arch/powerpc/platforms/pseries/setup.c

   1 /*
   2  *  64-bit pSeries and RS/6000 setup code.
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *  Adapted from 'alpha' version by Gary Thomas
   6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
   7  *  Modified by PPC64 Team, IBM Corp
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License
  11  * as published by the Free Software Foundation; either version
  12  * 2 of the License, or (at your option) any later version.
  13  */
  14
  15 /*
  16  * bootup setup stuff..
  17  */
  18
  19 #include <linux/cpu.h>
  20 #include <linux/errno.h>
  21 #include <linux/sched.h>
  22 #include <linux/kernel.h>
  23 #include <linux/mm.h>
  24 #include <linux/stddef.h>
  25 #include <linux/unistd.h>
  26 #include <linux/user.h>
  27 #include <linux/tty.h>
  28 #include <linux/major.h>
  29 #include <linux/interrupt.h>
  30 #include <linux/reboot.h>
  31 #include <linux/init.h>
  32 #include <linux/ioport.h>
  33 #include <linux/console.h>
  34 #include <linux/pci.h>
  35 #include <linux/utsname.h>
  36 #include <linux/adb.h>
  37 #include <linux/export.h>
  38 #include <linux/delay.h>
  39 #include <linux/irq.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/root_dev.h>
  42 #include <linux/of.h>
  43 #include <linux/of_pci.h>
  44 #include <linux/memblock.h>
  45
  46 #include <asm/mmu.h>
  47 #include <asm/processor.h>
  48 #include <asm/io.h>
  49 #include <asm/pgtable.h>
  50 #include <asm/prom.h>
  51 #include <asm/rtas.h>
  52 #include <asm/pci-bridge.h>
  53 #include <asm/iommu.h>
  54 #include <asm/dma.h>
  55 #include <asm/machdep.h>
  56 #include <asm/irq.h>
  57 #include <asm/time.h>
  58 #include <asm/nvram.h>
  59 #include <asm/pmc.h>
  60 #include <asm/xics.h>
  61 #include <asm/xive.h>
  62 #include <asm/ppc-pci.h>
  63 #include <asm/i8259.h>
  64 #include <asm/udbg.h>
  65 #include <asm/smp.h>
  66 #include <asm/firmware.h>
  67 #include <asm/eeh.h>
  68 #include <asm/reg.h>
  69 #include <asm/plpar_wrappers.h>
  70 #include <asm/kexec.h>
  71 #include <asm/isa-bridge.h>
  72 #include <asm/security_features.h>
  73 #include <asm/asm-const.h>
  74
  75 #include "pseries.h"
  76 #include "../../../../drivers/pci/pci.h"
  77
  78 int CMO_PrPSP = -1;
  79 int CMO_SecPSP = -1;
  80 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
  81 EXPORT_SYMBOL(CMO_PageSize);
  82
  83 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
  84
  85 static void pSeries_show_cpuinfo(struct seq_file *m)
  86 {
  87         struct device_node *root;
  88         const char *model = "";
  89
  90         root = of_find_node_by_path("/");
  91         if (root)
  92                 model = of_get_property(root, "model", NULL);
  93         seq_printf(m, "machine\t\t: CHRP %s\n", model);
  94         of_node_put(root);
  95         if (radix_enabled())
  96                 seq_printf(m, "MMU\t\t: Radix\n");
  97         else
  98                 seq_printf(m, "MMU\t\t: Hash\n");
  99 }
 100
 101 /* Initialize firmware assisted non-maskable interrupts if
 102  * the firmware supports this feature.
 103  */
 104 static void __init fwnmi_init(void)
 105 {
 106         unsigned long system_reset_addr, machine_check_addr;
 107         u8 *mce_data_buf;
 108         unsigned int i;
 109         int nr_cpus = num_possible_cpus();
 110 #ifdef CONFIG_PPC_BOOK3S_64
 111         struct slb_entry *slb_ptr;
 112         size_t size;
 113 #endif
 114
 115         int ibm_nmi_register = rtas_token("ibm,nmi-register");
 116         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
 117                 return;
 118
 119         /* If the kernel's not linked at zero we point the firmware at low
 120          * addresses anyway, and use a trampoline to get to the real code. */
 121         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
 122         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
 123
 124         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
 125                                 machine_check_addr))
 126                 fwnmi_active = 1;
 127
 128         /*
 129          * Allocate a chunk for per cpu buffer to hold rtas errorlog.
 130          * It will be used in real mode mce handler, hence it needs to be
 131          * below RMA.
