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/kexec.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/mpic.h>
  61 #include <asm/xics.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/security_features.h>
  71
  72 #include "pseries.h"
  73
  74 int CMO_PrPSP = -1;
  75 int CMO_SecPSP = -1;
  76 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
  77 EXPORT_SYMBOL(CMO_PageSize);
  78
  79 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
  80
  81 static struct device_node *pSeries_mpic_node;
  82
  83 static void pSeries_show_cpuinfo(struct seq_file *m)
  84 {
  85         struct device_node *root;
  86         const char *model = "";
  87
  88         root = of_find_node_by_path("/");
  89         if (root)
  90                 model = of_get_property(root, "model", NULL);
  91         seq_printf(m, "machine\t\t: CHRP %s\n", model);
  92         of_node_put(root);
  93 }
  94
  95 /* Initialize firmware assisted non-maskable interrupts if
  96  * the firmware supports this feature.
  97  */
  98 static void __init fwnmi_init(void)
  99 {
 100         unsigned long system_reset_addr, machine_check_addr;
 101
 102         int ibm_nmi_register = rtas_token("ibm,nmi-register");
 103         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
 104                 return;
 105
 106         /* If the kernel's not linked at zero we point the firmware at low
 107          * addresses anyway, and use a trampoline to get to the real code. */
 108         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
 109         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
 110
 111         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
 112                                 machine_check_addr))
 113                 fwnmi_active = 1;
 114 }
 115
 116 static void pseries_8259_cascade(struct irq_desc *desc)
 117 {
 118         struct irq_chip *chip = irq_desc_get_chip(desc);
 119         unsigned int cascade_irq = i8259_irq();
 120
 121         if (cascade_irq != NO_IRQ)
 122                 generic_handle_irq(cascade_irq);
 123
 124         chip->irq_eoi(&desc->irq_data);
 125 }
 126
 127 static void __init pseries_setup_i8259_cascade(void)
 128 {
 129         struct device_node *np, *old, *found = NULL;
 130         unsigned int cascade;
 131         const u32 *addrp;
 132         unsigned long intack = 0;
 133         int naddr;
 134
 135         for_each_node_by_type(np, "interrupt-controller") {
 136                 if (of_device_is_compatible(np, "chrp,iic")) {
 137                         found = np;
 138                         break;
 139                 }
 140         }
 141
 142         if (found == NULL) {
 143                 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
 144                 return;
 145         }
 146
 147         cascade = irq_of_parse_and_map(found, 0);
 148         if (cascade == NO_IRQ) {
 149                 printk(KERN_ERR "pic: failed to map cascade interrupt");
 150                 return;
 151         }
 152         pr_debug("pic: cascade mapped to irq %d\n", cascade);
 153
 154         for (old = of_node_get(found); old != NULL ; old = np) {
 155                 np = of_get_parent(old);
 156                 of_node_put(old);
 157                 if (np == NULL)
 158                         break;
 159                 if (strcmp(np->name, "pci") != 0)
 160                         continue;
 161                 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
 162                 if (addrp == NULL)
 163                         continue;
 164                 naddr = of_n_addr_cells(np);
 165                 intack = addrp[naddr-1];
 166                 if (naddr > 1)
 167                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
 168         }
 169         if (intack)
 170                 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
 171         i8259_init(found, intack);
 172         of_node_put(found);
 173         irq_set_chained_handler(cascade, pseries_8259_cascade);
 174 }
 175
 176 static void __init pseries_mpic_init_IRQ(void)
 177 {
 178         struct device_node *np;
 179         const unsigned int *opprop;
 180         unsigned long openpic_addr = 0;
 181         int naddr, n, i, opplen;
 182         struct mpic *mpic;
 183
 184         np = of_find_node_by_path("/");
 185         naddr = of_n_addr_cells(np);
 186         opprop = of_get_property(np, "platform-open-pic", &opplen);
 187         if (opprop != NULL) {
 188                 openpic_addr = of_read_number(opprop, naddr);
 189                 printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
 190         }
 191         of_node_put(np);
 192
 193         BUG_ON(openpic_addr == 0);
 194
 195         /* Setup the openpic driver */
 196         mpic = mpic_alloc(pSeries_mpic_node, openpic_addr,
 197                         MPIC_NO_RESET, 16, 0, " MPIC     ");
 198         BUG_ON(mpic == NULL);
 199
 200         /* Add ISUs */
 201         opplen /= sizeof(u32);
 202         for (n = 0, i = naddr; i < opplen; i += naddr, n++) {
 203                 unsigned long isuaddr = of_read_number(opprop + i, naddr);
 204                 mpic_assign_isu(mpic, n, isuaddr);
 205         }
 206
 207         /* Setup top-level get_irq */
 208         ppc_md.get_irq = mpic_get_irq;
