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/pgtable.h>
  47 #include <asm/prom.h>
  48 #include <asm/rtas.h>
  49 #include <asm/pci-bridge.h>
  50 #include <asm/iommu.h>
  51 #include <asm/dma.h>
  52 #include <asm/machdep.h>
  53 #include <asm/irq.h>
  54 #include <asm/time.h>
  55 #include <asm/nvram.h>
  56 #include <asm/pmc.h>
  57 #include <asm/xics.h>
  58 #include <asm/xive.h>
  59 #include <asm/ppc-pci.h>
  60 #include <asm/i8259.h>
  61 #include <asm/udbg.h>
  62 #include <asm/smp.h>
  63 #include <asm/firmware.h>
  64 #include <asm/eeh.h>
  65 #include <asm/reg.h>
  66 #include <asm/plpar_wrappers.h>
  67 #include <asm/kexec.h>
  68 #include <asm/isa-bridge.h>
  69 #include <asm/security_features.h>
  70 #include <asm/asm-const.h>
  71 #include <asm/swiotlb.h>
  72 #include <asm/svm.h>
  73
  74 #include "pseries.h"
  75 #include "../../../../drivers/pci/pci.h"
  76
  77 DEFINE_STATIC_KEY_FALSE(shared_processor);
  78 EXPORT_SYMBOL(shared_processor);
  79
  80 int CMO_PrPSP = -1;
  81 int CMO_SecPSP = -1;
  82 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
  83 EXPORT_SYMBOL(CMO_PageSize);
  84
  85 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
  86
  87 static void pSeries_show_cpuinfo(struct seq_file *m)
  88 {
  89         struct device_node *root;
  90         const char *model = "";
  91
  92         root = of_find_node_by_path("/");
  93         if (root)
  94                 model = of_get_property(root, "model", NULL);
  95         seq_printf(m, "machine\t\t: CHRP %s\n", model);
  96         of_node_put(root);
  97         if (radix_enabled())
  98                 seq_printf(m, "MMU\t\t: Radix\n");
  99         else
 100                 seq_printf(m, "MMU\t\t: Hash\n");
 101 }
 102
 103 /* Initialize firmware assisted non-maskable interrupts if
 104  * the firmware supports this feature.
 105  */
 106 static void __init fwnmi_init(void)
 107 {
 108         unsigned long system_reset_addr, machine_check_addr;
 109         u8 *mce_data_buf;
 110         unsigned int i;
 111         int nr_cpus = num_possible_cpus();
 112 #ifdef CONFIG_PPC_BOOK3S_64
 113         struct slb_entry *slb_ptr;
 114         size_t size;
 115 #endif
 116
 117         int ibm_nmi_register = rtas_token("ibm,nmi-register");
 118         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
 119                 return;
 120
 121         /* If the kernel's not linked at zero we point the firmware at low
 122          * addresses anyway, and use a trampoline to get to the real code. */
 123         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
 124         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
 125
 126         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
 127                                 machine_check_addr))
 128                 fwnmi_active = 1;
 129
 130         /*
 131          * Allocate a chunk for per cpu buffer to hold rtas errorlog.
 132          * It will be used in real mode mce handler, hence it needs to be
 133          * below RMA.
