arch/powerpc/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 <asm/processor.h>
  25 #include <asm/mmu.h>
  26 #include <asm/page.h>
  27 #include <linux/pgtable.h>
  28 #include <asm/machdep.h>
  29 #include <asm/mmu_context.h>
  30 #include <asm/iommu.h>
  31 #include <asm/tlb.h>
  32 #include <asm/prom.h>
  33 #include <asm/cputable.h>
  34 #include <asm/udbg.h>
  35 #include <asm/smp.h>
  36 #include <asm/trace.h>
  37 #include <asm/firmware.h>
  38 #include <asm/plpar_wrappers.h>
  39 #include <asm/kexec.h>
  40 #include <asm/fadump.h>
  41 #include <asm/asm-prototypes.h>
  42 #include <asm/debugfs.h>
  43
  44 #include "pseries.h"
  45
  46 /* Flag bits for H_BULK_REMOVE */
  47 #define HBR_REQUEST     0x4000000000000000UL
  48 #define HBR_RESPONSE    0x8000000000000000UL
  49 #define HBR_END         0xc000000000000000UL
  50 #define HBR_AVPN        0x0200000000000000UL
  51 #define HBR_ANDCOND     0x0100000000000000UL
  52
  53
  54 /* in hvCall.S */
  55 EXPORT_SYMBOL(plpar_hcall);
  56 EXPORT_SYMBOL(plpar_hcall9);
  57 EXPORT_SYMBOL(plpar_hcall_norets);
  58
  59 /*
  60  * H_BLOCK_REMOVE supported block size for this page size in segment who's base
  61  * page size is that page size.
  62  *
  63  * The first index is the segment base page size, the second one is the actual
  64  * page size.
  65  */
  66 static int hblkrm_size[MMU_PAGE_COUNT][MMU_PAGE_COUNT] __ro_after_init;
  67
  68 /*
  69  * Due to the involved complexity, and that the current hypervisor is only
  70  * returning this value or 0, we are limiting the support of the H_BLOCK_REMOVE
  71  * buffer size to 8 size block.
  72  */
  73 #define HBLKRM_SUPPORTED_BLOCK_SIZE 8
  74
  75 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
  76 static u8 dtl_mask = DTL_LOG_PREEMPT;
  77 #else
  78 static u8 dtl_mask;
  79 #endif
  80
  81 void alloc_dtl_buffers(unsigned long *time_limit)
  82 {
  83         int cpu;
  84         struct paca_struct *pp;
  85         struct dtl_entry *dtl;
  86
  87         for_each_possible_cpu(cpu) {
  88                 pp = paca_ptrs[cpu];
  89                 if (pp->dispatch_log)
  90                         continue;
  91                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
  92                 if (!dtl) {
  93                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
  94                                 cpu);
  95 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
  96                         pr_warn("Stolen time statistics will be unreliable\n");
  97 #endif
  98                         break;
  99                 }
 100
 101                 pp->dtl_ridx = 0;
 102                 pp->dispatch_log = dtl;
 103                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
 104                 pp->dtl_curr = dtl;
 105
 106                 if (time_limit && time_after(jiffies, *time_limit)) {
 107                         cond_resched();
 108                         *time_limit = jiffies + HZ;
 109                 }
 110         }
 111 }
 112
 113 void register_dtl_buffer(int cpu)
 114 {
 115         long ret;
 116         struct paca_struct *pp;
 117         struct dtl_entry *dtl;
 118         int hwcpu = get_hard_smp_processor_id(cpu);
 119
 120         pp = paca_ptrs[cpu];
 121         dtl = pp->dispatch_log;
 122         if (dtl && dtl_mask) {
 123                 pp->dtl_ridx = 0;
 124                 pp->dtl_curr = dtl;
 125                 lppaca_of(cpu).dtl_idx = 0;
 126
 127                 /* hypervisor reads buffer length from this field */
 128                 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 129                 ret = register_dtl(hwcpu, __pa(dtl));
 130                 if (ret)
 131                         pr_err("WARNING: DTL registration of cpu %d (hw %d) failed with %ld\n",
 132                                cpu, hwcpu, ret);
 133
 134                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 135         }
 136 }
 137
 138 #ifdef CONFIG_PPC_SPLPAR
 139 struct dtl_worker {
 140         struct delayed_work work;
 141         int cpu;
 142 };
 143
 144 struct vcpu_dispatch_data {
 145         int last_disp_cpu;
 146
 147         int total_disp;
 148
 149         int same_cpu_disp;
 150         int same_chip_disp;
 151         int diff_chip_disp;
 152         int far_chip_disp;
 153
 154         int numa_home_disp;
 155         int numa_remote_disp;
 156         int numa_far_disp;
 157 };
 158
 159 /*
 160  * This represents the number of cpus in the hypervisor. Since there is no
 161  * architected way to discover the number of processors in the host, we
 162  * provision for dealing with NR_CPUS. This is currently 2048 by default, and
 163  * is sufficient for our purposes. This will need to be tweaked if
 164  * CONFIG_NR_CPUS is changed.
 165  */
 166 #define NR_CPUS_H       NR_CPUS
 167
 168 DEFINE_RWLOCK(dtl_access_lock);
 169 static DEFINE_PER_CPU(struct vcpu_dispatch_data, vcpu_disp_data);
 170 static DEFINE_PER_CPU(u64, dtl_entry_ridx);
 171 static DEFINE_PER_CPU(struct dtl_worker, dtl_workers);
 172 static enum cpuhp_state dtl_worker_state;
 173 static DEFINE_MUTEX(dtl_enable_mutex);
 174 static int vcpudispatch_stats_on __read_mostly;
 175 static int vcpudispatch_stats_freq = 50;
 176 static __be32 *vcpu_associativity, *pcpu_associativity;
 177
 178
 179 static void free_dtl_buffers(unsigned long *time_limit)
 180 {
 181 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 182         int cpu;
 183         struct paca_struct *pp;
 184
 185         for_each_possible_cpu(cpu) {
 186                 pp = paca_ptrs[cpu];
 187                 if (!pp->dispatch_log)
 188                         continue;
 189                 kmem_cache_free(dtl_cache, pp->dispatch_log);
 190                 pp->dtl_ridx = 0;
 191                 pp->dispatch_log = 0;
 192                 pp->dispatch_log_end = 0;
 193                 pp->dtl_curr = 0;
 194
 195                 if (time_limit && time_after(jiffies, *time_limit)) {
 196                         cond_resched();
 197                         *time_limit = jiffies + HZ;
 198                 }
 199         }
 200 #endif
 201 }
 202
 203 static int init_cpu_associativity(void)
 204 {
 205         vcpu_associativity = kcalloc(num_possible_cpus() / threads_per_core,
 206                         VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
 207         pcpu_associativity = kcalloc(NR_CPUS_H / threads_per_core,
 208                         VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
 209
 210         if (!vcpu_associativity || !pcpu_associativity) {
 211                 pr_err("error allocating memory for associativity information\n");
 212                 return -ENOMEM;
 213         }
 214
 215         return 0;
 216 }
 217
 218 static void destroy_cpu_associativity(void)
 219 {
 220         kfree(vcpu_associativity);
 221         kfree(pcpu_associativity);
 222         vcpu_associativity = pcpu_associativity = 0;
 223 }
 224
 225 static __be32 *__get_cpu_associativity(int cpu, __be32 *cpu_assoc, int flag)
 226 {
 227         __be32 *assoc;
 228         int rc = 0;
 229
 230         assoc = &cpu_assoc[(int)(cpu / threads_per_core) * VPHN_ASSOC_BUFSIZE];
 231         if (!assoc[0]) {
 232                 rc = hcall_vphn(cpu, flag, &assoc[0]);
 233                 if (rc)
 234                         return NULL;
 235         }
 236
 237         return assoc;
 238 }
 239
 240 static __be32 *get_pcpu_associativity(int cpu)
 241 {
 242         return __get_cpu_associativity(cpu, pcpu_associativity, VPHN_FLAG_PCPU);
 243 }
 244
 245 static __be32 *get_vcpu_associativity(int cpu)
 246 {
 247         return __get_cpu_associativity(cpu, vcpu_associativity, VPHN_FLAG_VCPU);
 248 }
 249
 250 static int cpu_relative_dispatch_distance(int last_disp_cpu, int cur_disp_cpu)
 251 {
 252         __be32 *last_disp_cpu_assoc, *cur_disp_cpu_assoc;
 253
 254         if (last_disp_cpu >= NR_CPUS_H || cur_disp_cpu >= NR_CPUS_H)
 255                 return -EINVAL;
 256
 257         last_disp_cpu_assoc = get_pcpu_associativity(last_disp_cpu);
 258         cur_disp_cpu_assoc = get_pcpu_associativity(cur_disp_cpu);
 259
 260         if (!last_disp_cpu_assoc || !cur_disp_cpu_assoc)
 261                 return -EIO;
 262
 263         return cpu_distance(last_disp_cpu_assoc, cur_disp_cpu_assoc);