 132          */
 133         mce_data_buf = __va(memblock_alloc_base(RTAS_ERROR_LOG_MAX * nr_cpus,
 134                                         RTAS_ERROR_LOG_MAX, ppc64_rma_size));
 135         for_each_possible_cpu(i) {
 136                 paca_ptrs[i]->mce_data_buf = mce_data_buf +
 137                                                 (RTAS_ERROR_LOG_MAX * i);
 138         }
 139
 140 #ifdef CONFIG_PPC_BOOK3S_64
 141         /* Allocate per cpu slb area to save old slb contents during MCE */
 142         size = sizeof(struct slb_entry) * mmu_slb_size * nr_cpus;
 143         slb_ptr = __va(memblock_alloc_base(size, sizeof(struct slb_entry),
 144                                            ppc64_rma_size));
 145         for_each_possible_cpu(i)
 146                 paca_ptrs[i]->mce_faulty_slbs = slb_ptr + (mmu_slb_size * i);
 147 #endif
 148 }
 149
 150 static void pseries_8259_cascade(struct irq_desc *desc)
 151 {
 152         struct irq_chip *chip = irq_desc_get_chip(desc);
 153         unsigned int cascade_irq = i8259_irq();
 154
 155         if (cascade_irq)
 156                 generic_handle_irq(cascade_irq);
 157
 158         chip->irq_eoi(&desc->irq_data);
 159 }
 160
 161 static void __init pseries_setup_i8259_cascade(void)
 162 {
 163         struct device_node *np, *old, *found = NULL;
 164         unsigned int cascade;
 165         const u32 *addrp;
 166         unsigned long intack = 0;
 167         int naddr;
 168
 169         for_each_node_by_type(np, "interrupt-controller") {
 170                 if (of_device_is_compatible(np, "chrp,iic")) {
 171                         found = np;
 172                         break;
 173                 }
 174         }
 175
 176         if (found == NULL) {
 177                 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
 178                 return;
 179         }
 180
 181         cascade = irq_of_parse_and_map(found, 0);
 182         if (!cascade) {
 183                 printk(KERN_ERR "pic: failed to map cascade interrupt");
 184                 return;
 185         }
 186         pr_debug("pic: cascade mapped to irq %d\n", cascade);
 187
 188         for (old = of_node_get(found); old != NULL ; old = np) {
 189                 np = of_get_parent(old);
 190                 of_node_put(old);
 191                 if (np == NULL)
 192                         break;
 193                 if (strcmp(np->name, "pci") != 0)
 194                         continue;
 195                 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
 196                 if (addrp == NULL)
 197                         continue;
 198                 naddr = of_n_addr_cells(np);
 199                 intack = addrp[naddr-1];
 200                 if (naddr > 1)
 201                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
 202         }
 203         if (intack)
 204                 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
 205         i8259_init(found, intack);
 206         of_node_put(found);
 207         irq_set_chained_handler(cascade, pseries_8259_cascade);
 208 }
 209
 210 static void __init pseries_init_irq(void)
 211 {
 212         /* Try using a XIVE if available, otherwise use a XICS */
 213         if (!xive_spapr_init()) {
 214                 xics_init();
 215                 pseries_setup_i8259_cascade();
 216         }
 217 }
 218
 219 static void pseries_lpar_enable_pmcs(void)
 220 {
 221         unsigned long set, reset;
 222
 223         set = 1UL << 63;
 224         reset = 0;
 225         plpar_hcall_norets(H_PERFMON, set, reset);
 226 }
 227
 228 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
 229 {
 230         struct of_reconfig_data *rd = data;
 231         struct device_node *parent, *np = rd->dn;
 232         struct pci_dn *pdn;
 233         int err = NOTIFY_OK;
 234
 235         switch (action) {
 236         case OF_RECONFIG_ATTACH_NODE:
 237                 parent = of_get_parent(np);
 238                 pdn = parent ? PCI_DN(parent) : NULL;
 239                 if (pdn)
 240                         pci_add_device_node_info(pdn->phb, np);
 241
 242                 of_node_put(parent);
 243                 break;
 244         case OF_RECONFIG_DETACH_NODE:
 245                 pdn = PCI_DN(np);
 246                 if (pdn)
 247                         list_del(&pdn->list);
 248                 break;
 249         default:
 250                 err = NOTIFY_DONE;
 251                 break;
 252         }
 253         return err;
 254 }
 255
 256 static struct notifier_block pci_dn_reconfig_nb = {
 257         .notifier_call = pci_dn_reconfig_notifier,
 258 };
 259
 260 struct kmem_cache *dtl_cache;
 261
 262 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 263 /*
 264  * Allocate space for the dispatch trace log for all possible cpus
 265  * and register the buffers with the hypervisor.  This is used for
 266  * computing time stolen by the hypervisor.