 134          */
 135         mce_data_buf = memblock_alloc_try_nid_raw(RTAS_ERROR_LOG_MAX * nr_cpus,
 136                                         RTAS_ERROR_LOG_MAX, MEMBLOCK_LOW_LIMIT,
 137                                         ppc64_rma_size, NUMA_NO_NODE);
 138         if (!mce_data_buf)
 139                 panic("Failed to allocate %d bytes below %pa for MCE buffer\n",
 140                       RTAS_ERROR_LOG_MAX * nr_cpus, &ppc64_rma_size);
 141
 142         for_each_possible_cpu(i) {
 143                 paca_ptrs[i]->mce_data_buf = mce_data_buf +
 144                                                 (RTAS_ERROR_LOG_MAX * i);
 145         }
 146
 147 #ifdef CONFIG_PPC_BOOK3S_64
 148         if (!radix_enabled()) {
 149                 /* Allocate per cpu area to save old slb contents during MCE */
 150                 size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
 151                 slb_ptr = memblock_alloc_try_nid_raw(size,
 152                                 sizeof(struct slb_entry), MEMBLOCK_LOW_LIMIT,
 153                                 ppc64_rma_size, NUMA_NO_NODE);
 154                 if (!slb_ptr)
 155                         panic("Failed to allocate %zu bytes below %pa for slb area\n",
 156                               size, &ppc64_rma_size);
 157
 158                 for_each_possible_cpu(i)
 159                         paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
 160         }
 161 #endif
 162 }
 163
 164 static void pseries_8259_cascade(struct irq_desc *desc)
 165 {
 166         struct irq_chip *chip = irq_desc_get_chip(desc);
 167         unsigned int cascade_irq = i8259_irq();
 168
 169         if (cascade_irq)
 170                 generic_handle_irq(cascade_irq);
 171
 172         chip->irq_eoi(&desc->irq_data);
 173 }
 174
 175 static void __init pseries_setup_i8259_cascade(void)
 176 {
 177         struct device_node *np, *old, *found = NULL;
 178         unsigned int cascade;
 179         const u32 *addrp;
 180         unsigned long intack = 0;
 181         int naddr;
 182
 183         for_each_node_by_type(np, "interrupt-controller") {
 184                 if (of_device_is_compatible(np, "chrp,iic")) {
 185                         found = np;
 186                         break;
 187                 }
 188         }
 189
 190         if (found == NULL) {
 191                 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
 192                 return;
 193         }
 194
 195         cascade = irq_of_parse_and_map(found, 0);
 196         if (!cascade) {
 197                 printk(KERN_ERR "pic: failed to map cascade interrupt");
 198                 return;
 199         }
 200         pr_debug("pic: cascade mapped to irq %d\n", cascade);
 201
 202         for (old = of_node_get(found); old != NULL ; old = np) {
 203                 np = of_get_parent(old);
 204                 of_node_put(old);
 205                 if (np == NULL)
 206                         break;
 207                 if (!of_node_name_eq(np, "pci"))
 208                         continue;
 209                 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
 210                 if (addrp == NULL)
 211                         continue;
 212                 naddr = of_n_addr_cells(np);
 213                 intack = addrp[naddr-1];
 214                 if (naddr > 1)
 215                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
 216         }
 217         if (intack)
 218                 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
 219         i8259_init(found, intack);
 220         of_node_put(found);
 221         irq_set_chained_handler(cascade, pseries_8259_cascade);
 222 }
 223
 224 static void __init pseries_init_irq(void)
 225 {
 226         /* Try using a XIVE if available, otherwise use a XICS */
 227         if (!xive_spapr_init()) {
 228                 xics_init();
 229                 pseries_setup_i8259_cascade();
 230         }
 231 }
 232
 233 static void pseries_lpar_enable_pmcs(void)
 234 {
 235         unsigned long set, reset;
 236
 237         set = 1UL << 63;
 238         reset = 0;
 239         plpar_hcall_norets(H_PERFMON, set, reset);
 240 }
 241
 242 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
 243 {
 244         struct of_reconfig_data *rd = data;
 245         struct device_node *parent, *np = rd->dn;
 246         struct pci_dn *pdn;
 247         int err = NOTIFY_OK;
 248
 249         switch (action) {
 250         case OF_RECONFIG_ATTACH_NODE:
 251                 parent = of_get_parent(np);
 252                 pdn = parent ? PCI_DN(parent) : NULL;
 253                 if (pdn)
 254                         pci_add_device_node_info(pdn->phb, np);
 255
 256                 of_node_put(parent);
 257                 break;
 258         case OF_RECONFIG_DETACH_NODE:
 259                 pdn = PCI_DN(np);
 260                 if (pdn)
 261                         list_del(&pdn->list);
 262                 break;
 263         default:
 264                 err = NOTIFY_DONE;
 265                 break;
 266         }
 267         return err;
 268 }
 269
 270 static struct notifier_block pci_dn_reconfig_nb = {
 271         .notifier_call = pci_dn_reconfig_notifier,
 272 };
 273
 274 struct kmem_cache *dtl_cache;
 275
 276 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 277 /*
 278  * Allocate space for the dispatch trace log for all possible cpus
 279  * and register the buffers with the hypervisor.  This is used for
 280  * computing time stolen by the hypervisor.