 264 }
 265
 266 static int cpu_home_node_dispatch_distance(int disp_cpu)
 267 {
 268         __be32 *disp_cpu_assoc, *vcpu_assoc;
 269         int vcpu_id = smp_processor_id();
 270
 271         if (disp_cpu >= NR_CPUS_H) {
 272                 pr_debug_ratelimited("vcpu dispatch cpu %d > %d\n",
 273                                                 disp_cpu, NR_CPUS_H);
 274                 return -EINVAL;
 275         }
 276
 277         disp_cpu_assoc = get_pcpu_associativity(disp_cpu);
 278         vcpu_assoc = get_vcpu_associativity(vcpu_id);
 279
 280         if (!disp_cpu_assoc || !vcpu_assoc)
 281                 return -EIO;
 282
 283         return cpu_distance(disp_cpu_assoc, vcpu_assoc);
 284 }
 285
 286 static void update_vcpu_disp_stat(int disp_cpu)
 287 {
 288         struct vcpu_dispatch_data *disp;
 289         int distance;
 290
 291         disp = this_cpu_ptr(&vcpu_disp_data);
 292         if (disp->last_disp_cpu == -1) {
 293                 disp->last_disp_cpu = disp_cpu;
 294                 return;
 295         }
 296
 297         disp->total_disp++;
 298
 299         if (disp->last_disp_cpu == disp_cpu ||
 300                 (cpu_first_thread_sibling(disp->last_disp_cpu) ==
 301                                         cpu_first_thread_sibling(disp_cpu)))
 302                 disp->same_cpu_disp++;
 303         else {
 304                 distance = cpu_relative_dispatch_distance(disp->last_disp_cpu,
 305                                                                 disp_cpu);
 306                 if (distance < 0)
 307                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
 308                                         smp_processor_id());
 309                 else {
 310                         switch (distance) {
 311                         case 0:
 312                                 disp->same_chip_disp++;
 313                                 break;
 314                         case 1:
 315                                 disp->diff_chip_disp++;
 316                                 break;
 317                         case 2:
 318                                 disp->far_chip_disp++;
 319                                 break;
 320                         default:
 321                                 pr_debug_ratelimited("vcpudispatch_stats: cpu %d (%d -> %d): unexpected relative dispatch distance %d\n",
 322                                                  smp_processor_id(),
 323                                                  disp->last_disp_cpu,
 324                                                  disp_cpu,
 325                                                  distance);
 326                         }
 327                 }
 328         }
 329
 330         distance = cpu_home_node_dispatch_distance(disp_cpu);
 331         if (distance < 0)
 332                 pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
 333                                 smp_processor_id());
 334         else {
 335                 switch (distance) {
 336                 case 0:
 337                         disp->numa_home_disp++;
 338                         break;
 339                 case 1:
 340                         disp->numa_remote_disp++;
 341                         break;
 342                 case 2:
 343                         disp->numa_far_disp++;
 344                         break;
 345                 default:
 346                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d on %d: unexpected numa dispatch distance %d\n",
 347                                                  smp_processor_id(),
 348                                                  disp_cpu,
 349                                                  distance);
 350                 }
 351         }
 352
 353         disp->last_disp_cpu = disp_cpu;
 354 }
 355
 356 static void process_dtl_buffer(struct work_struct *work)
 357 {
 358         struct dtl_entry dtle;
 359         u64 i = __this_cpu_read(dtl_entry_ridx);
 360         struct dtl_entry *dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
 361         struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
 362         struct lppaca *vpa = local_paca->lppaca_ptr;
 363         struct dtl_worker *d = container_of(work, struct dtl_worker, work.work);
 364
 365         if (!local_paca->dispatch_log)
 366                 return;
 367
 368         /* if we have been migrated away, we cancel ourself */
 369         if (d->cpu != smp_processor_id()) {
 370                 pr_debug("vcpudispatch_stats: cpu %d worker migrated -- canceling worker\n",
 371                                                 smp_processor_id());
 372                 return;
 373         }
 374
 375         if (i == be64_to_cpu(vpa->dtl_idx))
 376                 goto out;
 377
 378         while (i < be64_to_cpu(vpa->dtl_idx)) {
 379                 dtle = *dtl;
 380                 barrier();
 381                 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
 382                         /* buffer has overflowed */
 383                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d lost %lld DTL samples\n",
 384                                 d->cpu,
 385                                 be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG - i);
 386                         i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
 387                         dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
 388                         continue;
 389                 }
 390                 update_vcpu_disp_stat(be16_to_cpu(dtle.processor_id));
 391                 ++i;
 392                 ++dtl;
 393                 if (dtl == dtl_end)
 394                         dtl = local_paca->dispatch_log;
 395         }
 396
 397         __this_cpu_write(dtl_entry_ridx, i);
 398
 399 out:
 400         schedule_delayed_work_on(d->cpu, to_delayed_work(work),
 401                                         HZ / vcpudispatch_stats_freq);
 402 }
 403
 404 static int dtl_worker_online(unsigned int cpu)
 405 {
 406         struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
 407
 408         memset(d, 0, sizeof(*d));
 409         INIT_DELAYED_WORK(&d->work, process_dtl_buffer);
 410         d->cpu = cpu;
 411
 412 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 413         per_cpu(dtl_entry_ridx, cpu) = 0;
 414         register_dtl_buffer(cpu);
 415 #else
 416         per_cpu(dtl_entry_ridx, cpu) = be64_to_cpu(lppaca_of(cpu).dtl_idx);
 417 #endif
 418
 419         schedule_delayed_work_on(cpu, &d->work, HZ / vcpudispatch_stats_freq);
 420         return 0;
 421 }
 422
 423 static int dtl_worker_offline(unsigned int cpu)
 424 {
 425         struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
 426
 427         cancel_delayed_work_sync(&d->work);
 428
 429 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 430         unregister_dtl(get_hard_smp_processor_id(cpu));
 431 #endif
 432
 433         return 0;
 434 }
 435
 436 static void set_global_dtl_mask(u8 mask)
 437 {
 438         int cpu;
 439
 440         dtl_mask = mask;
 441         for_each_present_cpu(cpu)
 442                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 443 }
 444
 445 static void reset_global_dtl_mask(void)
 446 {
 447         int cpu;
 448
 449 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 450         dtl_mask = DTL_LOG_PREEMPT;
 451 #else
 452         dtl_mask = 0;
 453 #endif
 454         for_each_present_cpu(cpu)
 455                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 456 }
 457
 458 static int dtl_worker_enable(unsigned long *time_limit)
 459 {
 460         int rc = 0, state;
 461
 462         if (!write_trylock(&dtl_access_lock)) {
 463                 rc = -EBUSY;
 464                 goto out;
 465         }
 466
 467         set_global_dtl_mask(DTL_LOG_ALL);
 468
 469         /* Setup dtl buffers and register those */
 470         alloc_dtl_buffers(time_limit);
 471
 472         state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powerpc/dtl:online",
 473                                         dtl_worker_online, dtl_worker_offline);
 474         if (state < 0) {
 475                 pr_err("vcpudispatch_stats: unable to setup workqueue for DTL processing\n");
 476                 free_dtl_buffers(time_limit);
 477                 reset_global_dtl_mask();