 267  */
 268 static int alloc_dispatch_logs(void)
 269 {
 270         int cpu, ret;
 271         struct paca_struct *pp;
 272         struct dtl_entry *dtl;
 273
 274         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 275                 return 0;
 276
 277         if (!dtl_cache)
 278                 return 0;
 279
 280         for_each_possible_cpu(cpu) {
 281                 pp = paca_ptrs[cpu];
 282                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
 283                 if (!dtl) {
 284                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
 285                                 cpu);
 286                         pr_warn("Stolen time statistics will be unreliable\n");
 287                         break;
 288                 }
 289
 290                 pp->dtl_ridx = 0;
 291                 pp->dispatch_log = dtl;
 292                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
 293                 pp->dtl_curr = dtl;
 294         }
 295
 296         /* Register the DTL for the current (boot) cpu */
 297         dtl = get_paca()->dispatch_log;
 298         get_paca()->dtl_ridx = 0;
 299         get_paca()->dtl_curr = dtl;
 300         get_paca()->lppaca_ptr->dtl_idx = 0;
 301
 302         /* hypervisor reads buffer length from this field */
 303         dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 304         ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
 305         if (ret)
 306                 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
 307                        "with %d\n", smp_processor_id(),
 308                        hard_smp_processor_id(), ret);
 309         get_paca()->lppaca_ptr->dtl_enable_mask = 2;
 310
 311         return 0;
 312 }
 313 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 314 static inline int alloc_dispatch_logs(void)
 315 {
 316         return 0;
 317 }
 318 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 319
 320 static int alloc_dispatch_log_kmem_cache(void)
 321 {
 322         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
 323                                                 DISPATCH_LOG_BYTES, 0, NULL);
 324         if (!dtl_cache) {
 325                 pr_warn("Failed to create dispatch trace log buffer cache\n");
 326                 pr_warn("Stolen time statistics will be unreliable\n");
 327                 return 0;
 328         }
 329
 330         return alloc_dispatch_logs();
 331 }
 332 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
 333
 334 static void pseries_lpar_idle(void)
 335 {
 336         /*
 337          * Default handler to go into low thread priority and possibly
 338          * low power mode by ceding processor to hypervisor
 339          */
 340
 341         /* Indicate to hypervisor that we are idle. */
 342         get_lppaca()->idle = 1;
 343
 344         /*
 345          * Yield the processor to the hypervisor.  We return if
 346          * an external interrupt occurs (which are driven prior
 347          * to returning here) or if a prod occurs from another
 348          * processor. When returning here, external interrupts
 349          * are enabled.
 350          */
 351         cede_processor();
 352
 353         get_lppaca()->idle = 0;
 354 }
 355
 356 /*
 357  * Enable relocation on during exceptions. This has partition wide scope and
 358  * may take a while to complete, if it takes longer than one second we will
 359  * just give up rather than wasting any more time on this - if that turns out
 360  * to ever be a problem in practice we can move this into a kernel thread to
 361  * finish off the process later in boot.
