arch/powerpc/platforms/pseries/setup.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  64-bit pSeries and RS/6000 setup code.
   4  *
   5  *  Copyright (C) 1995  Linus Torvalds
   6  *  Adapted from 'alpha' version by Gary Thomas
   7  *  Modified by Cort Dougan (cort@cs.nmt.edu)
   8  *  Modified by PPC64 Team, IBM Corp
   9  */
  10
  11 /*
  12  * bootup setup stuff..
  13  */
  14
  15 #include <linux/cpu.h>
  16 #include <linux/errno.h>
  17 #include <linux/sched.h>
  18 #include <linux/kernel.h>
  19 #include <linux/mm.h>
  20 #include <linux/stddef.h>
  21 #include <linux/unistd.h>
  22 #include <linux/user.h>
  23 #include <linux/tty.h>
  24 #include <linux/major.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/reboot.h>
  27 #include <linux/init.h>
  28 #include <linux/ioport.h>
  29 #include <linux/console.h>
  30 #include <linux/pci.h>
  31 #include <linux/utsname.h>
  32 #include <linux/adb.h>
  33 #include <linux/export.h>
  34 #include <linux/delay.h>
  35 #include <linux/irq.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/root_dev.h>
  38 #include <linux/of.h>
  39 #include <linux/of_pci.h>
  40 #include <linux/memblock.h>
  41 #include <linux/swiotlb.h>
  42
  43 #include <asm/mmu.h>
  44 #include <asm/processor.h>
  45 #include <asm/io.h>
  46 #include <asm/prom.h>
  47 #include <asm/rtas.h>
  48 #include <asm/pci-bridge.h>
  49 #include <asm/iommu.h>
  50 #include <asm/dma.h>
  51 #include <asm/machdep.h>
  52 #include <asm/irq.h>
  53 #include <asm/time.h>
  54 #include <asm/nvram.h>
  55 #include <asm/pmc.h>
  56 #include <asm/xics.h>
  57 #include <asm/xive.h>
  58 #include <asm/ppc-pci.h>
  59 #include <asm/i8259.h>
  60 #include <asm/udbg.h>
  61 #include <asm/smp.h>
  62 #include <asm/firmware.h>
  63 #include <asm/eeh.h>
  64 #include <asm/reg.h>
  65 #include <asm/plpar_wrappers.h>
  66 #include <asm/kexec.h>
  67 #include <asm/isa-bridge.h>
  68 #include <asm/security_features.h>
  69 #include <asm/asm-const.h>
  70 #include <asm/idle.h>
  71 #include <asm/swiotlb.h>
  72 #include <asm/svm.h>
  73 #include <asm/dtl.h>
  74
  75 #include "pseries.h"
  76 #include "../../../../drivers/pci/pci.h"
  77
  78 DEFINE_STATIC_KEY_FALSE(shared_processor);
  79 EXPORT_SYMBOL_GPL(shared_processor);
  80
  81 int CMO_PrPSP = -1;
  82 int CMO_SecPSP = -1;
  83 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
  84 EXPORT_SYMBOL(CMO_PageSize);
  85
  86 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
  87 int ibm_nmi_interlock_token;
  88
  89 static void pSeries_show_cpuinfo(struct seq_file *m)
  90 {
  91         struct device_node *root;
  92         const char *model = "";
  93
  94         root = of_find_node_by_path("/");
  95         if (root)
  96                 model = of_get_property(root, "model", NULL);
  97         seq_printf(m, "machine\t\t: CHRP %s\n", model);
  98         of_node_put(root);
  99         if (radix_enabled())
 100                 seq_printf(m, "MMU\t\t: Radix\n");
 101         else
 102                 seq_printf(m, "MMU\t\t: Hash\n");
 103 }
 104
 105 /* Initialize firmware assisted non-maskable interrupts if
 106  * the firmware supports this feature.
