10 bytes.
1468  * Currently MMU_PAGE_COUNT is 16, which means 42 bytes. Use a cache line size
1469  * (128 bytes) for the buffer to get plenty of space.
1470  */
1471 #define SPLPAR_TLB_BIC_MAXLENGTH        128
1472
1473 void __init pseries_lpar_read_hblkrm_characteristics(void)
1474 {
1475         static struct papr_sysparm_buf buf __initdata;
1476         int len, idx, bpsize;
1477
1478         if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
1479                 return;
1480
1481         if (papr_sysparm_get(PAPR_SYSPARM_TLB_BLOCK_INVALIDATE_ATTRS, &buf))
1482                 return;
1483
1484         len = be16_to_cpu(buf.len);
1485         if (len > SPLPAR_TLB_BIC_MAXLENGTH) {
1486                 pr_warn("%s too large returned buffer %d", __func__, len);
1487                 return;
1488         }
1489
1490         idx = 0;
1491         while (idx < len) {
1492                 u8 block_shift = buf.val[idx++];
1493                 u32 block_size;
1494                 unsigned int npsize;
1495
1496                 if (!block_shift)
1497                         break;
1498
1499                 block_size = 1 << block_shift;
1500
1501                 for (npsize = buf.val[idx++];
1502                      npsize > 0 && idx < len; npsize--)
1503                         check_lp_set_hblkrm((unsigned int)buf.val[idx++],
1504                                             block_size);
1505         }
1506
1507         for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
1508                 for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
1509                         if (hblkrm_size[bpsize][idx])
1510                                 pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
1511                                         bpsize, idx, hblkrm_size[bpsize][idx]);
1512 }
1513
1514 /*
1515  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
1516  * lock.
1517  */
1518 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
1519 {
1520         unsigned long vpn;
1521         unsigned long i, pix, rc;
1522         unsigned long flags = 0;
1523         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
1524         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1525         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1526         unsigned long index, shift, slot;
1527         real_pte_t pte;
1528         int psize, ssize;
1529
1530         if (lock_tlbie)
1531                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1532
1533         if (is_supported_hlbkrm(batch->psize, batch->psize)) {
1534                 do_block_remove(number, batch, param);
1535                 goto out;
1536         }
1537
1538         psize = batch->psize;
1539         ssize = batch->ssize;
1540         pix = 0;
1541         for (i = 0; i < number; i++) {
1542                 vpn = batch->vpn[i];
1543                 pte = batch->pte[i];
1544                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1545                         slot = compute_slot(pte, vpn, index, shift, ssize);
1546                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1547                                 /*
1548                                  * lpar doesn't use the passed actual page size
1549                                  */
1550                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
1551                                                              0, ssize, local);
1552                         } else {
1553                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
1554                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
1555                                                                 ssize);
1556                                 pix += 2;
1557                                 if (pix == 8) {
1558                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
1559                                                 param[0], param[1], param[2],
1560                                                 param[3], param[4], param[5],
1561                                                 param[6], param[7]);
1562                                         BUG_ON(rc != H_SUCCESS);
1563                                         pix = 0;
1564                                 }
1565                         }
1566                 } pte_iterate_hashed_end();
1567         }
1568         if (pix) {
1569                 param[pix] = HBR_END;
1570                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1571                                   param[2], param[3], param[4], param[5],
1572                                   param[6], param[7]);
1573                 BUG_ON(rc != H_SUCCESS);
1574         }
1575
1576 out:
1577         if (lock_tlbie)
1578                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1579 }
1580
1581 static int __init disable_bulk_remove(char *str)
1582 {
1583         if (strcmp(str, "off") == 0 &&
1584             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1585                 pr_info("Disabling BULK_REMOVE firmware feature");
1586                 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
1587         }
1588         return 1;
1589 }
1590
1591 __setup("bulk_remove=", disable_bulk_remove);
1592
1593 #define HPT_RESIZE_TIMEOUT      10000 /* ms */
1594
1595 struct hpt_resize_state {
1596         unsigned long shift;
1597         int commit_rc;
1598 };
1599
1600 static int pseries_lpar_resize_hpt_commit(void *data)
1601 {
1602         struct hpt_resize_state *state = data;
1603
1604         state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
1605         if (state->commit_rc != H_SUCCESS)
1606                 return -EIO;
1607
1608         /* Hypervisor has transitioned the HTAB, update our globals */
1609         ppc64_pft_size = state->shift;
1610         htab_size_bytes = 1UL << ppc64_pft_size;
1611         htab_hash_mask = (htab_size_bytes >> 7) - 1;
1612
1613         return 0;
1614 }
1615
1616 /*
1617  * Must be called in process context. The caller must hold the
1618  * cpus_lock.