 478                 write_unlock(&dtl_access_lock);
 479                 rc = -EINVAL;
 480                 goto out;
 481         }
 482         dtl_worker_state = state;
 483
 484 out:
 485         return rc;
 486 }
 487
 488 static void dtl_worker_disable(unsigned long *time_limit)
 489 {
 490         cpuhp_remove_state(dtl_worker_state);
 491         free_dtl_buffers(time_limit);
 492         reset_global_dtl_mask();
 493         write_unlock(&dtl_access_lock);
 494 }
 495
 496 static ssize_t vcpudispatch_stats_write(struct file *file, const char __user *p,
 497                 size_t count, loff_t *ppos)
 498 {
 499         unsigned long time_limit = jiffies + HZ;
 500         struct vcpu_dispatch_data *disp;
 501         int rc, cmd, cpu;
 502         char buf[16];
 503
 504         if (count > 15)
 505                 return -EINVAL;
 506
 507         if (copy_from_user(buf, p, count))
 508                 return -EFAULT;
 509
 510         buf[count] = 0;
 511         rc = kstrtoint(buf, 0, &cmd);
 512         if (rc || cmd < 0 || cmd > 1) {
 513                 pr_err("vcpudispatch_stats: please use 0 to disable or 1 to enable dispatch statistics\n");
 514                 return rc ? rc : -EINVAL;
 515         }
 516
 517         mutex_lock(&dtl_enable_mutex);
 518
 519         if ((cmd == 0 && !vcpudispatch_stats_on) ||
 520                         (cmd == 1 && vcpudispatch_stats_on))
 521                 goto out;
 522
 523         if (cmd) {
 524                 rc = init_cpu_associativity();
 525                 if (rc)
 526                         goto out;
 527
 528                 for_each_possible_cpu(cpu) {
 529                         disp = per_cpu_ptr(&vcpu_disp_data, cpu);
 530                         memset(disp, 0, sizeof(*disp));
 531                         disp->last_disp_cpu = -1;
 532                 }
 533
 534                 rc = dtl_worker_enable(&time_limit);
 535                 if (rc) {
 536                         destroy_cpu_associativity();
 537                         goto out;
 538                 }
 539         } else {
 540                 dtl_worker_disable(&time_limit);
 541                 destroy_cpu_associativity();
 542         }
 543
 544         vcpudispatch_stats_on = cmd;
 545
 546 out:
 547         mutex_unlock(&dtl_enable_mutex);
 548         if (rc)
 549                 return rc;
 550         return count;
 551 }
 552
 553 static int vcpudispatch_stats_display(struct seq_file *p, void *v)
 554 {
 555         int cpu;
 556         struct vcpu_dispatch_data *disp;
 557
 558         if (!vcpudispatch_stats_on) {
 559                 seq_puts(p, "off\n");
 560                 return 0;
 561         }
 562
 563         for_each_online_cpu(cpu) {
 564                 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
 565                 seq_printf(p, "cpu%d", cpu);
 566                 seq_put_decimal_ull(p, " ", disp->total_disp);
 567                 seq_put_decimal_ull(p, " ", disp->same_cpu_disp);
 568                 seq_put_decimal_ull(p, " ", disp->same_chip_disp);
 569                 seq_put_decimal_ull(p, " ", disp->diff_chip_disp);
 570                 seq_put_decimal_ull(p, " ", disp->far_chip_disp);
 571                 seq_put_decimal_ull(p, " ", disp->numa_home_disp);
 572                 seq_put_decimal_ull(p, " ", disp->numa_remote_disp);
 573                 seq_put_decimal_ull(p, " ", disp->numa_far_disp);
 574                 seq_puts(p, "\n");
 575         }
 576
 577         return 0;
 578 }
 579
 580 static int vcpudispatch_stats_open(struct inode *inode, struct file *file)
 581 {
 582         return single_open(file, vcpudispatch_stats_display, NULL);
 583 }
 584
 585 static const struct proc_ops vcpudispatch_stats_proc_ops = {
 586         .proc_open      = vcpudispatch_stats_open,
 587         .proc_read      = seq_read,
 588         .proc_write     = vcpudispatch_stats_write,
 589         .proc_lseek     = seq_lseek,
 590         .proc_release   = single_release,
 591 };
 592
 593 static ssize_t vcpudispatch_stats_freq_write(struct file *file,
 594                 const char __user *p, size_t count, loff_t *ppos)
 595 {
 596         int rc, freq;
 597         char buf[16];
 598
 599         if (count > 15)
 600                 return -EINVAL;
 601
 602         if (copy_from_user(buf, p, count))
 603                 return -EFAULT;
 604
 605         buf[count] = 0;
 606         rc = kstrtoint(buf, 0, &freq);
 607         if (rc || freq < 1 || freq > HZ) {
 608                 pr_err("vcpudispatch_stats_freq: please specify a frequency between 1 and %d\n",
 609                                 HZ);
 610                 return rc ? rc : -EINVAL;
 611         }
 612
 613         vcpudispatch_stats_freq = freq;
 614
 615         return count;
 616 }
 617
 618 static int vcpudispatch_stats_freq_display(struct seq_file *p, void *v)
 619 {
 620         seq_printf(p, "%d\n", vcpudispatch_stats_freq);
 621         return 0;
 622 }
 623
 624 static int vcpudispatch_stats_freq_open(struct inode *inode, struct file *file)
 625 {
 626         return single_open(file, vcpudispatch_stats_freq_display, NULL);
 627 }
 628
 629 static const struct proc_ops vcpudispatch_stats_freq_proc_ops = {
 630         .proc_open      = vcpudispatch_stats_freq_open,
 631         .proc_read      = seq_read,
 632         .proc_write     = vcpudispatch_stats_freq_write,
 633         .proc_lseek     = seq_lseek,
 634         .proc_release   = single_release,
 635 };
 636
 637 static int __init vcpudispatch_stats_procfs_init(void)
 638 {
 639         /*
 640          * Avoid smp_processor_id while preemptible. All CPUs should have
 641          * the same value for lppaca_shared_proc.
 642          */
 643         preempt_disable();
 644         if (!lppaca_shared_proc(get_lppaca())) {
 645                 preempt_enable();
 646                 return 0;
 647         }
 648         preempt_enable();
 649
 650         if (!proc_create("powerpc/vcpudispatch_stats", 0600, NULL,
 651                                         &vcpudispatch_stats_proc_ops))
 652                 pr_err("vcpudispatch_stats: error creating procfs file\n");
 653         else if (!proc_create("powerpc/vcpudispatch_stats_freq", 0600, NULL,
 654                                         &vcpudispatch_stats_freq_proc_ops))
 655                 pr_err("vcpudispatch_stats_freq: error creating procfs file\n");
 656
 657         return 0;
 658 }
 659
 660 machine_device_initcall(pseries, vcpudispatch_stats_procfs_init);
 661 #endif /* CONFIG_PPC_SPLPAR */
 662
 663 void vpa_init(int cpu)
 664 {
 665         int hwcpu = get_hard_smp_processor_id(cpu);
 666         unsigned long addr;
 667         long ret;
 668
 669         /*
 670          * The spec says it "may be problematic" if CPU x registers the VPA of
 671          * CPU y. We should never do that, but wail if we ever do.
 672          */
 673         WARN_ON(cpu != smp_processor_id());
 674
 675         if (cpu_has_feature(CPU_FTR_ALTIVEC))
 676                 lppaca_of(cpu).vmxregs_in_use = 1;
 677
 678         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 679                 lppaca_of(cpu).ebb_regs_in_use = 1;
 680
 681         addr = __pa(&lppaca_of(cpu));
 682         ret = register_vpa(hwcpu, addr);
 683
 684         if (ret) {
 685                 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
 686                        "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
 687                 return;
 688         }
 689
 690 #ifdef CONFIG_PPC_BOOK3S_64
 691         /*
 692          * PAPR says this feature is SLB-Buffer but firmware never
 693          * reports that.  All SPLPAR support SLB shadow buffer.
 694          */
 695         if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
 696                 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
 697                 ret = register_slb_shadow(hwcpu, addr);
 698                 if (ret)
 699                         pr_err("WARNING: SLB shadow buffer registration for "
 700                                "cpu %d (hw %d) of area %lx failed with %ld\n",
 701                                cpu, hwcpu, addr, ret);
 702         }
 703 #endif /* CONFIG_PPC_BOOK3S_64 */
 704
 705         /*
 706          * Register dispatch trace log, if one has been allocated.