 107  */
 108 static void __init fwnmi_init(void)
 109 {
 110         unsigned long system_reset_addr, machine_check_addr;
 111         u8 *mce_data_buf;
 112         unsigned int i;
 113         int nr_cpus = num_possible_cpus();
 114 #ifdef CONFIG_PPC_BOOK3S_64
 115         struct slb_entry *slb_ptr;
 116         size_t size;
 117 #endif
 118         int ibm_nmi_register_token;
 119
 120         ibm_nmi_register_token = rtas_token("ibm,nmi-register");
 121         if (ibm_nmi_register_token == RTAS_UNKNOWN_SERVICE)
 122                 return;
 123
 124         ibm_nmi_interlock_token = rtas_token("ibm,nmi-interlock");
 125         if (WARN_ON(ibm_nmi_interlock_token == RTAS_UNKNOWN_SERVICE))
 126                 return;
 127
 128         /* If the kernel's not linked at zero we point the firmware at low
 129          * addresses anyway, and use a trampoline to get to the real code. */
 130         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
 131         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
 132
 133         if (0 == rtas_call(ibm_nmi_register_token, 2, 1, NULL,
 134                            system_reset_addr, machine_check_addr))
 135                 fwnmi_active = 1;
 136
 137         /*
 138          * Allocate a chunk for per cpu buffer to hold rtas errorlog.
 139          * It will be used in real mode mce handler, hence it needs to be
 140          * below RMA.
 141          */
 142         mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
 143                                         RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
 144                                         ppc64_rma_size, NUMA_NO_NODE);
 145         if (!mce_data_buf)
 146                 panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
 147                       RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
 148
 149         for_each_possible_cpu(i) {
 150                 paca_ptrs[i]->mce_data_buf = mce_data_buf +
 151                                                 (RTAS_ERROR_LOG_MAX * i);
 152         }
 153
 154 #ifdef CONFIG_PPC_BOOK3S_64
 155         if (!radix_enabled()) {
 156                 /* Allocate per cpu area to save old slb contents during MCE */
 157                 size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
 158                 slb_ptr = memblock_alloc_try_nid_raw(size,
 159                                 sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
 160                                 ppc64_rma_size, NUMA_NO_NODE);
 161                 if (!slb_ptr)
 162                         panic("Failed to allocate %zu bytes below %pa for slb area\n",
 163                               size, &ppc64_rma_size);
 164
 165                 for_each_possible_cpu(i)
 166                         paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
 167         }
 168 #endif
 169 }
 170
 171 static void pseries_8259_cascade(struct irq_desc *desc)
 172 {
 173         struct irq_chip *chip = irq_desc_get_chip(desc);
 174         unsigned int cascade_irq = i8259_irq();
 175
 176         if (cascade_irq)
 177                 generic_handle_irq(cascade_irq);
 178
 179         chip->irq_eoi(&desc->irq_data);
 180 }
 181
 182 static void __init pseries_setup_i8259_cascade(void)
 183 {
 184         struct device_node *np, *old, *found = NULL;
 185         unsigned int cascade;
 186         const u32 *addrp;
 187         unsigned long intack = 0;
 188         int naddr;
 189
 190         for_each_node_by_type(np, "interrupt-controller") {
 191                 if (of_device_is_compatible(np, "chrp,iic")) {
 192                         found = np;
 193                         break;
 194                 }
 195         }
 196
 197         if (found == NULL) {
 198                 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
 199                 return;
 200         }
 201
 202         cascade = irq_of_parse_and_map(found, 0);
 203         if (!cascade) {
 204                 printk(KERN_ERR "pic: failed to map cascade interrupt");
 205                 return;
 206         }
 207         pr_debug("pic: cascade mapped to irq %d\n", cascade);
 208
 209         for (old = of_node_get(found); old != NULL ; old = np) {
 210                 np = of_get_parent(old);
 211                 of_node_put(old);
 212                 if (np == NULL)
 213                         break;
 214                 if (!of_node_name_eq(np, "pci"))
 215                         continue;
 216                 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
 217                 if (addrp == NULL)
 218                         continue;
 219                 naddr = of_n_addr_cells(np);
 220                 intack = addrp[naddr-1];
 221                 if (naddr > 1)
 222                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
 223         }
 224         if (intack)
 225                 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
 226         i8259_init(found, intack);
 227         of_node_put(found);
 228         irq_set_chained_handler(cascade, pseries_8259_cascade);
 229 }
 230
 231 static void __init pseries_init_irq(void)
 232 {
 233         /* Try using a XIVE if available, otherwise use a XICS */
 234         if (!xive_spapr_init()) {
 235                 xics_init();
 236                 pseries_setup_i8259_cascade();
 237         }
 238 }
 239
 240 static void pseries_lpar_enable_pmcs(void)
 241 {
 242         unsigned long set, reset;
 243
 244         set = 1UL << 63;
 245         reset = 0;
 246         plpar_hcall_norets(H_PERFMON, set, reset);
 247 }
 248
 249 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
 250 {
 251         struct of_reconfig_data *rd = data;
 252         struct device_node *parent, *np = rd->dn;
 253         struct pci_dn *pdn;
 254         int err = NOTIFY_OK;
 255
 256         switch (action) {
 257         case OF_RECONFIG_ATTACH_NODE:
 258                 parent = of_get_parent(np);
 259                 pdn = parent ? PCI_DN(parent) : NULL;
 260                 if (pdn)
 261                         pci_add_device_node_info(pdn->phb, np);
 262
 263                 of_node_put(parent);
 264                 break;
 265         case OF_RECONFIG_DETACH_NODE:
 266                 pdn = PCI_DN(np);
 267                 if (pdn)
 268                         list_del(&pdn->list);
 269                 break;
 270         default:
 271                 err = NOTIFY_DONE;
 272                 break;
 273         }
 274         return err;
 275 }
 276
 277 static struct notifier_block pci_dn_reconfig_nb = {
 278         .notifier_call = pci_dn_reconfig_notifier,
 279 };
 280
 281 struct kmem_cache *dtl_cache;
 282
 283 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 284 /*
 285  * Allocate space for the dispatch trace log for all possible cpus
 286  * and register the buffers with the hypervisor.  This is used for
 287  * computing time stolen by the hypervisor.