 281  */
 282 static int alloc_dispatch_logs(void)
 283 {
 284         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 285                 return 0;
 286
 287         if (!dtl_cache)
 288                 return 0;
 289
 290         alloc_dtl_buffers(0);
 291
 292         /* Register the DTL for the current (boot) cpu */
 293         register_dtl_buffer(smp_processor_id());
 294
 295         return 0;
 296 }
 297 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 298 static inline int alloc_dispatch_logs(void)
 299 {
 300         return 0;
 301 }
 302 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 303
 304 static int alloc_dispatch_log_kmem_cache(void)
 305 {
 306         void (*ctor)(void *) = get_dtl_cache_ctor();
 307
 308         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
 309                                                 DISPATCH_LOG_BYTES, 0, ctor);
 310         if (!dtl_cache) {
 311                 pr_warn("Failed to create dispatch trace log buffer cache\n");
 312                 pr_warn("Stolen time statistics will be unreliable\n");
 313                 return 0;
 314         }
 315
 316         return alloc_dispatch_logs();
 317 }
 318 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
 319
 320 static void pseries_lpar_idle(void)
 321 {
 322         /*
 323          * Default handler to go into low thread priority and possibly
 324          * low power mode by ceding processor to hypervisor
 325          */
 326
 327         if (!prep_irq_for_idle())
 328                 return;
 329
 330         /* Indicate to hypervisor that we are idle. */
 331         get_lppaca()->idle = 1;
 332
 333         /*
 334          * Yield the processor to the hypervisor.  We return if
 335          * an external interrupt occurs (which are driven prior
 336          * to returning here) or if a prod occurs from another
 337          * processor. When returning here, external interrupts
 338          * are enabled.
 339          */
 340         cede_processor();
 341
 342         get_lppaca()->idle = 0;
 343 }
 344
 345 /*
 346  * Enable relocation on during exceptions. This has partition wide scope and
 347  * may take a while to complete, if it takes longer than one second we will
 348  * just give up rather than wasting any more time on this - if that turns out
 349  * to ever be a problem in practice we can move this into a kernel thread to
 350  * finish off the process later in boot.
 351  */
 352 void pseries_enable_reloc_on_exc(void)
 353 {
 354         long rc;
 355         unsigned int delay, total_delay = 0;
 356
 357         while (1) {
 358                 rc = enable_reloc_on_exceptions();
 359                 if (!H_IS_LONG_BUSY(rc)) {
 360                         if (rc == H_P2) {
 361                                 pr_info("Relocation on exceptions not"
 362                                         " supported\n");
 363                         } else if (rc != H_SUCCESS) {
 364                                 pr_warn("Unable to enable relocation"
 365                                         " on exceptions: %ld\n", rc);
 366                         }
 367                         break;
 368                 }
 369
 370                 delay = get_longbusy_msecs(rc);
 371                 total_delay += delay;
 372                 if (total_delay > 1000) {
 373                         pr_warn("Warning: Giving up waiting to enable "
 374                                 "relocation on exceptions (%u msec)!\n",
 375                                 total_delay);
 376                         return;
 377                 }
 378
 379                 mdelay(delay);
 380         }
 381 }
 382 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
 383
 384 void pseries_disable_reloc_on_exc(void)
 385 {
 386         long rc;
 387
 388         while (1) {
 389                 rc = disable_reloc_on_exceptions();
 390                 if (!H_IS_LONG_BUSY(rc))
 391                         break;
 392                 mdelay(get_longbusy_msecs(rc));
 393         }
 394         if (rc != H_SUCCESS)
 395                 pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
 396                         rc);
 397 }
 398 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
 399
 400 #ifdef CONFIG_KEXEC_CORE
 401 static void pSeries_machine_kexec(struct kimage *image)
 402 {
 403         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 404                 pseries_disable_reloc_on_exc();
 405
 406         default_machine_kexec(image);
 407 }
 408 #endif
 409
 410 #ifdef __LITTLE_ENDIAN__
 411 void pseries_big_endian_exceptions(void)
 412 {
 413         long rc;
 414
 415         while (1) {
 416                 rc = enable_big_endian_exceptions();
 417                 if (!H_IS_LONG_BUSY(rc))
 418                         break;
 419                 mdelay(get_longbusy_msecs(rc));
 420         }
 421
 422         /*
 423          * At this point it is unlikely panic() will get anything
 424          * out to the user, since this is called very late in kexec
 425          * but at least this will stop us from continuing on further
 426          * and creating an even more difficult to debug situation.