 707          */
 708         register_dtl_buffer(cpu);
 709 }
 710
 711 #ifdef CONFIG_PPC_BOOK3S_64
 712
 713 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 714                                      unsigned long vpn, unsigned long pa,
 715                                      unsigned long rflags, unsigned long vflags,
 716                                      int psize, int apsize, int ssize)
 717 {
 718         unsigned long lpar_rc;
 719         unsigned long flags;
 720         unsigned long slot;
 721         unsigned long hpte_v, hpte_r;
 722
 723         if (!(vflags & HPTE_V_BOLTED))
 724                 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
 725                          "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
 726                          hpte_group, vpn,  pa, rflags, vflags, psize);
 727
 728         hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
 729         hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
 730
 731         if (!(vflags & HPTE_V_BOLTED))
 732                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 733
 734         /* Now fill in the actual HPTE */
 735         /* Set CEC cookie to 0         */
 736         /* Zero page = 0               */
 737         /* I-cache Invalidate = 0      */
 738         /* I-cache synchronize = 0     */
 739         /* Exact = 0                   */
 740         flags = 0;
 741
 742         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 743                 flags |= H_COALESCE_CAND;
 744
 745         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 746         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 747                 pr_devel("Hash table group is full\n");
 748                 return -1;
 749         }
 750
 751         /*
 752          * Since we try and ioremap PHBs we don't own, the pte insert
 753          * will fail. However we must catch the failure in hash_page
 754          * or we will loop forever, so return -2 in this case.
 755          */
 756         if (unlikely(lpar_rc != H_SUCCESS)) {
 757                 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
 758                 return -2;
 759         }
 760         if (!(vflags & HPTE_V_BOLTED))
 761                 pr_devel(" -> slot: %lu\n", slot & 7);
 762
 763         /* Because of iSeries, we have to pass down the secondary
 764          * bucket bit here as well
 765          */
 766         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 767 }
 768
 769 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 770
 771 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 772 {
 773         unsigned long slot_offset;
 774         unsigned long lpar_rc;
 775         int i;
 776         unsigned long dummy1, dummy2;
 777
 778         /* pick a random slot to start at */
 779         slot_offset = mftb() & 0x7;
 780
 781         for (i = 0; i < HPTES_PER_GROUP; i++) {
 782
 783                 /* don't remove a bolted entry */
 784                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 785                                            HPTE_V_BOLTED, &dummy1, &dummy2);
 786                 if (lpar_rc == H_SUCCESS)
 787                         return i;
 788
 789                 /*
 790                  * The test for adjunct partition is performed before the
 791                  * ANDCOND test.  H_RESOURCE may be returned, so we need to
 792                  * check for that as well.
 793                  */
 794                 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
 795
 796                 slot_offset++;
 797                 slot_offset &= 0x7;
 798         }
 799
 800         return -1;
 801 }
 802
 803 static void manual_hpte_clear_all(void)
 804 {
 805         unsigned long size_bytes = 1UL << ppc64_pft_size;
 806         unsigned long hpte_count = size_bytes >> 4;
 807         struct {
 808                 unsigned long pteh;
 809                 unsigned long ptel;
 810         } ptes[4];
 811         long lpar_rc;
 812         unsigned long i, j;
 813
 814         /* Read in batches of 4,
 815          * invalidate only valid entries not in the VRMA
 816          * hpte_count will be a multiple of 4
 817          */
 818         for (i = 0; i < hpte_count; i += 4) {
 819                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 820                 if (lpar_rc != H_SUCCESS) {
 821                         pr_info("Failed to read hash page table at %ld err %ld\n",
 822                                 i, lpar_rc);
 823                         continue;
 824                 }
 825                 for (j = 0; j < 4; j++){
 826                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 827                                 HPTE_V_VRMA_MASK)
 828                                 continue;
 829                         if (ptes[j].pteh & HPTE_V_VALID)
 830                                 plpar_pte_remove_raw(0, i + j, 0,
 831                                         &(ptes[j].pteh), &(ptes[j].ptel));
 832                 }
 833         }
 834 }
 835
 836 static int hcall_hpte_clear_all(void)
 837 {
 838         int rc;
 839
 840         do {
 841                 rc = plpar_hcall_norets(H_CLEAR_HPT);
 842         } while (rc == H_CONTINUE);
 843
 844         return rc;
 845 }
 846
 847 static void pseries_hpte_clear_all(void)
 848 {
 849         int rc;
 850
 851         rc = hcall_hpte_clear_all();
 852         if (rc != H_SUCCESS)
 853                 manual_hpte_clear_all();
 854
 855 #ifdef __LITTLE_ENDIAN__
 856         /*
 857          * Reset exceptions to big endian.
 858          *
 859          * FIXME this is a hack for kexec, we need to reset the exception
 860          * endian before starting the new kernel and this is a convenient place
 861          * to do it.
 862          *
 863          * This is also called on boot when a fadump happens. In that case we
 864          * must not change the exception endian mode.
 865          */
 866         if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
 867                 pseries_big_endian_exceptions();
 868 #endif
 869 }
 870
 871 /*
 872  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 873  * the low 3 bits of flags happen to line up.  So no transform is needed.
 874  * We can probably optimize here and assume the high bits of newpp are
 875  * already zero.  For now I am paranoid.
 876  */
 877 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 878                                        unsigned long newpp,
 879                                        unsigned long vpn,
 880                                        int psize, int apsize,
 881                                        int ssize, unsigned long inv_flags)
 882 {
 883         unsigned long lpar_rc;
 884         unsigned long flags;
 885         unsigned long want_v;
 886
 887         want_v = hpte_encode_avpn(vpn, psize, ssize);
 888
 889         flags = (newpp & 7) | H_AVPN;
 890         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 891                 /* Move pp0 into bit 8 (IBM 55) */
 892                 flags |= (newpp & HPTE_R_PP0) >> 55;
 893
 894         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 895                  want_v, slot, flags, psize);
 896
 897         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 898
 899         if (lpar_rc == H_NOT_FOUND) {
 900                 pr_devel("not found !\n");
 901                 return -1;
 902         }
 903
 904         pr_devel("ok\n");
 905
 906         BUG_ON(lpar_rc != H_SUCCESS);
 907
 908         return 0;
 909 }
 910
 911 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
 912 {
 913         long lpar_rc;
 914         unsigned long i, j;
 915         struct {
 916                 unsigned long pteh;
 917                 unsigned long ptel;
 918         } ptes[4];
 919
 920         for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
 921
 922                 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
 923                 if (lpar_rc != H_SUCCESS) {
 924                         pr_info("Failed to read hash page table at %ld err %ld\n",
 925                                 hpte_group, lpar_rc);
 926                         continue;
 927                 }
 928
 929                 for (j = 0; j < 4; j++) {
 930                         if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
 931                             (ptes[j].pteh & HPTE_V_VALID))
 932                                 return i + j;
 933                 }
 934         }
 935
 936         return -1;
 937 }
 938
 939 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 940 {
 941         long slot;
 942         unsigned long hash;
 943         unsigned long want_v;
 944         unsigned long hpte_group;
 945
 946         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 947         want_v = hpte_encode_avpn(vpn, psize, ssize);
 948
 949         /*
 950          * We try to keep bolted entries always in primary hash
 951          * But in some case we can find them in secondary too.
 952          */
 953         hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 954         slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 955         if (slot < 0) {
 956                 /* Try in secondary */
 957                 hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
 958                 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 959                 if (slot < 0)
 960                         return -1;
 961         }
 962         return hpte_group + slot;
 963 }
 964
 965 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 966                                              unsigned long ea,
 967                                              int psize, int ssize)
 968 {
 969         unsigned long vpn;
 970         unsigned long lpar_rc, slot, vsid, flags;
 971
 972         vsid = get_kernel_vsid(ea, ssize);
 973         vpn = hpt_vpn(ea, vsid, ssize);
 974
 975         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 976         BUG_ON(slot == -1);
 977
 978         flags = newpp & 7;
 979         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 980                 /* Move pp0 into bit 8 (IBM 55) */
 981                 flags |= (newpp & HPTE_R_PP0) >> 55;
 982
 983         lpar_rc = plpar_pte_protect(flags, slot, 0);
 984
 985         BUG_ON(lpar_rc != H_SUCCESS);
 986 }
 987
 988 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
 989                                          int psize, int apsize,
 990                                          int ssize, int local)
 991 {
 992         unsigned long want_v;
 993         unsigned long lpar_rc;
 994         unsigned long dummy1, dummy2;
 995
 996         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
 997                  slot, vpn, psize, local);
 998
 999         want_v = hpte_encode_avpn(vpn, psize, ssize);
1000         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
1001         if (lpar_rc == H_NOT_FOUND)
1002                 return;
1003
1004         BUG_ON(lpar_rc != H_SUCCESS);
1005 }
1006
1007
1008 /*
1009  * As defined in the PAPR's section 14.5.4.1.8
1010  * The control mask doesn't include the returned reference and change bit from
1011  * the processed PTE.