 362  */
 363 void pseries_enable_reloc_on_exc(void)
 364 {
 365         long rc;
 366         unsigned int delay, total_delay = 0;
 367
 368         while (1) {
 369                 rc = enable_reloc_on_exceptions();
 370                 if (!H_IS_LONG_BUSY(rc)) {
 371                         if (rc == H_P2) {
 372                                 pr_info("Relocation on exceptions not"
 373                                         " supported\n");
 374                         } else if (rc != H_SUCCESS) {
 375                                 pr_warn("Unable to enable relocation"
 376                                         " on exceptions: %ld\n", rc);
 377                         }
 378                         break;
 379                 }
 380
 381                 delay = get_longbusy_msecs(rc);
 382                 total_delay += delay;
 383                 if (total_delay > 1000) {
 384                         pr_warn("Warning: Giving up waiting to enable "
 385                                 "relocation on exceptions (%u msec)!\n",
 386                                 total_delay);
 387                         return;
 388                 }
 389
 390                 mdelay(delay);
 391         }
 392 }
 393 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
 394
 395 void pseries_disable_reloc_on_exc(void)
 396 {
 397         long rc;
 398
 399         while (1) {
 400                 rc = disable_reloc_on_exceptions();
 401                 if (!H_IS_LONG_BUSY(rc))
 402                         break;
 403                 mdelay(get_longbusy_msecs(rc));
 404         }
 405         if (rc != H_SUCCESS)
 406                 pr_warn("Warning: Failed to disable relocation on exceptions: %ld\n",
 407                         rc);
 408 }
 409 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
 410
 411 #ifdef CONFIG_KEXEC_CORE
 412 static void pSeries_machine_kexec(struct kimage *image)
 413 {
 414         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 415                 pseries_disable_reloc_on_exc();
 416
 417         default_machine_kexec(image);
 418 }
 419 #endif
 420
 421 #ifdef __LITTLE_ENDIAN__
 422 void pseries_big_endian_exceptions(void)
 423 {
 424         long rc;
 425
 426         while (1) {
 427                 rc = enable_big_endian_exceptions();
 428                 if (!H_IS_LONG_BUSY(rc))
 429                         break;
 430                 mdelay(get_longbusy_msecs(rc));
 431         }
 432
 433         /*
 434          * At this point it is unlikely panic() will get anything
 435          * out to the user, since this is called very late in kexec
 436          * but at least this will stop us from continuing on further
 437          * and creating an even more difficult to debug situation.
 438          *
 439          * There is a known problem when kdump'ing, if cpus are offline
 440          * the above call will fail. Rather than panicking again, keep
 441          * going and hope the kdump kernel is also little endian, which
 442          * it usually is.
 443          */
 444         if (rc && !kdump_in_progress())
 445                 panic("Could not enable big endian exceptions");
 446 }
 447
 448 void pseries_little_endian_exceptions(void)
 449 {
 450         long rc;
 451
 452         while (1) {
 453                 rc = enable_little_endian_exceptions();
 454                 if (!H_IS_LONG_BUSY(rc))
 455                         break;
 456                 mdelay(get_longbusy_msecs(rc));
 457         }
 458         if (rc) {
 459                 ppc_md.progress("H_SET_MODE LE exception fail", 0);
 460                 panic("Could not enable little endian exceptions");
 461         }
 462 }
 463 #endif
 464
 465 static void __init find_and_init_phbs(void)
 466 {
 467         struct device_node *node;
 468         struct pci_controller *phb;
 469         struct device_node *root = of_find_node_by_path("/");
 470
 471         for_each_child_of_node(root, node) {
 472                 if (node->type == NULL || (strcmp(node->type, "pci") != 0 &&
 473                                            strcmp(node->type, "pciex") != 0))
 474                         continue;
 475
 476                 phb = pcibios_alloc_controller(node);
 477                 if (!phb)
 478                         continue;
 479                 rtas_setup_phb(phb);
 480                 pci_process_bridge_OF_ranges(phb, node, 0);
 481                 isa_bridge_find_early(phb);
 482                 phb->controller_ops = pseries_pci_controller_ops;
 483         }
 484
 485         of_node_put(root);
 486
 487         /*
 488          * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
 489          * in chosen.
 490          */
 491         of_pci_check_probe_only();
 492 }
 493
 494 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
 495 {
 496         /*
 497          * The features below are disabled by default, so we instead look to see
 498          * if firmware has *enabled* them, and set them if so.
 499          */
 500         if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
 501                 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
 502
 503         if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
 504                 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
 505
 506         if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
 507                 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
 508
 509         if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
 510                 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
 511
 512         if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
 513                 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
 514
 515         if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
 516                 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
 517
 518         if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
 519                 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
 520
 521         if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
 522                 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
 523
 524         /*
 525          * The features below are enabled by default, so we instead look to see
 526          * if firmware has *disabled* them, and clear them if so.
 527          */
 528         if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
 529                 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
 530
 531         if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
 532                 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
 533
 534         if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
 535                 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
 536 }
 537
 538 void pseries_setup_rfi_flush(void)
 539 {
 540         struct h_cpu_char_result result;
 541         enum l1d_flush_type types;
 542         bool enable;
 543         long rc;
 544
 545         /*
 546          * Set features to the defaults assumed by init_cpu_char_feature_flags()
 547          * so it can set/clear again any features that might have changed after
 548          * migration, and in case the hypercall fails and it is not even called.
 549          */
 550         powerpc_security_features = SEC_FTR_DEFAULT;
 551
 552         rc = plpar_get_cpu_characteristics(&result);
 553         if (rc == H_SUCCESS)
 554                 init_cpu_char_feature_flags(&result);
 555
 556         /*
 557          * We're the guest so this doesn't apply to us, clear it to simplify
 558          * handling of it elsewhere.