 427          *
 428          * There is a known problem when kdump'ing, if cpus are offline
 429          * the above call will fail. Rather than panicking again, keep
 430          * going and hope the kdump kernel is also little endian, which
 431          * it usually is.
 432          */
 433         if (rc && !kdump_in_progress())
 434                 panic("Could not enable big endian exceptions");
 435 }
 436
 437 void pseries_little_endian_exceptions(void)
 438 {
 439         long rc;
 440
 441         while (1) {
 442                 rc = enable_little_endian_exceptions();
 443                 if (!H_IS_LONG_BUSY(rc))
 444                         break;
 445                 mdelay(get_longbusy_msecs(rc));
 446         }
 447         if (rc) {
 448                 ppc_md.progress("H_SET_MODE LE exception fail", 0);
 449                 panic("Could not enable little endian exceptions");
 450         }
 451 }
 452 #endif
 453
 454 static void __init find_and_init_phbs(void)
 455 {
 456         struct device_node *node;
 457         struct pci_controller *phb;
 458         struct device_node *root = of_find_node_by_path("/");
 459
 460         for_each_child_of_node(root, node) {
 461                 if (!of_node_is_type(node, "pci") &&
 462                     !of_node_is_type(node, "pciex"))
 463                         continue;
 464
 465                 phb = pcibios_alloc_controller(node);
 466                 if (!phb)
 467                         continue;
 468                 rtas_setup_phb(phb);
 469                 pci_process_bridge_OF_ranges(phb, node, 0);
 470                 isa_bridge_find_early(phb);
 471                 phb->controller_ops = pseries_pci_controller_ops;
 472         }
 473
 474         of_node_put(root);
 475
 476         /*
 477          * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
 478          * in chosen.
 479          */
 480         of_pci_check_probe_only();
 481 }
 482
 483 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
 484 {
 485         /*
 486          * The features below are disabled by default, so we instead look to see
 487          * if firmware has *enabled* them, and set them if so.
 488          */
 489         if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
 490                 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
 491
 492         if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
 493                 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
 494
 495         if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
 496                 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
 497
 498         if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
 499                 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
 500
 501         if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
 502                 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
 503
 504         if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
 505                 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
 506
 507         if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
 508                 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
 509
 510         if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
 511                 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
 512
 513         /*
 514          * The features below are enabled by default, so we instead look to see
 515          * if firmware has *disabled* them, and clear them if so.
 516          */
 517         if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
 518                 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
 519
 520         if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
 521                 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
 522
 523         if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
 524                 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
 525 }
 526
 527 void pseries_setup_rfi_flush(void)
 528 {
 529         struct h_cpu_char_result result;
 530         enum l1d_flush_type types;
 531         bool enable;
 532         long rc;
 533
 534         /*
 535          * Set features to the defaults assumed by init_cpu_char_feature_flags()
 536          * so it can set/clear again any features that might have changed after
 537          * migration, and in case the hypercall fails and it is not even called.
 538          */
 539         powerpc_security_features = SEC_FTR_DEFAULT;
 540
 541         rc = plpar_get_cpu_characteristics(&result);
 542         if (rc == H_SUCCESS)
 543                 init_cpu_char_feature_flags(&result);
 544
 545         /*
 546          * We're the guest so this doesn't apply to us, clear it to simplify
 547          * handling of it elsewhere.