1012  */
1013 #define HBLKR_AVPN              0x0100000000000000UL
1014 #define HBLKR_CTRL_MASK         0xf800000000000000UL
1015 #define HBLKR_CTRL_SUCCESS      0x8000000000000000UL
1016 #define HBLKR_CTRL_ERRNOTFOUND  0x8800000000000000UL
1017 #define HBLKR_CTRL_ERRBUSY      0xa000000000000000UL
1018
1019 /*
1020  * Returned true if we are supporting this block size for the specified segment
1021  * base page size and actual page size.
1022  *
1023  * Currently, we only support 8 size block.
1024  */
1025 static inline bool is_supported_hlbkrm(int bpsize, int psize)
1026 {
1027         return (hblkrm_size[bpsize][psize] == HBLKRM_SUPPORTED_BLOCK_SIZE);
1028 }
1029
1030 /**
1031  * H_BLOCK_REMOVE caller.
1032  * @idx should point to the latest @param entry set with a PTEX.
1033  * If PTE cannot be processed because another CPUs has already locked that
1034  * group, those entries are put back in @param starting at index 1.
1035  * If entries has to be retried and @retry_busy is set to true, these entries
1036  * are retried until success. If @retry_busy is set to false, the returned
1037  * is the number of entries yet to process.
1038  */
1039 static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
1040                                        bool retry_busy)
1041 {
1042         unsigned long i, rc, new_idx;
1043         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1044
1045         if (idx < 2) {
1046                 pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
1047                 return 0;
1048         }
1049 again:
1050         new_idx = 0;
1051         if (idx > PLPAR_HCALL9_BUFSIZE) {
1052                 pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
1053                 idx = PLPAR_HCALL9_BUFSIZE;
1054         } else if (idx < PLPAR_HCALL9_BUFSIZE)
1055                 param[idx] = HBR_END;
1056
1057         rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
1058                           param[0], /* AVA */
1059                           param[1],  param[2],  param[3],  param[4], /* TS0-7 */
1060                           param[5],  param[6],  param[7],  param[8]);
1061         if (rc == H_SUCCESS)
1062                 return 0;
1063
1064         BUG_ON(rc != H_PARTIAL);
1065
1066         /* Check that the unprocessed entries were 'not found' or 'busy' */
1067         for (i = 0; i < idx-1; i++) {
1068                 unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
1069
1070                 if (ctrl == HBLKR_CTRL_ERRBUSY) {
1071                         param[++new_idx] = param[i+1];
1072                         continue;
1073                 }
1074
1075                 BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
1076                        && ctrl != HBLKR_CTRL_ERRNOTFOUND);
1077         }
1078
1079         /*
1080          * If there were entries found busy, retry these entries if requested,
1081          * of if all the entries have to be retried.
1082          */
1083         if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
1084                 idx = new_idx + 1;
1085                 goto again;
1086         }
1087
1088         return new_idx;
1089 }
1090
1091 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1092 /*
1093  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
1094  * to make sure that we avoid bouncing the hypervisor tlbie lock.
1095  */
1096 #define PPC64_HUGE_HPTE_BATCH 12
1097
1098 static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
1099                                       int count, int psize, int ssize)
1100 {
1101         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1102         unsigned long shift, current_vpgb, vpgb;
1103         int i, pix = 0;
1104
1105         shift = mmu_psize_defs[psize].shift;
1106
1107         for (i = 0; i < count; i++) {
1108                 /*
1109                  * Shifting 3 bits more on the right to get a
1110                  * 8 pages aligned virtual addresse.
1111                  */
1112                 vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
1113                 if (!pix || vpgb != current_vpgb) {
1114                         /*
1115                          * Need to start a new 8 pages block, flush
1116                          * the current one if needed.
1117                          */
1118                         if (pix)
1119                                 (void)call_block_remove(pix, param, true);
1120                         current_vpgb = vpgb;
1121                         param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
1122                         pix = 1;
1123                 }
1124
1125                 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
1126                 if (pix == PLPAR_HCALL9_BUFSIZE) {
1127                         pix = call_block_remove(pix, param, false);
1128                         /*
1129                          * pix = 0 means that all the entries were
1130                          * removed, we can start a new block.
1131                          * Otherwise, this means that there are entries
1132                          * to retry, and pix points to latest one, so
1133                          * we should increment it and try to continue
1134                          * the same block.
1135                          */
1136                         if (pix)
1137                                 pix++;
1138                 }
1139         }
1140         if (pix)
1141                 (void)call_block_remove(pix, param, true);
1142 }
1143
1144 static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
1145                                      int count, int psize, int ssize)
1146 {
1147         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1148         int i = 0, pix = 0, rc;
1149
1150         for (i = 0; i < count; i++) {
1151
1152                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1153                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
1154                                                      ssize, 0);
1155                 } else {
1156                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
1157                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
1158                         pix += 2;
1159                         if (pix == 8) {
1160                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
1161                                                   param[0], param[1], param[2],
1162                                                   param[3], param[4], param[5],
1163                                                   param[6], param[7]);
1164                                 BUG_ON(rc != H_SUCCESS);
1165                                 pix = 0;
1166                         }
1167                 }
1168         }
1169         if (pix) {
1170                 param[pix] = HBR_END;
1171                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1172                                   param[2], param[3], param[4], param[5],
1173                                   param[6], param[7]);
1174                 BUG_ON(rc != H_SUCCESS);
1175         }
1176 }
1177
1178 static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
1179                                                       unsigned long *vpn,
1180                                                       int count, int psize,
1181                                                       int ssize)
1182 {
1183         unsigned long flags = 0;
1184         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1185
1186         if (lock_tlbie)
1187                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1188
1189         /* Assuming THP size is 16M */
1190         if (is_supported_hlbkrm(psize, MMU_PAGE_16M))
1191                 hugepage_block_invalidate(slot, vpn, count, psize, ssize);
1192         else
1193                 hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
1194
1195         if (lock_tlbie)
1196                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1197 }
1198
1199 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1200                                              unsigned long addr,
1201                                              unsigned char *hpte_slot_array,
1202                                              int psize, int ssize, int local)
1203 {
1204         int i, index = 0;
1205         unsigned long s_addr = addr;
1206         unsigned int max_hpte_count, valid;
1207         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
1208         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
1209         unsigned long shift, hidx, vpn = 0, hash, slot;
1210
1211         shift = mmu_psize_defs[psize].shift;
1212         max_hpte_count = 1U << (PMD_SHIFT - shift);
1213
1214         for (i = 0; i < max_hpte_count; i++) {
1215                 valid = hpte_valid(hpte_slot_array, i);
1216                 if (!valid)
1217                         continue;
1218                 hidx =  hpte_hash_index(hpte_slot_array, i);
1219
1220                 /* get the vpn */
1221                 addr = s_addr + (i * (1ul << shift));
1222                 vpn = hpt_vpn(addr, vsid, ssize);
1223                 hash = hpt_hash(vpn, shift, ssize);
1224                 if (hidx & _PTEIDX_SECONDARY)
1225                         hash = ~hash;
1226
1227                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1228                 slot += hidx & _PTEIDX_GROUP_IX;
1229
1230                 slot_array[index] = slot;
1231                 vpn_array[index] = vpn;
1232                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
1233                         /*
1234                          * Now do a bluk invalidate
1235                          */
1236                         __pSeries_lpar_hugepage_invalidate(slot_array,
1237                                                            vpn_array,
1238                                                            PPC64_HUGE_HPTE_BATCH,
1239                                                            psize, ssize);
1240                         index = 0;
1241                 } else
1242                         index++;
1243         }
1244         if (index)
1245                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
1246                                                    index, psize, ssize);
1247 }
1248 #else
1249 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1250                                              unsigned long addr,
1251                                              unsigned char *hpte_slot_array,
1252                                              int psize, int ssize, int local)
1253 {
1254         WARN(1, "%s called without THP support\n", __func__);
1255 }
1256 #endif
1257
1258 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
1259                                           int psize, int ssize)
1260 {
1261         unsigned long vpn;
1262         unsigned long slot, vsid;
1263
1264         vsid = get_kernel_vsid(ea, ssize);
1265         vpn = hpt_vpn(ea, vsid, ssize);
1266
1267         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1268         if (slot == -1)
1269                 return -ENOENT;
1270
1271         /*
1272          * lpar doesn't use the passed actual page size
1273          */
1274         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
1275         return 0;
1276 }
1277
1278
1279 static inline unsigned long compute_slot(real_pte_t pte,
1280                                          unsigned long vpn,
1281                                          unsigned long index,
1282                                          unsigned long shift,
1283                                          int ssize)
1284 {
1285         unsigned long slot, hash, hidx;
1286
1287         hash = hpt_hash(vpn, shift, ssize);
1288         hidx = __rpte_to_hidx(pte, index);
1289         if (hidx & _PTEIDX_SECONDARY)
1290                 hash = ~hash;
1291         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1292         slot += hidx & _PTEIDX_GROUP_IX;
1293         return slot;
1294 }
1295
1296 /**
1297  * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
1298  * "all within the same naturally aligned 8 page virtual address block".