 288  */
 289 static int alloc_dispatch_logs(void)
 290 {
 291         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 292                 return 0;
 293
 294         if (!dtl_cache)
 295                 return 0;
 296
 297         alloc_dtl_buffers(0);
 298
 299         /* Register the DTL for the current (boot) cpu */
 300         register_dtl_buffer(smp_processor_id());
 301
 302         return 0;
 303 }
 304 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 305 static inline int alloc_dispatch_logs(void)
 306 {
 307         return 0;
 308 }
 309 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 310
 311 static int alloc_dispatch_log_kmem_cache(void)
 312 {
 313         void (*ctor)(void *) = get_dtl_cache_ctor();
 314
 315         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
 316                                                 DISPATCH_LOG_BYTES, 0, ctor);
 317         if (!dtl_cache) {
 318                 pr_warn("Failed to create dispatch trace log buffer cache\n");
 319                 pr_warn("Stolen time statistics will be unreliable\n");
 320                 return 0;
 321         }
 322
 323         return alloc_dispatch_logs();
 324 }
 325 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
 326
 327 DEFINE_PER_CPU(u64, idle_spurr_cycles);
 328 DEFINE_PER_CPU(u64, idle_entry_purr_snap);
 329 DEFINE_PER_CPU(u64, idle_entry_spurr_snap);
 330 static void pseries_lpar_idle(void)
 331 {
 332         /*
 333          * Default handler to go into low thread priority and possibly
 334          * low power mode by ceding processor to hypervisor
 335          */
 336
 337         if (!prep_irq_for_idle())
 338                 return;
 339
 340         /* Indicate to hypervisor that we are idle. */
 341         pseries_idle_prolog();
 342
 343         /*
 344          * Yield the processor to the hypervisor.  We return if
 345          * an external interrupt occurs (which are driven prior
 346          * to returning here) or if a prod occurs from another
 347          * processor. When returning here, external interrupts
 348          * are enabled.
 349          */
 350         cede_processor();
 351
 352         pseries_idle_epilog();
 353 }
 354
 355 /*
 356  * Enable relocation on during exceptions. This has partition wide scope and
 357  * may take a while to complete, if it takes longer than one second we will
 358  * just give up rather than wasting any more time on this - if that turns out
 359  * to ever be a problem in practice we can move this into a kernel thread to
 360  * finish off the process later in boot.