 548          */
 549         security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
 550
 551         types = L1D_FLUSH_FALLBACK;
 552
 553         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
 554                 types |= L1D_FLUSH_MTTRIG;
 555
 556         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
 557                 types |= L1D_FLUSH_ORI;
 558
 559         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
 560                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
 561
 562         setup_rfi_flush(types, enable);
 563         setup_count_cache_flush();
 564
 565         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 566                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
 567         setup_entry_flush(enable);
 568
 569         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 570                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
 571         setup_uaccess_flush(enable);
 572 }
 573
 574 #ifdef CONFIG_PCI_IOV
 575 enum rtas_iov_fw_value_map {
 576         NUM_RES_PROPERTY  = 0, /* Number of Resources */
 577         LOW_INT           = 1, /* Lowest 32 bits of Address */
 578         START_OF_ENTRIES  = 2, /* Always start of entry */
 579         APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
 580         WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
 581         NEXT_ENTRY        = 7  /* Go to next entry on array */
 582 };
 583
 584 enum get_iov_fw_value_index {
 585         BAR_ADDRS     = 1,    /*  Get Bar Address */
 586         APERTURE_SIZE = 2,    /*  Get Aperture Size */
 587         WDW_SIZE      = 3     /*  Get Window Size */
 588 };
 589
 590 resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
 591                                          enum get_iov_fw_value_index value)
 592 {
 593         const int *indexes;
 594         struct device_node *dn = pci_device_to_OF_node(dev);
 595         int i, num_res, ret = 0;
 596
 597         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 598         if (!indexes)
 599                 return  0;
 600
 601         /*
 602          * First element in the array is the number of Bars
 603          * returned.  Search through the list to find the matching
 604          * bar
 605          */
 606         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 607         if (resno >= num_res)
 608                 return 0; /* or an errror */
 609
 610         i = START_OF_ENTRIES + NEXT_ENTRY * resno;
 611         switch (value) {
 612         case BAR_ADDRS:
 613                 ret = of_read_number(&indexes[i], 2);
 614                 break;
 615         case APERTURE_SIZE:
 616                 ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
 617                 break;
 618         case WDW_SIZE:
 619                 ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
 620                 break;
 621         }
 622
 623         return ret;
 624 }
 625
 626 void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
 627 {
 628         struct resource *res;
 629         resource_size_t base, size;
 630         int i, r, num_res;
 631
 632         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 633         num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
 634         for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
 635              i += NEXT_ENTRY, r++) {
 636                 res = &dev->resource[r + PCI_IOV_RESOURCES];
 637                 base = of_read_number(&indexes[i], 2);
 638                 size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
 639                 res->flags = pci_parse_of_flags(of_read_number
 640                                                 (&indexes[i + LOW_INT], 1), 0);
 641                 res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
 642                 res->name = pci_name(dev);
 643                 res->start = base;
 644                 res->end = base + size - 1;
 645         }
 646 }
 647
 648 void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
 649 {
 650         struct resource *res, *root, *conflict;
 651         resource_size_t base, size;
 652         int i, r, num_res;
 653
 654         /*
 655          * First element in the array is the number of Bars
 656          * returned.  Search through the list to find the matching
 657          * bars assign them from firmware into resources structure.