1299  */
1300 static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
1301                             unsigned long *param)
1302 {
1303         unsigned long vpn;
1304         unsigned long i, pix = 0;
1305         unsigned long index, shift, slot, current_vpgb, vpgb;
1306         real_pte_t pte;
1307         int psize, ssize;
1308
1309         psize = batch->psize;
1310         ssize = batch->ssize;
1311
1312         for (i = 0; i < number; i++) {
1313                 vpn = batch->vpn[i];
1314                 pte = batch->pte[i];
1315                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1316                         /*
1317                          * Shifting 3 bits more on the right to get a
1318                          * 8 pages aligned virtual addresse.
1319                          */
1320                         vpgb = (vpn >> (shift - VPN_SHIFT + 3));
1321                         if (!pix || vpgb != current_vpgb) {
1322                                 /*
1323                                  * Need to start a new 8 pages block, flush
1324                                  * the current one if needed.
1325                                  */
1326                                 if (pix)
1327                                         (void)call_block_remove(pix, param,
1328                                                                 true);
1329                                 current_vpgb = vpgb;
1330                                 param[0] = hpte_encode_avpn(vpn, psize,
1331                                                             ssize);
1332                                 pix = 1;
1333                         }
1334
1335                         slot = compute_slot(pte, vpn, index, shift, ssize);
1336                         param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
1337
1338                         if (pix == PLPAR_HCALL9_BUFSIZE) {
1339                                 pix = call_block_remove(pix, param, false);
1340                                 /*
1341                                  * pix = 0 means that all the entries were
1342                                  * removed, we can start a new block.
1343                                  * Otherwise, this means that there are entries
1344                                  * to retry, and pix points to latest one, so
1345                                  * we should increment it and try to continue
1346                                  * the same block.
1347                                  */
1348                                 if (pix)
1349                                         pix++;
1350                         }
1351                 } pte_iterate_hashed_end();
1352         }
1353
1354         if (pix)
1355                 (void)call_block_remove(pix, param, true);
1356 }
1357
1358 /*
1359  * TLB Block Invalidate Characteristics
1360  *
1361  * These characteristics define the size of the block the hcall H_BLOCK_REMOVE
1362  * is able to process for each couple segment base page size, actual page size.
1363  *
1364  * The ibm,get-system-parameter properties is returning a buffer with the
1365  * following layout:
1366  *
1367  * [ 2 bytes size of the RTAS buffer (excluding these 2 bytes) ]
1368  * -----------------
1369  * TLB Block Invalidate Specifiers:
1370  * [ 1 byte LOG base 2 of the TLB invalidate block size being specified ]
1371  * [ 1 byte Number of page sizes (N) that are supported for the specified
1372  *          TLB invalidate block size ]
1373  * [ 1 byte Encoded segment base page size and actual page size
1374  *          MSB=0 means 4k segment base page size and actual page size
1375  *          MSB=1 the penc value in mmu_psize_def ]
1376  * ...
1377  * -----------------
1378  * Next TLB Block Invalidate Specifiers...
1379  * -----------------
1380  * [ 0 ]
1381  */
1382 static inline void set_hblkrm_bloc_size(int bpsize, int psize,
1383                                         unsigned int block_size)
1384 {
1385         if (block_size > hblkrm_size[bpsize][psize])
1386                 hblkrm_size[bpsize][psize] = block_size;
1387 }
1388
1389 /*
1390  * Decode the Encoded segment base page size and actual page size.
1391  * PAPR specifies:
1392  *   - bit 7 is the L bit
1393  *   - bits 0-5 are the penc value
1394  * If the L bit is 0, this means 4K segment base page size and actual page size
1395  * otherwise the penc value should be read.
1396  */
1397 #define HBLKRM_L_MASK           0x80
1398 #define HBLKRM_PENC_MASK        0x3f
1399 static inline void __init check_lp_set_hblkrm(unsigned int lp,
1400                                               unsigned int block_size)
1401 {
1402         unsigned int bpsize, psize;
1403
1404         /* First, check the L bit, if not set, this means 4K */
1405         if ((lp & HBLKRM_L_MASK) == 0) {
1406                 set_hblkrm_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
1407                 return;
1408         }
1409
1410         lp &= HBLKRM_PENC_MASK;
1411         for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
1412                 struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
1413
1414                 for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1415                         if (def->penc[psize] == lp) {
1416                                 set_hblkrm_bloc_size(bpsize, psize, block_size);
1417                                 return;
1418                         }
1419                 }
1420         }
1421 }
1422
1423 #define SPLPAR_TLB_BIC_TOKEN            50
1424
1425 /*
1426  * The size of the TLB Block Invalidate Characteristics is variable. But at the
1427  * maximum it will be the number of possible page sizes *2 + 10 bytes.
1428  * Currently MMU_PAGE_COUNT is 16, which means 42 bytes. Use a cache line size
1429  * (128 bytes) for the buffer to get plenty of space.
1430  */
1431 #define SPLPAR_TLB_BIC_MAXLENGTH        128
1432
1433 void __init pseries_lpar_read_hblkrm_characteristics(void)
1434 {
1435         unsigned char local_buffer[SPLPAR_TLB_BIC_MAXLENGTH];
1436         int call_status, len, idx, bpsize;
1437
1438         if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
1439                 return;
1440
1441         spin_lock(&rtas_data_buf_lock);
1442         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
1443         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
1444                                 NULL,
1445                                 SPLPAR_TLB_BIC_TOKEN,
1446                                 __pa(rtas_data_buf),
1447                                 RTAS_DATA_BUF_SIZE);
1448         memcpy(local_buffer, rtas_data_buf, SPLPAR_TLB_BIC_MAXLENGTH);
1449         local_buffer[SPLPAR_TLB_BIC_MAXLENGTH - 1] = '\0';
1450         spin_unlock(&rtas_data_buf_lock);
1451
1452         if (call_status != 0) {
1453                 pr_warn("%s %s Error calling get-system-parameter (0x%x)\n",
1454                         __FILE__, __func__, call_status);
1455                 return;
1456         }
1457
1458         /*
1459          * The first two (2) bytes of the data in the buffer are the length of
1460          * the returned data, not counting these first two (2) bytes.