 361  */
 362 bool pseries_enable_reloc_on_exc(void)
 363 {
 364         long rc;
 365         unsigned int delay, total_delay = 0;
 366
 367         while (1) {
 368                 rc = enable_reloc_on_exceptions();
 369                 if (!H_IS_LONG_BUSY(rc)) {
 370                         if (rc == H_P2) {
 371                                 pr_info("Relocation on exceptions not"
 372                                         " supported\n");
 373                                 return false;
 374                         } else if (rc != H_SUCCESS) {
 375                                 pr_warn("Unable to enable relocation"
 376                                         " on exceptions: %ld\n", rc);
 377                                 return false;
 378                         }
 379                         return true;
 380                 }
 381
 382                 delay = get_longbusy_msecs(rc);
 383                 total_delay += delay;
 384                 if (total_delay > 1000) {
 385                         pr_warn("Warning: Giving up waiting to enable "
 386                                 "relocation on exceptions (%u msec)!\n",
 387                                 total_delay);
 388                         return false;
 389                 }
 390
 391                 mdelay(delay);
 392         }
 393 }
 394 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
 395
 396 void pseries_disable_reloc_on_exc(void)
 397 {
 398         long rc;
 399
 400         while (1) {
 401                 rc = disable_reloc_on_exceptions();
 402                 if (!H_IS_LONG_BUSY(rc))
 403                         break;
 404                 mdelay(get_longbusy_msecs(rc));
 405         }
 406         if (rc != H_SUCCESS)
 407                 pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
 408                         rc);
 409 }
 410 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
 411
 412 #ifdef CONFIG_KEXEC_CORE
 413 static void pSeries_machine_kexec(struct kimage *image)
 414 {
 415         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 416                 pseries_disable_reloc_on_exc();
 417
 418         default_machine_kexec(image);
 419 }
 420 #endif
 421
 422 #ifdef __LITTLE_ENDIAN__
 423 void pseries_big_endian_exceptions(void)
 424 {
 425         long rc;
 426
 427         while (1) {
 428                 rc = enable_big_endian_exceptions();
 429                 if (!H_IS_LONG_BUSY(rc))
 430                         break;
 431                 mdelay(get_longbusy_msecs(rc));
 432         }
 433
 434         /*
 435          * At this point it is unlikely panic() will get anything
 436          * out to the user, since this is called very late in kexec
 437          * but at least this will stop us from continuing on further
 438          * and creating an even more difficult to debug situation.
 439          *
 440          * There is a known problem when kdump'ing, if cpus are offline
 441          * the above call will fail. Rather than panicking again, keep
 442          * going and hope the kdump kernel is also little endian, which
 443          * it usually is.
 444          */
 445         if (rc && !kdump_in_progress())
 446                 panic("Could not enable big endian exceptions");
 447 }
 448
 449 void pseries_little_endian_exceptions(void)
 450 {
 451         long rc;
 452
 453         while (1) {
 454                 rc = enable_little_endian_exceptions();
 455                 if (!H_IS_LONG_BUSY(rc))
 456                         break;
 457                 mdelay(get_longbusy_msecs(rc));
 458         }
 459         if (rc) {
 460                 ppc_md.progress("H_SET_MODE LE exception fail", 0);
 461                 panic("Could not enable little endian exceptions");
 462         }
 463 }
 464 #endif
 465
 466 static void __init find_and_init_phbs(void)
 467 {
 468         struct device_node *node;
 469         struct pci_controller *phb;
 470         struct device_node *root = of_find_node_by_path("/");
 471
 472         for_each_child_of_node(root, node) {
 473                 if (!of_node_is_type(node, "pci") &&
 474                     !of_node_is_type(node, "pciex"))
 475                         continue;
 476
 477                 phb = pcibios_alloc_controller(node);
 478                 if (!phb)
 479                         continue;
 480                 rtas_setup_phb(phb);
 481                 pci_process_bridge_OF_ranges(phb, node, 0);
 482                 isa_bridge_find_early(phb);
 483                 phb->controller_ops = pseries_pci_controller_ops;
 484         }
 485
 486         of_node_put(root);
 487
 488         /*
 489          * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
 490          * in chosen.
 491          */
 492         of_pci_check_probe_only();
 493 }
 494
 495 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
 496 {
 497         /*
 498          * The features below are disabled by default, so we instead look to see
 499          * if firmware has *enabled* them, and set them if so.
 500          */
 501         if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
 502                 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
 503
 504         if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
 505                 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
 506
 507         if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
 508                 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
 509
 510         if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
 511                 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
 512
 513         if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
 514                 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
 515
 516         if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
 517                 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
 518
 519         if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
 520                 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
 521
 522         if (result->character & H_CPU_CHAR_BCCTR_LINK_FLUSH_ASSIST)
 523                 security_ftr_set(SEC_FTR_BCCTR_LINK_FLUSH_ASSIST);
 524
 525         if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
 526                 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
 527
 528         if (result->behaviour & H_CPU_BEHAV_FLUSH_LINK_STACK)
 529                 security_ftr_set(SEC_FTR_FLUSH_LINK_STACK);
 530
 531         /*
 532          * The features below are enabled by default, so we instead look to see
 533          * if firmware has *disabled* them, and clear them if so.