 658          */
 659         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 660         for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
 661              i += NEXT_ENTRY, r++) {
 662                 res = &dev->resource[r + PCI_IOV_RESOURCES];
 663                 base = of_read_number(&indexes[i], 2);
 664                 size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
 665                 res->name = pci_name(dev);
 666                 res->start = base;
 667                 res->end = base + size - 1;
 668                 root = &iomem_resource;
 669                 dev_dbg(&dev->dev,
 670                         "pSeries IOV BAR %d: trying firmware assignment %pR\n",
 671                          r + PCI_IOV_RESOURCES, res);
 672                 conflict = request_resource_conflict(root, res);
 673                 if (conflict) {
 674                         dev_info(&dev->dev,
 675                                  "BAR %d: %pR conflicts with %s %pR\n",
 676                                  r + PCI_IOV_RESOURCES, res,
 677                                  conflict->name, conflict);
 678                         res->flags |= IORESOURCE_UNSET;
 679                 }
 680         }
 681 }
 682
 683 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
 684 {
 685         int i;
 686
 687         pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
 688         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
 689                 pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
 690 }
 691
 692 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
 693 {
 694         const int *indexes;
 695         struct device_node *dn = pci_device_to_OF_node(pdev);
 696
 697         /*Firmware must support open sriov otherwise dont configure*/
 698         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 699         if (indexes)
 700                 of_pci_set_vf_bar_size(pdev, indexes);
 701         else
 702                 pseries_disable_sriov_resources(pdev);
 703 }
 704
 705 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
 706 {
 707         const int *indexes;
 708         struct device_node *dn = pci_device_to_OF_node(pdev);
 709
 710         if (!pdev->is_physfn || pci_dev_is_added(pdev))
 711                 return;
 712         /*Firmware must support open sriov otherwise dont configure*/
 713         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 714         if (indexes)
 715                 of_pci_parse_iov_addrs(pdev, indexes);
 716         else
 717                 pseries_disable_sriov_resources(pdev);
 718 }
 719
 720 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
 721                                                           int resno)
 722 {
 723         const __be32 *reg;
 724         struct device_node *dn = pci_device_to_OF_node(pdev);
 725
 726         /*Firmware must support open sriov otherwise report regular alignment*/
 727         reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
 728         if (!reg)
 729                 return pci_iov_resource_size(pdev, resno);
 730
 731         if (!pdev->is_physfn)
 732                 return 0;
 733         return pseries_get_iov_fw_value(pdev,
 734                                         resno - PCI_IOV_RESOURCES,
 735                                         APERTURE_SIZE);
 736 }
 737 #endif
 738
 739 static void __init pSeries_setup_arch(void)
 740 {
 741         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
 742
 743         /* Discover PIC type and setup ppc_md accordingly */
 744         smp_init_pseries();
 745
 746
 747         /* openpic global configuration register (64-bit format). */
 748         /* openpic Interrupt Source Unit pointer (64-bit format). */
 749         /* python0 facility area (mmio) (64-bit format) REAL address. */
 750
 751         /* init to some ~sane value until calibrate_delay() runs */
 752         loops_per_jiffy = 50000000;
 753
 754         fwnmi_init();
 755
 756         pseries_setup_rfi_flush();
 757         setup_stf_barrier();
 758         pseries_lpar_read_hblkrm_characteristics();
 759
 760         /* By default, only probe PCI (can be overridden by rtas_pci) */
 761         pci_add_flags(PCI_PROBE_ONLY);
 762
 763         /* Find and initialize PCI host bridges */
 764         init_pci_config_tokens();
 765         find_and_init_phbs();
 766         of_reconfig_notifier_register(&pci_dn_reconfig_nb);
 767
 768         pSeries_nvram_init();
 769
 770         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 771                 vpa_init(boot_cpuid);
 772
 773                 if (lppaca_shared_proc(get_lppaca()))
 774                         static_branch_enable(&shared_processor);
 775
 776                 ppc_md.power_save = pseries_lpar_idle;
 777                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
 778 #ifdef CONFIG_PCI_IOV
 779                 ppc_md.pcibios_fixup_resources =
 780                         pseries_pci_fixup_resources;
 781                 ppc_md.pcibios_fixup_sriov =
 782                         pseries_pci_fixup_iov_resources;
 783                 ppc_md.pcibios_iov_resource_alignment =
 784                         pseries_pci_iov_resource_alignment;
 785 #endif
 786         } else {
 787                 /* No special idle routine */
 788                 ppc_md.enable_pmcs = power4_enable_pmcs;
 789         }
 790
 791         ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
 792
 793         if (swiotlb_force == SWIOTLB_FORCE)
 794                 ppc_swiotlb_enable = 1;
 795
 796         pseries_rng_init();
 797 }
 798
 799 static void pseries_panic(char *str)
 800 {
 801         panic_flush_kmsg_end();
 802         rtas_os_term(str);
 803 }
 804
 805 static int __init pSeries_init_panel(void)
 806 {
 807         /* Manually leave the kernel version on the panel. */
 808 #ifdef __BIG_ENDIAN__