1461          */
1462         len = be16_to_cpu(*((u16 *)local_buffer)) + 2;
1463         if (len > SPLPAR_TLB_BIC_MAXLENGTH) {
1464                 pr_warn("%s too large returned buffer %d", __func__, len);
1465                 return;
1466         }
1467
1468         idx = 2;
1469         while (idx < len) {
1470                 u8 block_shift = local_buffer[idx++];
1471                 u32 block_size;
1472                 unsigned int npsize;
1473
1474                 if (!block_shift)
1475                         break;
1476
1477                 block_size = 1 << block_shift;
1478
1479                 for (npsize = local_buffer[idx++];
1480                      npsize > 0 && idx < len; npsize--)
1481                         check_lp_set_hblkrm((unsigned int) local_buffer[idx++],
1482                                             block_size);
1483         }
1484
1485         for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
1486                 for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
1487                         if (hblkrm_size[bpsize][idx])
1488                                 pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
1489                                         bpsize, idx, hblkrm_size[bpsize][idx]);
1490 }
1491
1492 /*
1493  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
1494  * lock.
1495  */
1496 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
1497 {
1498         unsigned long vpn;
1499         unsigned long i, pix, rc;
1500         unsigned long flags = 0;
1501         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
1502         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1503         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1504         unsigned long index, shift, slot;
1505         real_pte_t pte;
1506         int psize, ssize;
1507
1508         if (lock_tlbie)
1509                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1510
1511         if (is_supported_hlbkrm(batch->psize, batch->psize)) {
1512                 do_block_remove(number, batch, param);
1513                 goto out;
1514         }
1515
1516         psize = batch->psize;
1517         ssize = batch->ssize;
1518         pix = 0;
1519         for (i = 0; i < number; i++) {
1520                 vpn = batch->vpn[i];
1521                 pte = batch->pte[i];
1522                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1523                         slot = compute_slot(pte, vpn, index, shift, ssize);
1524                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1525                                 /*
1526                                  * lpar doesn't use the passed actual page size
1527                                  */
1528                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
1529                                                              0, ssize, local);
1530                         } else {
1531                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
1532                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
1533                                                                 ssize);
1534                                 pix += 2;
1535                                 if (pix == 8) {
1536                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
1537                                                 param[0], param[1], param[2],
1538                                                 param[3], param[4], param[5],
1539                                                 param[6], param[7]);
1540                                         BUG_ON(rc != H_SUCCESS);
1541                                         pix = 0;
1542                                 }
1543                         }
1544                 } pte_iterate_hashed_end();
1545         }
1546         if (pix) {
1547                 param[pix] = HBR_END;
1548                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1549                                   param[2], param[3], param[4], param[5],
1550                                   param[6], param[7]);
1551                 BUG_ON(rc != H_SUCCESS);
1552         }
1553
1554 out:
1555         if (lock_tlbie)
1556                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1557 }
1558
1559 static int __init disable_bulk_remove(char *str)
1560 {
1561         if (strcmp(str, "off") == 0 &&
1562             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1563                 pr_info("Disabling BULK_REMOVE firmware feature");
1564                 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
1565         }
1566         return 1;
1567 }
1568
1569 __setup("bulk_remove=", disable_bulk_remove);
1570
1571 #define HPT_RESIZE_TIMEOUT      10000 /* ms */
1572
1573 struct hpt_resize_state {
1574         unsigned long shift;
1575         int commit_rc;
1576 };
1577
1578 static int pseries_lpar_resize_hpt_commit(void *data)
1579 {
1580         struct hpt_resize_state *state = data;
1581
1582         state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
1583         if (state->commit_rc != H_SUCCESS)
1584                 return -EIO;
1585
1586         /* Hypervisor has transitioned the HTAB, update our globals */
1587         ppc64_pft_size = state->shift;
1588         htab_size_bytes = 1UL << ppc64_pft_size;
1589         htab_hash_mask = (htab_size_bytes >> 7) - 1;
1590
1591         return 0;
1592 }
1593
1594 /*
1595  * Must be called in process context. The caller must hold the
1596  * cpus_lock.
1597  */
1598 static int pseries_lpar_resize_hpt(unsigned long shift)
1599 {
1600         struct hpt_resize_state state = {
1601                 .shift = shift,
1602                 .commit_rc = H_FUNCTION,
1603         };
1604         unsigned int delay, total_delay = 0;
1605         int rc;
1606         ktime_t t0, t1, t2;
1607
1608         might_sleep();
1609
1610         if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1611                 return -ENODEV;
1612
1613         pr_info("Attempting to resize HPT to shift %lu\n", shift);
1614
1615         t0 = ktime_get();
1616
1617         rc = plpar_resize_hpt_prepare(0, shift);
1618         while (H_IS_LONG_BUSY(rc)) {
1619                 delay = get_longbusy_msecs(rc);
1620                 total_delay += delay;
1621                 if (total_delay > HPT_RESIZE_TIMEOUT) {
1622                         /* prepare with shift==0 cancels an in-progress resize */
1623                         rc = plpar_resize_hpt_prepare(0, 0);
1624                         if (rc != H_SUCCESS)
1625                                 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
1626                                        rc);
1627                         return -ETIMEDOUT;
1628                 }
1629                 msleep(delay);
1630                 rc = plpar_resize_hpt_prepare(0, shift);
1631         };
1632
1633         switch (rc) {
1634         case H_SUCCESS:
1635                 /* Continue on */
1636                 break;
1637
1638         case H_PARAMETER:
1639                 pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
1640                 return -EINVAL;
1641         case H_RESOURCE:
1642                 pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
1643                 return -EPERM;
1644         default:
1645                 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
1646                 return -EIO;
1647         }
1648
1649         t1 = ktime_get();
1650
1651         rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
1652                                      &state, NULL);
1653
1654         t2 = ktime_get();
1655
1656         if (rc != 0) {
1657                 switch (state.commit_rc) {
1658                 case H_PTEG_FULL:
1659                         return -ENOSPC;
1660
1661                 default:
1662                         pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
1663                                 state.commit_rc);
1664                         return -EIO;
1665                 };
1666         }
1667
1668         pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
1669                 shift, (long long) ktime_ms_delta(t1, t0),
1670                 (long long) ktime_ms_delta(t2, t1));
1671
1672         return 0;
1673 }
1674
1675 static int pseries_lpar_register_process_table(unsigned long base,
1676                         unsigned long page_size, unsigned long table_size)
1677 {
1678         long rc;
1679         unsigned long flags = 0;
1680
1681         if (table_size)
1682                 flags |= PROC_TABLE_NEW;
1683         if (radix_enabled())
1684                 flags |= PROC_TABLE_RADIX | PROC_TABLE_GTSE;
1685         else
1686                 flags |= PROC_TABLE_HPT_SLB;
1687         for (;;) {
1688                 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
1689                                         page_size, table_size);
1690                 if (!H_IS_LONG_BUSY(rc))
1691                         break;
1692                 mdelay(get_longbusy_msecs(rc));
1693         }
1694         if (rc != H_SUCCESS) {
1695                 pr_err("Failed to register process table (rc=%ld)\n", rc);
1696                 BUG();
1697         }
1698         return rc;
1699 }
1700
1701 void __init hpte_init_pseries(void)
1702 {
1703         mmu_hash_ops.hpte_invalidate     = pSeries_lpar_hpte_invalidate;
1704         mmu_hash_ops.hpte_updatepp       = pSeries_lpar_hpte_updatepp;
1705         mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
1706         mmu_hash_ops.hpte_insert         = pSeries_lpar_hpte_insert;
1707         mmu_hash_ops.hpte_remove         = pSeries_lpar_hpte_remove;
1708         mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
1709         mmu_hash_ops.flush_hash_range    = pSeries_lpar_flush_hash_range;
1710         mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
1711         mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
1712
1713         if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1714                 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
1715
1716         /*
1717          * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
1718          * to inform the hypervisor that we wish to use the HPT.