 559          */
 560         security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
 561
 562         types = L1D_FLUSH_FALLBACK;
 563
 564         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
 565                 types |= L1D_FLUSH_MTTRIG;
 566
 567         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
 568                 types |= L1D_FLUSH_ORI;
 569
 570         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
 571                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
 572
 573         setup_rfi_flush(types, enable);
 574         setup_count_cache_flush();
 575 }
 576
 577 #ifdef CONFIG_PCI_IOV
 578 enum rtas_iov_fw_value_map {
 579         NUM_RES_PROPERTY  = 0, /* Number of Resources */
 580         LOW_INT           = 1, /* Lowest 32 bits of Address */
 581         START_OF_ENTRIES  = 2, /* Always start of entry */
 582         APERTURE_PROPERTY = 2, /* Start of entry+ to  Aperture Size */
 583         WDW_SIZE_PROPERTY = 4, /* Start of entry+ to Window Size */
 584         NEXT_ENTRY        = 7  /* Go to next entry on array */
 585 };
 586
 587 enum get_iov_fw_value_index {
 588         BAR_ADDRS     = 1,    /*  Get Bar Address */
 589         APERTURE_SIZE = 2,    /*  Get Aperture Size */
 590         WDW_SIZE      = 3     /*  Get Window Size */
 591 };
 592
 593 resource_size_t pseries_get_iov_fw_value(struct pci_dev *dev, int resno,
 594                                          enum get_iov_fw_value_index value)
 595 {
 596         const int *indexes;
 597         struct device_node *dn = pci_device_to_OF_node(dev);
 598         int i, num_res, ret = 0;
 599
 600         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 601         if (!indexes)
 602                 return  0;
 603
 604         /*
 605          * First element in the array is the number of Bars
 606          * returned.  Search through the list to find the matching
 607          * bar
 608          */
 609         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 610         if (resno >= num_res)
 611                 return 0; /* or an errror */
 612
 613         i = START_OF_ENTRIES + NEXT_ENTRY * resno;
 614         switch (value) {
 615         case BAR_ADDRS:
 616                 ret = of_read_number(&indexes[i], 2);
 617                 break;
 618         case APERTURE_SIZE:
 619                 ret = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
 620                 break;
 621         case WDW_SIZE:
 622                 ret = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
 623                 break;
 624         }
 625
 626         return ret;
 627 }
 628
 629 void of_pci_set_vf_bar_size(struct pci_dev *dev, const int *indexes)
 630 {
 631         struct resource *res;
 632         resource_size_t base, size;
 633         int i, r, num_res;
 634
 635         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 636         num_res = min_t(int, num_res, PCI_SRIOV_NUM_BARS);
 637         for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
 638              i += NEXT_ENTRY, r++) {
 639                 res = &dev->resource[r + PCI_IOV_RESOURCES];
 640                 base = of_read_number(&indexes[i], 2);
 641                 size = of_read_number(&indexes[i + APERTURE_PROPERTY], 2);
 642                 res->flags = pci_parse_of_flags(of_read_number
 643                                                 (&indexes[i + LOW_INT], 1), 0);
 644                 res->flags |= (IORESOURCE_MEM_64 | IORESOURCE_PCI_FIXED);
 645                 res->name = pci_name(dev);
 646                 res->start = base;
 647                 res->end = base + size - 1;
 648         }
 649 }
 650
 651 void of_pci_parse_iov_addrs(struct pci_dev *dev, const int *indexes)
 652 {
 653         struct resource *res, *root, *conflict;
 654         resource_size_t base, size;
 655         int i, r, num_res;
 656
 657         /*
 658          * First element in the array is the number of Bars
 659          * returned.  Search through the list to find the matching
 660          * bars assign them from firmware into resources structure.