 534          */
 535         if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
 536                 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
 537
 538         if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
 539                 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
 540
 541         if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
 542                 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
 543 }
 544
 545 void pseries_setup_rfi_flush(void)
 546 {
 547         struct h_cpu_char_result result;
 548         enum l1d_flush_type types;
 549         bool enable;
 550         long rc;
 551
 552         /*
 553          * Set features to the defaults assumed by init_cpu_char_feature_flags()
 554          * so it can set/clear again any features that might have changed after
 555          * migration, and in case the hypercall fails and it is not even called.
 556          */
 557         powerpc_security_features = SEC_FTR_DEFAULT;
 558
 559         rc = plpar_get_cpu_characteristics(&result);
 560         if (rc == H_SUCCESS)
 561                 init_cpu_char_feature_flags(&result);
 562
 563         /*
 564          * We're the guest so this doesn't apply to us, clear it to simplify
 565          * handling of it elsewhere.
 566          */
 567         security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
 568
 569         types = L1D_FLUSH_FALLBACK;
 570
 571         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
 572                 types |= L1D_FLUSH_MTTRIG;
 573
 574         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
 575                 types |= L1D_FLUSH_ORI;
 576
 577         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
 578                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
 579
 580         setup_rfi_flush(types, enable);
 581         setup_count_cache_flush();
 582 }
 583
 584 #ifdef CONFIG_PCI_IOV
 585 enum rtas_iov_fw_value_map {
 586         NUM_RES_PROPERTY  = 0, /* Number of Resources */
 587         LOW_INT           = 1, /* Lowest 32 bits of Address */
 588         START_OF_ENTRIES  = 2, /* Always start of entry */
 589         APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
 590         WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
 591         NEXT_ENTRY        = 7  /* Go to next entry on array */
 592 };
 593
 594 enum get_iov_fw_value_index {
 595         BAR_ADDRS     = 1,    /*  Get Bar Address */
 596         APERTURE_SIZE = 2,    /*  Get Aperture Size */
 597         WDW_SIZE      = 3     /*  Get Window Size */
 598 };
 599
 600 resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
 601                                          enum get_iov_fw_value_index value)
 602 {
 603         const int *indexes;
 604         struct device_node *dn = pci_device_to_OF_node(dev);
 605         int i, num_res, ret = 0;
 606
 607         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 608         if (!indexes)
 609                 return  0;
 610
 611         /*
 612          * First element in the array is the number of Bars
 613          * returned.  Search through the list to find the matching
 614          * bar
 615          */
 616         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 617         if (resno >= num_res)
 618                 return 0; /* or an errror */
 619
 620         i = START_OF_ENTRIES + NEXT_ENTRY * resno;
 621         switch (value) {
 622         case BAR_ADDRS:
 623                 ret = of_read_number(&indexes[i], 2);
 624                 break;
 625         case APERTURE_SIZE:
 626                 ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
 627                 break;
 628         case WDW_SIZE:
 629                 ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
 630                 break;
 631         }
 632
 633         return ret;
 634 }
 635
 636 void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
 637 {
 638         struct resource *res;
 639         resource_size_t base, size;
 640         int i, r, num_res;
 641
 642         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 643         num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
 644         for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
 645              i += NEXT_ENTRY, r++) {
 646                 res = &dev->resource[r + PCI_IOV_RESOURCES];
 647                 base = of_read_number(&indexes[i], 2);
 648                 size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
 649                 res->flags = pci_parse_of_flags(of_read_number
 650                                                 (&indexes[i + LOW_INT], 1), 0);
 651                 res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
 652                 res->name = pci_name(dev);
 653                 res->start = base;
 654                 res->end = base + size - 1;
 655         }
 656 }
 657
 658 void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
 659 {
 660         struct resource *res, *root, *conflict;
 661         resource_size_t base, size;
 662         int i, r, num_res;
 663
 664         /*
 665          * First element in the array is the number of Bars
 666          * returned.  Search through the list to find the matching
 667          * bars assign them from firmware into resources structure.
