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