 809         ppc_md.progress("Linux ppc64\n", 0);
 810 #else
 811         ppc_md.progress("Linux ppc64le\n", 0);
 812 #endif
 813         ppc_md.progress(init_utsname()->version, 0);
 814
 815         return 0;
 816 }
 817 machine_arch_initcall(pseries, pSeries_init_panel);
 818
 819 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
 820 {
 821         return plpar_hcall_norets(H_SET_DABR, dabr);
 822 }
 823
 824 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
 825 {
 826         /* Have to set at least one bit in the DABRX according to PAPR */
 827         if (dabrx == 0 && dabr == 0)
 828                 dabrx = DABRX_USER;
 829         /* PAPR says we can only set kernel and user bits */
 830         dabrx &= DABRX_KERNEL | DABRX_USER;
 831
 832         return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
 833 }
 834
 835 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
 836 {
 837         /* PAPR says we can't set HYP */
 838         dawrx &= ~DAWRX_HYP;
 839
 840         return  plpar_set_watchpoint0(dawr, dawrx);
 841 }
 842
 843 #define CMO_CHARACTERISTICS_TOKEN 44
 844 #define CMO_MAXLENGTH 1026
 845
 846 void pSeries_coalesce_init(void)
 847 {
 848         struct hvcall_mpp_x_data mpp_x_data;
 849
 850         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
 851                 powerpc_firmware_features |= FW_FEATURE_XCMO;
 852         else
 853                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
 854 }
 855
 856 /**
 857  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
 858  * handle that here. (Stolen from parse_system_parameter_string)
 859  */
 860 static void pSeries_cmo_feature_init(void)
 861 {
 862         char *ptr, *key, *value, *end;
 863         int call_status;
 864         int page_order = IOMMU_PAGE_SHIFT_4K;
 865
 866         pr_debug(" -> fw_cmo_feature_init()\n");
 867         spin_lock(&rtas_data_buf_lock);
 868         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
 869         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 870                                 NULL,
 871                                 CMO_CHARACTERISTICS_TOKEN,
 872                                 __pa(rtas_data_buf),
 873                                 RTAS_DATA_BUF_SIZE);
 874
 875         if (call_status != 0) {
 876                 spin_unlock(&rtas_data_buf_lock);
 877                 pr_debug("CMO not available\n");
 878                 pr_debug(" <- fw_cmo_feature_init()\n");
 879                 return;
 880         }
 881
 882         end = rtas_data_buf + CMO_MAXLENGTH - 2;
 883         ptr = rtas_data_buf + 2;        /* step over strlen value */
 884         key = value = ptr;
 885
 886         while (*ptr && (ptr <= end)) {
 887                 /* Separate the key and value by replacing '=' with '\0' and
 888                  * point the value at the string after the '='
 889                  */
 890                 if (ptr[0] == '=') {
 891                         ptr[0] = '\0';
 892                         value = ptr + 1;
 893                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
 894                         /* Terminate the string containing the key/value pair */
 895                         ptr[0] = '\0';
 896
 897                         if (key == value) {
 898                                 pr_debug("Malformed key/value pair\n");
 899                                 /* Never found a '=', end processing */
 900                                 break;
 901                         }
 902
 903                         if (0 == strcmp(key, "CMOPageSize"))
 904                                 page_order = simple_strtol(value, NULL, 10);
 905                         else if (0 == strcmp(key, "PrPSP"))
 906                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
 907                         else if (0 == strcmp(key, "SecPSP"))
 908                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
 909                         value = key = ptr + 1;
 910                 }
 911                 ptr++;
 912         }
 913
 914         /* Page size is returned as the power of 2 of the page size,
 915          * convert to the page size in bytes before returning
 916          */
 917         CMO_PageSize = 1 << page_order;
 918         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
 919
 920         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
 921                 pr_info("CMO enabled\n");
 922                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 923                          CMO_SecPSP);
 924                 powerpc_firmware_features |= FW_FEATURE_CMO;
 925                 pSeries_coalesce_init();
 926         } else
 927                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 928                          CMO_SecPSP);
 929         spin_unlock(&rtas_data_buf_lock);
 930         pr_debug(" <- fw_cmo_feature_init()\n");
 931 }
 932
 933 /*
 934  * Early initialization.  Relocation is on but do not reference unbolted pages
 935  */
 936 static void __init pseries_init(void)
 937 {
 938         pr_debug(" -> pseries_init()\n");
 939
 940 #ifdef CONFIG_HVC_CONSOLE
 941         if (firmware_has_feature(FW_FEATURE_LPAR))
 942                 hvc_vio_init_early();
 943 #endif
 944         if (firmware_has_feature(FW_FEATURE_XDABR))
 945                 ppc_md.set_dabr = pseries_set_xdabr;
 946         else if (firmware_has_feature(FW_FEATURE_DABR))
 947                 ppc_md.set_dabr = pseries_set_dabr;
 948
 949         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 950                 ppc_md.set_dawr = pseries_set_dawr;
 951
 952         pSeries_cmo_feature_init();
 953         iommu_init_early_pSeries();
 954
 955         pr_debug(" <- pseries_init()\n");
 956 }
 957
 958 /**
 959  * pseries_power_off - tell firmware about how to power off the system.