 668          */
 669         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 670         for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
 671              i += NEXT_ENTRY, r++) {
 672                 res = &dev->resource[r + PCI_IOV_RESOURCES];
 673                 base = of_read_number(&indexes[i], 2);
 674                 size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
 675                 res->name = pci_name(dev);
 676                 res->start = base;
 677                 res->end = base + size - 1;
 678                 root = &iomem_resource;
 679                 dev_dbg(&dev->dev,
 680                         "pSeries IOV BAR %d: trying firmware assignment %pR\n",
 681                          r + PCI_IOV_RESOURCES, res);
 682                 conflict = request_resource_conflict(root, res);
 683                 if (conflict) {
 684                         dev_info(&dev->dev,
 685                                  "BAR %d: %pR conflicts with %s %pR\n",
 686                                  r + PCI_IOV_RESOURCES, res,
 687                                  conflict->name, conflict);
 688                         res->flags |= IORESOURCE_UNSET;
 689                 }
 690         }
 691 }
 692
 693 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
 694 {
 695         int i;
 696
 697         pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
 698         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
 699                 pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
 700 }
 701
 702 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
 703 {
 704         const int *indexes;
 705         struct device_node *dn = pci_device_to_OF_node(pdev);
 706
 707         /*Firmware must support open sriov otherwise dont configure*/
 708         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 709         if (indexes)
 710                 of_pci_set_vf_bar_size(pdev, indexes);
 711         else
 712                 pseries_disable_sriov_resources(pdev);
 713 }
 714
 715 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
 716 {
 717         const int *indexes;
 718         struct device_node *dn = pci_device_to_OF_node(pdev);
 719
 720         if (!pdev->is_physfn || pci_dev_is_added(pdev))
 721                 return;
 722         /*Firmware must support open sriov otherwise dont configure*/
 723         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 724         if (indexes)
 725                 of_pci_parse_iov_addrs(pdev, indexes);
 726         else
 727                 pseries_disable_sriov_resources(pdev);
 728 }
 729
 730 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
 731                                                           int resno)
 732 {
 733         const __be32 *reg;
 734         struct device_node *dn = pci_device_to_OF_node(pdev);
 735
 736         /*Firmware must support open sriov otherwise report regular alignment*/
 737         reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
 738         if (!reg)
 739                 return pci_iov_resource_size(pdev, resno);
 740
 741         if (!pdev->is_physfn)
 742                 return 0;
 743         return pseries_get_iov_fw_value(pdev,
 744                                         resno - PCI_IOV_RESOURCES,
 745                                         APERTURE_SIZE);
 746 }
 747 #endif
 748
 749 static void __init pSeries_setup_arch(void)
 750 {
 751         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
 752
 753         /* Discover PIC type and setup ppc_md accordingly */
 754         smp_init_pseries();
 755
 756
 757         if (radix_enabled() && !mmu_has_feature(MMU_FTR_GTSE))
 758                 if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
 759                         panic("BUG: Radix support requires either GTSE or RPT_INVALIDATE\n");
 760
 761
 762         /* openpic global configuration register (64-bit format). */
 763         /* openpic Interrupt Source Unit pointer (64-bit format). */
 764         /* python0 facility area (mmio) (64-bit format) REAL address. */
 765
 766         /* init to some ~sane value until calibrate_delay() runs */
 767         loops_per_jiffy = 50000000;
 768
 769         fwnmi_init();
 770
 771         pseries_setup_rfi_flush();
 772         setup_stf_barrier();
 773         pseries_lpar_read_hblkrm_characteristics();
 774
 775         /* By default, only probe PCI (can be overridden by rtas_pci) */
 776         pci_add_flags(PCI_PROBE_ONLY);
 777
 778         /* Find and initialize PCI host bridges */
 779         init_pci_config_tokens();
 780         find_and_init_phbs();
 781         of_reconfig_notifier_register(&pci_dn_reconfig_nb);
 782
 783         pSeries_nvram_init();
 784
 785         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 786                 vpa_init(boot_cpuid);
 787
 788                 if (lppaca_shared_proc(get_lppaca())) {
 789                         static_branch_enable(&shared_processor);
 790                         pv_spinlocks_init();
 791                 }
 792
 793                 ppc_md.power_save = pseries_lpar_idle;
 794                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
 795 #ifdef CONFIG_PCI_IOV
 796                 ppc_md.pcibios_fixup_resources =
 797                         pseries_pci_fixup_resources;
 798                 ppc_md.pcibios_fixup_sriov =
 799                         pseries_pci_fixup_iov_resources;
 800                 ppc_md.pcibios_iov_resource_alignment =
 801                         pseries_pci_iov_resource_alignment;
 802 #endif
 803         } else {
 804                 /* No special idle routine */
 805                 ppc_md.enable_pmcs = power4_enable_pmcs;
 806         }
 807
 808         ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
 809
 810         if (swiotlb_force == SWIOTLB_FORCE)
 811                 ppc_swiotlb_enable = 1;
 812 }
 813
 814 static void pseries_panic(char *str)
 815 {
 816         panic_flush_kmsg_end();
 817         rtas_os_term(str);
 818 }
 819
 820 static int __init pSeries_init_panel(void)
 821 {