1719          */
1720         if (cpu_has_feature(CPU_FTR_ARCH_300))
1721                 pseries_lpar_register_process_table(0, 0, 0);
1722 }
1723
1724 void radix_init_pseries(void)
1725 {
1726         pr_info("Using radix MMU under hypervisor\n");
1727
1728         pseries_lpar_register_process_table(__pa(process_tb),
1729                                                 0, PRTB_SIZE_SHIFT - 12);
1730 }
1731
1732 #ifdef CONFIG_PPC_SMLPAR
1733 #define CMO_FREE_HINT_DEFAULT 1
1734 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
1735
1736 static int __init cmo_free_hint(char *str)
1737 {
1738         char *parm;
1739         parm = strstrip(str);
1740
1741         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
1742                 pr_info("%s: CMO free page hinting is not active.\n", __func__);
1743                 cmo_free_hint_flag = 0;
1744                 return 1;
1745         }
1746
1747         cmo_free_hint_flag = 1;
1748         pr_info("%s: CMO free page hinting is active.\n", __func__);
1749
1750         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1751                 return 1;
1752
1753         return 0;
1754 }
1755
1756 __setup("cmo_free_hint=", cmo_free_hint);
1757
1758 static void pSeries_set_page_state(struct page *page, int order,
1759                                    unsigned long state)
1760 {
1761         int i, j;
1762         unsigned long cmo_page_sz, addr;
1763
1764         cmo_page_sz = cmo_get_page_size();
1765         addr = __pa((unsigned long)page_address(page));
1766
1767         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1768                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1769                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1770         }
1771 }
1772
1773 void arch_free_page(struct page *page, int order)
1774 {
1775         if (radix_enabled())
1776                 return;
1777         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1778                 return;
1779
1780         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1781 }
1782 EXPORT_SYMBOL(arch_free_page);
1783
1784 #endif /* CONFIG_PPC_SMLPAR */
1785 #endif /* CONFIG_PPC_BOOK3S_64 */
1786
1787 #ifdef CONFIG_TRACEPOINTS
1788 #ifdef CONFIG_JUMP_LABEL
1789 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1790
1791 int hcall_tracepoint_regfunc(void)
1792 {
1793         static_key_slow_inc(&hcall_tracepoint_key);
1794         return 0;
1795 }
1796
1797 void hcall_tracepoint_unregfunc(void)
1798 {
1799         static_key_slow_dec(&hcall_tracepoint_key);
1800 }
1801 #else
1802 /*
1803  * We optimise our hcall path by placing hcall_tracepoint_refcount
1804  * directly in the TOC so we can check if the hcall tracepoints are
1805  * enabled via a single load.
1806  */
1807
1808 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1809 extern long hcall_tracepoint_refcount;
1810
1811 int hcall_tracepoint_regfunc(void)
1812 {
1813         hcall_tracepoint_refcount++;
1814         return 0;
1815 }
1816
1817 void hcall_tracepoint_unregfunc(void)
1818 {
1819         hcall_tracepoint_refcount--;
1820 }
1821 #endif
1822
1823 /*
1824  * Since the tracing code might execute hcalls we need to guard against
1825  * recursion. One example of this are spinlocks calling H_YIELD on
1826  * shared processor partitions.
1827  */
1828 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1829
1830
1831 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1832 {
1833         unsigned long flags;
1834         unsigned int *depth;
1835
1836         /*
1837          * We cannot call tracepoints inside RCU idle regions which
1838          * means we must not trace H_CEDE.
1839          */
1840         if (opcode == H_CEDE)
1841                 return;
1842
1843         local_irq_save(flags);
1844
1845         depth = this_cpu_ptr(&hcall_trace_depth);
1846
1847         if (*depth)
1848                 goto out;
1849
1850         (*depth)++;
1851         preempt_disable();
1852         trace_hcall_entry(opcode, args);
1853         (*depth)--;
1854
1855 out:
1856         local_irq_restore(flags);
1857 }
1858
1859 void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1860 {
1861         unsigned long flags;
1862         unsigned int *depth;
1863
1864         if (opcode == H_CEDE)
1865                 return;
1866
1867         local_irq_save(flags);
1868
1869         depth = this_cpu_ptr(&hcall_trace_depth);
1870
1871         if (*depth)
1872                 goto out;
1873
1874         (*depth)++;
1875         trace_hcall_exit(opcode, retval, retbuf);
1876         preempt_enable();
1877         (*depth)--;
1878
1879 out:
1880         local_irq_restore(flags);
1881 }
1882 #endif
1883
1884 /**
1885  * h_get_mpp
1886  * H_GET_MPP hcall returns info in 7 parms
1887  */
1888 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
1889 {
1890         int rc;
1891         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1892
1893         rc = plpar_hcall9(H_GET_MPP, retbuf);
1894
1895         mpp_data->entitled_mem = retbuf[0];
1896         mpp_data->mapped_mem = retbuf[1];
1897
1898         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1899         mpp_data->pool_num = retbuf[2] & 0xffff;
1900
1901         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1902         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1903         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1904
1905         mpp_data->pool_size = retbuf[4];
1906         mpp_data->loan_request = retbuf[5];
1907         mpp_data->backing_mem = retbuf[6];
1908
1909         return rc;
1910 }
1911 EXPORT_SYMBOL(h_get_mpp);
1912
1913 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1914 {
1915         int rc;
1916         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1917
1918         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1919
1920         mpp_x_data->coalesced_bytes = retbuf[0];
1921         mpp_x_data->pool_coalesced_bytes = retbuf[1];
1922         mpp_x_data->pool_purr_cycles = retbuf[2];
1923         mpp_x_data->pool_spurr_cycles = retbuf[3];
1924
1925         return rc;
1926 }
1927
1928 static unsigned long vsid_unscramble(unsigned long vsid, int ssize)
1929 {
1930         unsigned long protovsid;
1931         unsigned long va_bits = VA_BITS;
1932         unsigned long modinv, vsid_modulus;
1933         unsigned long max_mod_inv, tmp_modinv;
1934
1935         if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1936                 va_bits = 65;
1937
1938         if (ssize == MMU_SEGSIZE_256M) {
1939                 modinv = VSID_MULINV_256M;
1940                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1941         } else {
1942                 modinv = VSID_MULINV_1T;
1943                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1944         }
1945
1946         /*
1947          * vsid outside our range.
1948          */
1949         if (vsid >= vsid_modulus)
1950                 return 0;
1951
1952         /*
1953          * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1954          * and vsid = (protovsid * x) % vsid_modulus, then we say:
1955          *   protovsid = (vsid * modinv) % vsid_modulus
1956          */
1957
1958         /* Check if (vsid * modinv) overflow (63 bits) */
1959         max_mod_inv = 0x7fffffffffffffffull / vsid;
1960         if (modinv < max_mod_inv)
1961                 return (vsid * modinv) % vsid_modulus;
1962
1963         tmp_modinv = modinv/max_mod_inv;
1964         modinv %= max_mod_inv;
1965
1966         protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1967         protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1968
1969         return protovsid;
1970 }
1971
1972 static int __init reserve_vrma_context_id(void)
1973 {
1974         unsigned long protovsid;
1975
1976         /*
1977          * Reserve context ids which map to reserved virtual addresses. For now
1978          * we only reserve the context id which maps to the VRMA VSID. We ignore
1979          * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1980          * enable adjunct support via the "ibm,client-architecture-support"
1981          * interface.
1982          */
1983         protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1984         hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1985         return 0;
1986 }
1987 machine_device_initcall(pseries, reserve_vrma_context_id);
1988
1989 #ifdef CONFIG_DEBUG_FS
1990 /* debugfs file interface for vpa data */
1991 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1992                               loff_t *pos)
1993 {
1994         int cpu = (long)filp->private_data;
1995         struct lppaca *lppaca = &lppaca_of(cpu);
1996
1997         return simple_read_from_buffer(buf, len, pos, lppaca,
1998                                 sizeof(struct lppaca));
1999 }
2000
2001 static const struct file_operations vpa_fops = {
2002         .open           = simple_open,
2003         .read           = vpa_file_read,
2004         .llseek         = default_llseek,
2005 };
2006
2007 static int __init vpa_debugfs_init(void)
2008 {
2009         char name[16];
2010         long i;
2011         struct dentry *vpa_dir;
2012
2013         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
2014                 return 0;
2015
2016         vpa_dir = debugfs_create_dir("vpa", powerpc_debugfs_root);
2017
2018         /* set up the per-cpu vpa file*/
2019         for_each_possible_cpu(i) {
2020                 sprintf(name, "cpu-%ld", i);
2021                 debugfs_create_file(name, 0400, vpa_dir, (void *)i, &vpa_fops);
2022         }
2023
2024         return 0;
2025 }
2026 machine_arch_initcall(pseries, vpa_debugfs_init);
2027 #endif /* CONFIG_DEBUG_FS */