 960  *
 961  * This function calls either the power-off rtas token in normal cases
 962  * or the ibm,power-off-ups token (if present & requested) in case of
 963  * a power failure. If power-off token is used, power on will only be
 964  * possible with power button press. If ibm,power-off-ups token is used
 965  * it will allow auto poweron after power is restored.
 966  */
 967 static void pseries_power_off(void)
 968 {
 969         int rc;
 970         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
 971
 972         if (rtas_flash_term_hook)
 973                 rtas_flash_term_hook(SYS_POWER_OFF);
 974
 975         if (rtas_poweron_auto == 0 ||
 976                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
 977                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
 978                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
 979         } else {
 980                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
 981                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
 982         }
 983         for (;;);
 984 }
 985
 986 static int __init pSeries_probe(void)
 987 {
 988         if (!of_node_is_type(of_root, "chrp"))
 989                 return 0;
 990
 991         /* Cell blades firmware claims to be chrp while it's not. Until this
 992          * is fixed, we need to avoid those here.
 993          */
 994         if (of_machine_is_compatible("IBM,CPBW-1.0") ||
 995             of_machine_is_compatible("IBM,CBEA"))
 996                 return 0;
 997
 998         pm_power_off = pseries_power_off;
 999
1000         pr_debug("Machine is%s LPAR !\n",
1001                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
1002
1003         pseries_init();
1004
1005         return 1;
1006 }
1007
1008 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1009 {
1010         if (firmware_has_feature(FW_FEATURE_LPAR))
1011                 return PCI_PROBE_DEVTREE;
1012         return PCI_PROBE_NORMAL;
1013 }
1014
1015 struct pci_controller_ops pseries_pci_controller_ops = {
1016         .probe_mode             = pSeries_pci_probe_mode,
1017 };
1018
1019 define_machine(pseries) {
1020         .name                   = "pSeries",
1021         .probe                  = pSeries_probe,
1022         .setup_arch             = pSeries_setup_arch,
1023         .init_IRQ               = pseries_init_irq,
1024         .show_cpuinfo           = pSeries_show_cpuinfo,
1025         .log_error              = pSeries_log_error,
1026         .pcibios_fixup          = pSeries_final_fixup,
1027         .restart                = rtas_restart,
1028         .halt                   = rtas_halt,
1029         .panic                  = pseries_panic,
1030         .get_boot_time          = rtas_get_boot_time,
1031         .get_rtc_time           = rtas_get_rtc_time,
1032         .set_rtc_time           = rtas_set_rtc_time,
1033         .calibrate_decr         = generic_calibrate_decr,
1034         .progress               = rtas_progress,
1035         .system_reset_exception = pSeries_system_reset_exception,
1036         .machine_check_early    = pseries_machine_check_realmode,
1037         .machine_check_exception = pSeries_machine_check_exception,
1038 #ifdef CONFIG_KEXEC_CORE
1039         .machine_kexec          = pSeries_machine_kexec,
1040         .kexec_cpu_down         = pseries_kexec_cpu_down,
1041 #endif
1042 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
1043         .memory_block_size      = pseries_memory_block_size,
1044 #endif
1045 };