 822         /* Manually leave the kernel version on the panel. */
 823 #ifdef __BIG_ENDIAN__
 824         ppc_md.progress("Linux ppc64\n", 0);
 825 #else
 826         ppc_md.progress("Linux ppc64le\n", 0);
 827 #endif
 828         ppc_md.progress(init_utsname()->version, 0);
 829
 830         return 0;
 831 }
 832 machine_arch_initcall(pseries, pSeries_init_panel);
 833
 834 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
 835 {
 836         return plpar_hcall_norets(H_SET_DABR, dabr);
 837 }
 838
 839 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
 840 {
 841         /* Have to set at least one bit in the DABRX according to PAPR */
 842         if (dabrx == 0 && dabr == 0)
 843                 dabrx = DABRX_USER;
 844         /* PAPR says we can only set kernel and user bits */
 845         dabrx &= DABRX_KERNEL | DABRX_USER;
 846
 847         return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
 848 }
 849
 850 static int pseries_set_dawr(int nr, unsigned long dawr, unsigned long dawrx)
 851 {
 852         /* PAPR says we can't set HYP */
 853         dawrx &= ~DAWRX_HYP;
 854
 855         if (nr == 0)
 856                 return plpar_set_watchpoint0(dawr, dawrx);
 857         else
 858                 return plpar_set_watchpoint1(dawr, dawrx);
 859 }
 860
 861 #define CMO_CHARACTERISTICS_TOKEN 44
 862 #define CMO_MAXLENGTH 1026
 863
 864 void pSeries_coalesce_init(void)
 865 {
 866         struct hvcall_mpp_x_data mpp_x_data;
 867
 868         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
 869                 powerpc_firmware_features |= FW_FEATURE_XCMO;
 870         else
 871                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
 872 }
 873
 874 /**
 875  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
 876  * handle that here. (Stolen from parse_system_parameter_string)
 877  */
 878 static void pSeries_cmo_feature_init(void)
 879 {
 880         char *ptr, *key, *value, *end;
 881         int call_status;
 882         int page_order = IOMMU_PAGE_SHIFT_4K;
 883
 884         pr_debug(" -> fw_cmo_feature_init()\n");
 885         spin_lock(&rtas_data_buf_lock);
 886         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
 887         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 888                                 NULL,
 889                                 CMO_CHARACTERISTICS_TOKEN,
 890                                 __pa(rtas_data_buf),
 891                                 RTAS_DATA_BUF_SIZE);
 892
 893         if (call_status != 0) {
 894                 spin_unlock(&rtas_data_buf_lock);
 895                 pr_debug("CMO not available\n");
 896                 pr_debug(" <- fw_cmo_feature_init()\n");
 897                 return;
 898         }
 899
 900         end = rtas_data_buf + CMO_MAXLENGTH - 2;
 901         ptr = rtas_data_buf + 2;        /* step over strlen value */
 902         key = value = ptr;
 903
 904         while (*ptr && (ptr <= end)) {
 905                 /* Separate the key and value by replacing '=' with '\0' and
 906                  * point the value at the string after the '='
 907                  */
 908                 if (ptr[0] == '=') {
 909                         ptr[0] = '\0';
 910                         value = ptr + 1;
 911                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
 912                         /* Terminate the string containing the key/value pair */
 913                         ptr[0] = '\0';
 914
 915                         if (key == value) {
 916                                 pr_debug("Malformed key/value pair\n");
 917                                 /* Never found a '=', end processing */
 918                                 break;
 919                         }
 920
 921                         if (0 == strcmp(key, "CMOPageSize"))
 922                                 page_order = simple_strtol(value, NULL, 10);
 923                         else if (0 == strcmp(key, "PrPSP"))
 924                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
 925                         else if (0 == strcmp(key, "SecPSP"))
 926                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
 927                         value = key = ptr + 1;
 928                 }
 929                 ptr++;
 930         }
 931
 932         /* Page size is returned as the power of 2 of the page size,
 933          * convert to the page size in bytes before returning
 934          */
 935         CMO_PageSize = 1 << page_order;
 936         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
 937
 938         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
 939                 pr_info("CMO enabled\n");
 940                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 941                          CMO_SecPSP);
 942                 powerpc_firmware_features |= FW_FEATURE_CMO;
 943                 pSeries_coalesce_init();
 944         } else
 945                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 946                          CMO_SecPSP);
 947         spin_unlock(&rtas_data_buf_lock);
 948         pr_debug(" <- fw_cmo_feature_init()\n");
 949 }
 950
 951 /*
 952  * Early initialization.  Relocation is on but do not reference unbolted pages
 953  */
 954 static void __init pseries_init(void)
 955 {
 956         pr_debug(" -> pseries_init()\n");
 957
 958 #ifdef CONFIG_HVC_CONSOLE
 959         if (firmware_has_feature(FW_FEATURE_LPAR))
 960                 hvc_vio_init_early();
 961 #endif
 962         if (firmware_has_feature(FW_FEATURE_XDABR))
 963                 ppc_md.set_dabr = pseries_set_xdabr;
 964         else if (firmware_has_feature(FW_FEATURE_DABR))
 965                 ppc_md.set_dabr = pseries_set_dabr;
 966
 967         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 968                 ppc_md.set_dawr = pseries_set_dawr;
 969
 970         pSeries_cmo_feature_init();
 971         iommu_init_early_pSeries();
 972
 973         pr_debug(" <- pseries_init()\n");
 974 }
 975
 976 /**
 977  * pseries_power_off - tell firmware about how to power off the system.