 661          */
 662         num_res = of_read_number(&indexes[NUM_RES_PROPERTY], 1);
 663         for (i = START_OF_ENTRIES, r = 0; r < num_res && r < PCI_SRIOV_NUM_BARS;
 664              i += NEXT_ENTRY, r++) {
 665                 res = &dev->resource[r + PCI_IOV_RESOURCES];
 666                 base = of_read_number(&indexes[i], 2);
 667                 size = of_read_number(&indexes[i + WDW_SIZE_PROPERTY], 2);
 668                 res->name = pci_name(dev);
 669                 res->start = base;
 670                 res->end = base + size - 1;
 671                 root = &iomem_resource;
 672                 dev_dbg(&dev->dev,
 673                         "pSeries IOV BAR %d: trying firmware assignment %pR\n",
 674                          r + PCI_IOV_RESOURCES, res);
 675                 conflict = request_resource_conflict(root, res);
 676                 if (conflict) {
 677                         dev_info(&dev->dev,
 678                                  "BAR %d: %pR conflicts with %s %pR\n",
 679                                  r + PCI_IOV_RESOURCES, res,
 680                                  conflict->name, conflict);
 681                         res->flags |= IORESOURCE_UNSET;
 682                 }
 683         }
 684 }
 685
 686 static void pseries_disable_sriov_resources(struct pci_dev *pdev)
 687 {
 688         int i;
 689
 690         pci_warn(pdev, "No hypervisor support for SR-IOV on this device, IOV BARs disabled.\n");
 691         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
 692                 pdev->resource[i + PCI_IOV_RESOURCES].flags = 0;
 693 }
 694
 695 static void pseries_pci_fixup_resources(struct pci_dev *pdev)
 696 {
 697         const int *indexes;
 698         struct device_node *dn = pci_device_to_OF_node(pdev);
 699
 700         /*Firmware must support open sriov otherwise dont configure*/
 701         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 702         if (indexes)
 703                 of_pci_set_vf_bar_size(pdev, indexes);
 704         else
 705                 pseries_disable_sriov_resources(pdev);
 706 }
 707
 708 static void pseries_pci_fixup_iov_resources(struct pci_dev *pdev)
 709 {
 710         const int *indexes;
 711         struct device_node *dn = pci_device_to_OF_node(pdev);
 712
 713         if (!pdev->is_physfn || pci_dev_is_added(pdev))
 714                 return;
 715         /*Firmware must support open sriov otherwise dont configure*/
 716         indexes = of_get_property(dn, "ibm,open-sriov-vf-bar-info", NULL);
 717         if (indexes)
 718                 of_pci_parse_iov_addrs(pdev, indexes);
 719         else
 720                 pseries_disable_sriov_resources(pdev);
 721 }
 722
 723 static resource_size_t pseries_pci_iov_resource_alignment(struct pci_dev *pdev,
 724                                                           int resno)
 725 {
 726         const __be32 *reg;
 727         struct device_node *dn = pci_device_to_OF_node(pdev);
 728
 729         /*Firmware must support open sriov otherwise report regular alignment*/
 730         reg = of_get_property(dn, "ibm,is-open-sriov-pf", NULL);
 731         if (!reg)
 732                 return pci_iov_resource_size(pdev, resno);
 733
 734         if (!pdev->is_physfn)
 735                 return 0;
 736         return pseries_get_iov_fw_value(pdev,
 737                                         resno - PCI_IOV_RESOURCES,
 738                                         APERTURE_SIZE);
 739 }
 740 #endif
 741
 742 static void __init pSeries_setup_arch(void)
 743 {
 744         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
 745
 746         /* Discover PIC type and setup ppc_md accordingly */
 747         smp_init_pseries();
 748
 749
 750         /* openpic global configuration register (64-bit format). */
 751         /* openpic Interrupt Source Unit pointer (64-bit format). */
 752         /* python0 facility area (mmio) (64-bit format) REAL address. */
 753
 754         /* init to some ~sane value until calibrate_delay() runs */
 755         loops_per_jiffy = 50000000;
 756
 757         fwnmi_init();
 758
 759         pseries_setup_rfi_flush();
 760         setup_stf_barrier();
 761
 762         /* By default, only probe PCI (can be overridden by rtas_pci) */
 763         pci_add_flags(PCI_PROBE_ONLY);
 764
 765         /* Find and initialize PCI host bridges */
 766         init_pci_config_tokens();
 767         find_and_init_phbs();
 768         of_reconfig_notifier_register(&pci_dn_reconfig_nb);
 769
 770         pSeries_nvram_init();
 771
 772         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 773                 vpa_init(boot_cpuid);
 774                 ppc_md.power_save = pseries_lpar_idle;
 775                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
 776 #ifdef CONFIG_PCI_IOV
 777                 ppc_md.pcibios_fixup_resources =
 778                         pseries_pci_fixup_resources;
 779                 ppc_md.pcibios_fixup_sriov =
 780                         pseries_pci_fixup_iov_resources;
 781                 ppc_md.pcibios_iov_resource_alignment =
 782                         pseries_pci_iov_resource_alignment;
 783 #endif
 784         } else {
 785                 /* No special idle routine */
 786                 ppc_md.enable_pmcs = power4_enable_pmcs;
 787         }
 788
 789         ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
 790 }
 791
 792 static void pseries_panic(char *str)
 793 {
 794         panic_flush_kmsg_end();
 795         rtas_os_term(str);
 796 }
 797
 798 static int __init pSeries_init_panel(void)
 799 {
 800         /* Manually leave the kernel version on the panel. */
 801 #ifdef __BIG_ENDIAN__
 802         ppc_md.progress("Linux ppc64\n", 0);
 803 #else
 804         ppc_md.progress("Linux ppc64le\n", 0);