1619  */
1620 static int pseries_lpar_resize_hpt(unsigned long shift)
1621 {
1622         struct hpt_resize_state state = {
1623                 .shift = shift,
1624                 .commit_rc = H_FUNCTION,
1625         };
1626         unsigned int delay, total_delay = 0;
1627         int rc;
1628         ktime_t t0, t1, t2;
1629
1630         might_sleep();
1631
1632         if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1633                 return -ENODEV;
1634
1635         pr_info("Attempting to resize HPT to shift %lu\n", shift);
1636
1637         t0 = ktime_get();
1638
1639         rc = plpar_resize_hpt_prepare(0, shift);
1640         while (H_IS_LONG_BUSY(rc)) {
1641                 delay = get_longbusy_msecs(rc);
1642                 total_delay += delay;
1643                 if (total_delay > HPT_RESIZE_TIMEOUT) {
1644                         /* prepare with shift==0 cancels an in-progress resize */
1645                         rc = plpar_resize_hpt_prepare(0, 0);
1646                         if (rc != H_SUCCESS)
1647                                 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
1648                                        rc);
1649                         return -ETIMEDOUT;
1650                 }
1651                 msleep(delay);
1652                 rc = plpar_resize_hpt_prepare(0, shift);
1653         }
1654
1655         switch (rc) {
1656         case H_SUCCESS:
1657                 /* Continue on */
1658                 break;
1659
1660         case H_PARAMETER:
1661                 pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
1662                 return -EINVAL;
1663         case H_RESOURCE:
1664                 pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
1665                 return -EPERM;
1666         default:
1667                 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
1668                 return -EIO;
1669         }
1670
1671         t1 = ktime_get();
1672
1673         rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
1674                                      &state, NULL);
1675
1676         t2 = ktime_get();
1677
1678         if (rc != 0) {
1679                 switch (state.commit_rc) {
1680                 case H_PTEG_FULL:
1681                         return -ENOSPC;
1682
1683                 default:
1684                         pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
1685                                 state.commit_rc);
1686                         return -EIO;
1687                 };
1688         }
1689
1690         pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
1691                 shift, (long long) ktime_ms_delta(t1, t0),
1692                 (long long) ktime_ms_delta(t2, t1));
1693
1694         return 0;
1695 }
1696
1697 void __init hpte_init_pseries(void)
1698 {
1699         mmu_hash_ops.hpte_invalidate     = pSeries_lpar_hpte_invalidate;
1700         mmu_hash_ops.hpte_updatepp       = pSeries_lpar_hpte_updatepp;
1701         mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
1702         mmu_hash_ops.hpte_insert         = pSeries_lpar_hpte_insert;
1703         mmu_hash_ops.hpte_remove         = pSeries_lpar_hpte_remove;
1704         mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
1705         mmu_hash_ops.flush_hash_range    = pSeries_lpar_flush_hash_range;
1706         mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
1707         mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
1708
1709         if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1710                 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
1711
1712         /*
1713          * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
1714          * to inform the hypervisor that we wish to use the HPT.