 978  *
 979  * This function calls either the power-off rtas token in normal cases
 980  * or the ibm,power-off-ups token (if present & requested) in case of
 981  * a power failure. If power-off token is used, power on will only be
 982  * possible with power button press. If ibm,power-off-ups token is used
 983  * it will allow auto poweron after power is restored.
 984  */
 985 static void pseries_power_off(void)
 986 {
 987         int rc;
 988         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
 989
 990         if (rtas_flash_term_hook)
 991                 rtas_flash_term_hook(SYS_POWER_OFF);
 992
 993         if (rtas_poweron_auto == 0 ||
 994                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
 995                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
 996                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
 997         } else {
 998                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
 999                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
1000         }
1001         for (;;);
1002 }
1003
1004 static int __init pSeries_probe(void)
1005 {
1006         if (!of_node_is_type(of_root, "chrp"))
1007                 return 0;
1008
1009         /* Cell blades firmware claims to be chrp while it's not. Until this
1010          * is fixed, we need to avoid those here.
1011          */
1012         if (of_machine_is_compatible("IBM,CPBW-1.0") ||
1013             of_machine_is_compatible("IBM,CBEA"))
1014                 return 0;
1015
1016         pm_power_off = pseries_power_off;
1017
1018         pr_debug("Machine is%s LPAR !\n",
1019                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1020
1021         pseries_init();
1022
1023         return 1;
1024 }
1025
1026 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1027 {
1028         if (firmware_has_feature(FW_FEATURE_LPAR))
1029                 return PCI_PROBE_DEVTREE;
1030         return PCI_PROBE_NORMAL;
1031 }
1032
1033 struct pci_controller_ops pseries_pci_controller_ops = {
1034         .probe_mode             = pSeries_pci_probe_mode,
1035 };
1036
1037 define_machine(pseries) {
1038         .name                   = "pSeries",
1039         .probe                  = pSeries_probe,
1040         .setup_arch             = pSeries_setup_arch,
1041         .init_IRQ               = pseries_init_irq,
1042         .show_cpuinfo           = pSeries_show_cpuinfo,
1043         .log_error              = pSeries_log_error,
1044         .pcibios_fixup          = pSeries_final_fixup,
1045         .restart                = rtas_restart,
1046         .halt                   = rtas_halt,
1047         .panic                  = pseries_panic,
1048         .get_boot_time          = rtas_get_boot_time,
1049         .get_rtc_time           = rtas_get_rtc_time,
1050         .set_rtc_time           = rtas_set_rtc_time,
1051         .calibrate_decr         = generic_calibrate_decr,
1052         .progress               = rtas_progress,
1053         .system_reset_exception = pSeries_system_reset_exception,
1054         .machine_check_early    = pseries_machine_check_realmode,
1055         .machine_check_exception = pSeries_machine_check_exception,
1056 #ifdef CONFIG_KEXEC_CORE
1057         .machine_kexec          = pSeries_machine_kexec,
1058         .kexec_cpu_down         = pseries_kexec_cpu_down,
1059 #endif
1060 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
1061         .memory_block_size      = pseries_memory_block_size,
1062 #endif
1063 };