 805 #endif
 806         ppc_md.progress(init_utsname()->version, 0);
 807
 808         return 0;
 809 }
 810 machine_arch_initcall(pseries, pSeries_init_panel);
 811
 812 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
 813 {
 814         return plpar_hcall_norets(H_SET_DABR, dabr);
 815 }
 816
 817 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
 818 {
 819         /* Have to set at least one bit in the DABRX according to PAPR */
 820         if (dabrx == 0 && dabr == 0)
 821                 dabrx = DABRX_USER;
 822         /* PAPR says we can only set kernel and user bits */
 823         dabrx &= DABRX_KERNEL | DABRX_USER;
 824
 825         return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
 826 }
 827
 828 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
 829 {
 830         /* PAPR says we can't set HYP */
 831         dawrx &= ~DAWRX_HYP;
 832
 833         return  plpar_set_watchpoint0(dawr, dawrx);
 834 }
 835
 836 #define CMO_CHARACTERISTICS_TOKEN 44
 837 #define CMO_MAXLENGTH 1026
 838
 839 void pSeries_coalesce_init(void)
 840 {
 841         struct hvcall_mpp_x_data mpp_x_data;
 842
 843         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
 844                 powerpc_firmware_features |= FW_FEATURE_XCMO;
 845         else
 846                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
 847 }
 848
 849 /**
 850  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
 851  * handle that here. (Stolen from parse_system_parameter_string)
 852  */
 853 static void pSeries_cmo_feature_init(void)
 854 {
 855         char *ptr, *key, *value, *end;
 856         int call_status;
 857         int page_order = IOMMU_PAGE_SHIFT_4K;
 858
 859         pr_debug(" -> fw_cmo_feature_init()\n");
 860         spin_lock(&rtas_data_buf_lock);
 861         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
 862         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 863                                 NULL,
 864                                 CMO_CHARACTERISTICS_TOKEN,
 865                                 __pa(rtas_data_buf),
 866                                 RTAS_DATA_BUF_SIZE);
 867
 868         if (call_status != 0) {
 869                 spin_unlock(&rtas_data_buf_lock);
 870                 pr_debug("CMO not available\n");
 871                 pr_debug(" <- fw_cmo_feature_init()\n");
 872                 return;
 873         }
 874
 875         end = rtas_data_buf + CMO_MAXLENGTH - 2;
 876         ptr = rtas_data_buf + 2;        /* step over strlen value */
 877         key = value = ptr;
 878
 879         while (*ptr && (ptr <= end)) {
 880                 /* Separate the key and value by replacing '=' with '\0' and
 881                  * point the value at the string after the '='
 882                  */
 883                 if (ptr[0] == '=') {
 884                         ptr[0] = '\0';
 885                         value = ptr + 1;
 886                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
 887                         /* Terminate the string containing the key/value pair */
 888                         ptr[0] = '\0';
 889
 890                         if (key == value) {
 891                                 pr_debug("Malformed key/value pair\n");
 892                                 /* Never found a '=', end processing */
 893                                 break;
 894                         }
 895
 896                         if (0 == strcmp(key, "CMOPageSize"))
 897                                 page_order = simple_strtol(value, NULL, 10);
 898                         else if (0 == strcmp(key, "PrPSP"))
 899                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
 900                         else if (0 == strcmp(key, "SecPSP"))
 901                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
 902                         value = key = ptr + 1;
 903                 }
 904                 ptr++;
 905         }
 906
 907         /* Page size is returned as the power of 2 of the page size,
 908          * convert to the page size in bytes before returning
 909          */
 910         CMO_PageSize = 1 << page_order;
 911         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
 912
 913         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
 914                 pr_info("CMO enabled\n");
 915                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 916                          CMO_SecPSP);
 917                 powerpc_firmware_features |= FW_FEATURE_CMO;
 918                 pSeries_coalesce_init();
 919         } else
 920                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 921                          CMO_SecPSP);
 922         spin_unlock(&rtas_data_buf_lock);
 923         pr_debug(" <- fw_cmo_feature_init()\n");
 924 }
 925
 926 /*
 927  * Early initialization.  Relocation is on but do not reference unbolted pages
 928  */
 929 static void __init pseries_init(void)
 930 {
 931         pr_debug(" -> pseries_init()\n");
 932
 933 #ifdef CONFIG_HVC_CONSOLE
 934         if (firmware_has_feature(FW_FEATURE_LPAR))
 935                 hvc_vio_init_early();
 936 #endif
 937         if (firmware_has_feature(FW_FEATURE_XDABR))
 938                 ppc_md.set_dabr = pseries_set_xdabr;
 939         else if (firmware_has_feature(FW_FEATURE_DABR))
 940                 ppc_md.set_dabr = pseries_set_dabr;
 941
 942         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 943                 ppc_md.set_dawr = pseries_set_dawr;
 944
 945         pSeries_cmo_feature_init();
 946         iommu_init_early_pSeries();
 947
 948         pr_debug(" <- pseries_init()\n");
 949 }
 950
 951 /**
 952  * pseries_power_off - tell firmware about how to power off the system.