1715          */
1716         if (cpu_has_feature(CPU_FTR_ARCH_300))
1717                 pseries_lpar_register_process_table(0, 0, 0);
1718 }
1719 #endif /* CONFIG_PPC_64S_HASH_MMU */
1720
1721 #ifdef CONFIG_PPC_RADIX_MMU
1722 void __init radix_init_pseries(void)
1723 {
1724         pr_info("Using radix MMU under hypervisor\n");
1725
1726         pseries_lpar_register_process_table(__pa(process_tb),
1727                                                 0, PRTB_SIZE_SHIFT - 12);
1728 }
1729 #endif
1730
1731 #ifdef CONFIG_PPC_SMLPAR
1732 #define CMO_FREE_HINT_DEFAULT 1
1733 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
1734
1735 static int __init cmo_free_hint(char *str)
1736 {
1737         char *parm;
1738         parm = strstrip(str);
1739
1740         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
1741                 pr_info("%s: CMO free page hinting is not active.\n", __func__);
1742                 cmo_free_hint_flag = 0;
1743                 return 1;
1744         }
1745
1746         cmo_free_hint_flag = 1;
1747         pr_info("%s: CMO free page hinting is active.\n", __func__);
1748
1749         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1750                 return 1;
1751
1752         return 0;
1753 }
1754
1755 __setup("cmo_free_hint=", cmo_free_hint);
1756
1757 static void pSeries_set_page_state(struct page *page, int order,
1758                                    unsigned long state)
1759 {
1760         int i, j;
1761         unsigned long cmo_page_sz, addr;
1762
1763         cmo_page_sz = cmo_get_page_size();
1764         addr = __pa((unsigned long)page_address(page));
1765
1766         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1767                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1768                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1769         }
1770 }
1771
1772 void arch_free_page(struct page *page, int order)
1773 {
1774         if (radix_enabled())
1775                 return;
1776         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1777                 return;
1778
1779         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1780 }
1781 EXPORT_SYMBOL(arch_free_page);
1782
1783 #endif /* CONFIG_PPC_SMLPAR */
1784 #endif /* CONFIG_PPC_BOOK3S_64 */
1785
1786 #ifdef CONFIG_TRACEPOINTS
1787 #ifdef CONFIG_JUMP_LABEL
1788 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1789
1790 int hcall_tracepoint_regfunc(void)
1791 {
1792         static_key_slow_inc(&hcall_tracepoint_key);
1793         return 0;
1794 }
1795
1796 void hcall_tracepoint_unregfunc(void)
1797 {
1798         static_key_slow_dec(&hcall_tracepoint_key);
1799 }
1800 #else
1801 /*
1802  * We optimise our hcall path by placing hcall_tracepoint_refcount
1803  * directly in the TOC so we can check if the hcall tracepoints are
1804  * enabled via a single load.
1805  */
1806
1807 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1808 extern long hcall_tracepoint_refcount;
1809
1810 int hcall_tracepoint_regfunc(void)
1811 {
1812         hcall_tracepoint_refcount++;
1813         return 0;
1814 }
1815
1816 void hcall_tracepoint_unregfunc(void)
1817 {
1818         hcall_tracepoint_refcount--;
1819 }
1820 #endif
1821
1822 /*
1823  * Keep track of hcall tracing depth and prevent recursion. Warn if any is
1824  * detected because it may indicate a problem. This will not catch all
1825  * problems with tracing code making hcalls, because the tracing might have
1826  * been invoked from a non-hcall, so the first hcall could recurse into it
1827  * without warning here, but this better than nothing.
1828  *
1829  * Hcalls with specific problems being traced should use the _notrace
1830  * plpar_hcall variants.
1831  */
1832 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1833
1834
1835 notrace void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1836 {
1837         unsigned long flags;
1838         unsigned int *depth;
1839
1840         local_irq_save(flags);
1841
1842         depth = this_cpu_ptr(&hcall_trace_depth);
1843
1844         if (WARN_ON_ONCE(*depth))
1845                 goto out;
1846
1847         (*depth)++;
1848         preempt_disable();
1849         trace_hcall_entry(opcode, args);
1850         (*depth)--;
1851
1852 out:
1853         local_irq_restore(flags);
1854 }
1855
1856 notrace void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1857 {
1858         unsigned long flags;
1859         unsigned int *depth;
1860
1861         local_irq_save(flags);
1862
1863         depth = this_cpu_ptr(&hcall_trace_depth);
1864
1865         if (*depth) /* Don't warn again on the way out */
1866                 goto out;
1867
1868         (*depth)++;
1869         trace_hcall_exit(opcode, retval, retbuf);
1870         preempt_enable();
1871         (*depth)--;
1872
1873 out:
1874         local_irq_restore(flags);
1875 }
1876 #endif
1877
1878 /**
1879  * h_get_mpp
1880  * H_GET_MPP hcall returns info in 7 parms
1881  */
1882 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