 953  *
 954  * This function calls either the power-off rtas token in normal cases
 955  * or the ibm,power-off-ups token (if present & requested) in case of
 956  * a power failure. If power-off token is used, power on will only be
 957  * possible with power button press. If ibm,power-off-ups token is used
 958  * it will allow auto poweron after power is restored.
 959  */
 960 static void pseries_power_off(void)
 961 {
 962         int rc;
 963         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
 964
 965         if (rtas_flash_term_hook)
 966                 rtas_flash_term_hook(SYS_POWER_OFF);
 967
 968         if (rtas_poweron_auto == 0 ||
 969                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
 970                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
 971                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
 972         } else {
 973                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
 974                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
 975         }
 976         for (;;);
 977 }
 978
 979 static int __init pSeries_probe(void)
 980 {
 981         const char *dtype = of_get_property(of_root, "device_type", NULL);
 982
 983         if (dtype == NULL)
 984                 return 0;
 985         if (strcmp(dtype, "chrp"))
 986                 return 0;
 987
 988         /* Cell blades firmware claims to be chrp while it's not. Until this
 989          * is fixed, we need to avoid those here.
 990          */
 991         if (of_machine_is_compatible("IBM,CPBW-1.0") ||
 992             of_machine_is_compatible("IBM,CBEA"))
 993                 return 0;
 994
 995         pm_power_off = pseries_power_off;
 996
 997         pr_debug("Machine is%s LPAR !\n",
 998                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
 999
1000         pseries_init();
1001
1002         return 1;
1003 }
1004
1005 static int pSeries_pci_probe_mode(struct pci_bus *bus)
1006 {
1007         if (firmware_has_feature(FW_FEATURE_LPAR))
1008                 return PCI_PROBE_DEVTREE;
1009         return PCI_PROBE_NORMAL;
1010 }
1011
1012 struct pci_controller_ops pseries_pci_controller_ops = {
1013         .probe_mode             = pSeries_pci_probe_mode,
1014 };
1015
1016 define_machine(pseries) {
1017         .name                   = "pSeries",
1018         .probe                  = pSeries_probe,
1019         .setup_arch             = pSeries_setup_arch,
1020         .init_IRQ               = pseries_init_irq,
1021         .show_cpuinfo           = pSeries_show_cpuinfo,
1022         .log_error              = pSeries_log_error,
1023         .pcibios_fixup          = pSeries_final_fixup,
1024         .restart                = rtas_restart,
1025         .halt                   = rtas_halt,
1026         .panic                  = pseries_panic,
1027         .get_boot_time          = rtas_get_boot_time,
1028         .get_rtc_time           = rtas_get_rtc_time,
1029         .set_rtc_time           = rtas_set_rtc_time,
1030         .calibrate_decr         = generic_calibrate_decr,
1031         .progress               = rtas_progress,
1032         .system_reset_exception = pSeries_system_reset_exception,
1033         .machine_check_early    = pseries_machine_check_realmode,
1034         .machine_check_exception = pSeries_machine_check_exception,
1035 #ifdef CONFIG_KEXEC_CORE
1036         .machine_kexec          = pSeries_machine_kexec,
1037         .kexec_cpu_down         = pseries_kexec_cpu_down,
1038 #endif
1039 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
1040         .memory_block_size      = pseries_memory_block_size,
1041 #endif
1042 };