1883 {
1884         int rc;
1885         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1886
1887         rc = plpar_hcall9(H_GET_MPP, retbuf);
1888
1889         mpp_data->entitled_mem = retbuf[0];
1890         mpp_data->mapped_mem = retbuf[1];
1891
1892         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1893         mpp_data->pool_num = retbuf[2] & 0xffff;
1894
1895         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1896         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1897         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1898
1899         mpp_data->pool_size = retbuf[4];
1900         mpp_data->loan_request = retbuf[5];
1901         mpp_data->backing_mem = retbuf[6];
1902
1903         return rc;
1904 }
1905 EXPORT_SYMBOL(h_get_mpp);
1906
1907 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1908 {
1909         int rc;
1910         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1911
1912         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1913
1914         mpp_x_data->coalesced_bytes = retbuf[0];
1915         mpp_x_data->pool_coalesced_bytes = retbuf[1];
1916         mpp_x_data->pool_purr_cycles = retbuf[2];
1917         mpp_x_data->pool_spurr_cycles = retbuf[3];
1918
1919         return rc;
1920 }
1921
1922 #ifdef CONFIG_PPC_64S_HASH_MMU
1923 static unsigned long __init vsid_unscramble(unsigned long vsid, int ssize)
1924 {
1925         unsigned long protovsid;
1926         unsigned long va_bits = VA_BITS;
1927         unsigned long modinv, vsid_modulus;
1928         unsigned long max_mod_inv, tmp_modinv;
1929
1930         if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1931                 va_bits = 65;
1932
1933         if (ssize == MMU_SEGSIZE_256M) {
1934                 modinv = VSID_MULINV_256M;
1935                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1936         } else {
1937                 modinv = VSID_MULINV_1T;
1938                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1939         }
1940
1941         /*
1942          * vsid outside our range.
1943          */
1944         if (vsid >= vsid_modulus)
1945                 return 0;
1946
1947         /*
1948          * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1949          * and vsid = (protovsid * x) % vsid_modulus, then we say:
1950          *   protovsid = (vsid * modinv) % vsid_modulus
1951          */
1952
1953         /* Check if (vsid * modinv) overflow (63 bits) */
1954         max_mod_inv = 0x7fffffffffffffffull / vsid;
1955         if (modinv < max_mod_inv)
1956                 return (vsid * modinv) % vsid_modulus;
1957
1958         tmp_modinv = modinv/max_mod_inv;
1959         modinv %= max_mod_inv;
1960
1961         protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1962         protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1963
1964         return protovsid;
1965 }
1966
1967 static int __init reserve_vrma_context_id(void)
1968 {
1969         unsigned long protovsid;
1970
1971         /*
1972          * Reserve context ids which map to reserved virtual addresses. For now
1973          * we only reserve the context id which maps to the VRMA VSID. We ignore
1974          * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1975          * enable adjunct support via the "ibm,client-architecture-support"
1976          * interface.
1977          */
1978         protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1979         hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1980         return 0;
1981 }
1982 machine_device_initcall(pseries, reserve_vrma_context_id);
1983 #endif
1984
1985 #ifdef CONFIG_DEBUG_FS
1986 /* debugfs file interface for vpa data */
1987 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1988                               loff_t *pos)
1989 {
1990         int cpu = (long)filp->private_data;
1991         struct lppaca *lppaca = &lppaca_of(cpu);
1992
1993         return simple_read_from_buffer(buf, len, pos, lppaca,
1994                                 sizeof(struct lppaca));
1995 }
1996
1997 static const struct file_operations vpa_fops = {
1998         .open           = simple_open,
1999         .read           = vpa_file_read,
2000         .llseek         = default_llseek,
2001 };
2002
2003 static int __init vpa_debugfs_init(void)
2004 {
2005         char name[16];
2006         long i;
2007         struct dentry *vpa_dir;
2008
2009         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
2010                 return 0;
2011
2012         vpa_dir = debugfs_create_dir("vpa", arch_debugfs_dir);
2013
2014         /* set up the per-cpu vpa file*/
2015         for_each_possible_cpu(i) {
2016                 sprintf(name, "cpu-%ld", i);
2017                 debugfs_create_file(name, 0400, vpa_dir, (void *)i, &vpa_fops);
2018         }
2019
2020         return 0;
2021 }
2022 machine_arch_initcall(pseries, vpa_debugfs_init);
2023 #endif /* CONFIG_DEBUG_FS */