 209
 210         /* All ISUs are setup, complete initialization */
 211         mpic_init(mpic);
 212
 213         /* Look for cascade */
 214         pseries_setup_i8259_cascade();
 215 }
 216
 217 static void __init pseries_xics_init_IRQ(void)
 218 {
 219         xics_init();
 220         pseries_setup_i8259_cascade();
 221 }
 222
 223 static void pseries_lpar_enable_pmcs(void)
 224 {
 225         unsigned long set, reset;
 226
 227         set = 1UL << 63;
 228         reset = 0;
 229         plpar_hcall_norets(H_PERFMON, set, reset);
 230 }
 231
 232 static void __init pseries_discover_pic(void)
 233 {
 234         struct device_node *np;
 235         const char *typep;
 236
 237         for_each_node_by_name(np, "interrupt-controller") {
 238                 typep = of_get_property(np, "compatible", NULL);
 239                 if (strstr(typep, "open-pic")) {
 240                         pSeries_mpic_node = of_node_get(np);
 241                         ppc_md.init_IRQ       = pseries_mpic_init_IRQ;
 242                         setup_kexec_cpu_down_mpic();
 243                         smp_init_pseries_mpic();
 244                         return;
 245                 } else if (strstr(typep, "ppc-xicp")) {
 246                         ppc_md.init_IRQ       = pseries_xics_init_IRQ;
 247                         setup_kexec_cpu_down_xics();
 248                         smp_init_pseries_xics();
 249                         return;
 250                 }
 251         }
 252         printk(KERN_ERR "pSeries_discover_pic: failed to recognize"
 253                " interrupt-controller\n");
 254 }
 255
 256 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
 257 {
 258         struct of_reconfig_data *rd = data;
 259         struct device_node *parent, *np = rd->dn;
 260         struct pci_dn *pdn;
 261         int err = NOTIFY_OK;
 262
 263         switch (action) {
 264         case OF_RECONFIG_ATTACH_NODE:
 265                 parent = of_get_parent(np);
 266                 pdn = parent ? PCI_DN(parent) : NULL;
 267                 if (pdn) {
 268                         /* Create pdn and EEH device */
 269                         update_dn_pci_info(np, pdn->phb);
 270                         eeh_dev_init(PCI_DN(np), pdn->phb);
 271                 }
 272
 273                 of_node_put(parent);
 274                 break;
 275         case OF_RECONFIG_DETACH_NODE:
 276                 pdn = PCI_DN(np);
 277                 if (pdn)
 278                         list_del(&pdn->list);
 279                 break;
 280         default:
 281                 err = NOTIFY_DONE;
 282                 break;
 283         }
 284         return err;
 285 }
 286
 287 static struct notifier_block pci_dn_reconfig_nb = {
 288         .notifier_call = pci_dn_reconfig_notifier,
 289 };
 290
 291 struct kmem_cache *dtl_cache;
 292
 293 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 294 /*
 295  * Allocate space for the dispatch trace log for all possible cpus
 296  * and register the buffers with the hypervisor.  This is used for
 297  * computing time stolen by the hypervisor.
 298  */
 299 static int alloc_dispatch_logs(void)
 300 {
 301         int cpu, ret;
 302         struct paca_struct *pp;
 303         struct dtl_entry *dtl;
 304
 305         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 306                 return 0;
 307
 308         if (!dtl_cache)
 309                 return 0;
 310
 311         for_each_possible_cpu(cpu) {
 312                 pp = &paca[cpu];
 313                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
 314                 if (!dtl) {
 315                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
 316                                 cpu);
 317                         pr_warn("Stolen time statistics will be unreliable\n");
 318                         break;
 319                 }
 320
 321                 pp->dtl_ridx = 0;
 322                 pp->dispatch_log = dtl;
 323                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
 324                 pp->dtl_curr = dtl;
 325         }
 326
 327         /* Register the DTL for the current (boot) cpu */
 328         dtl = get_paca()->dispatch_log;
 329         get_paca()->dtl_ridx = 0;
 330         get_paca()->dtl_curr = dtl;
 331         get_paca()->lppaca_ptr->dtl_idx = 0;
 332
 333         /* hypervisor reads buffer length from this field */
 334         dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 335         ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
 336         if (ret)
 337                 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
 338                        "with %d\n", smp_processor_id(),
 339                        hard_smp_processor_id(), ret);
 340         get_paca()->lppaca_ptr->dtl_enable_mask = 2;
 341
 342         return 0;
 343 }
 344 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 345 static inline int alloc_dispatch_logs(void)
 346 {
 347         return 0;
 348 }
 349 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 350
 351 static int alloc_dispatch_log_kmem_cache(void)
 352 {
 353         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
 354                                                 DISPATCH_LOG_BYTES, 0, NULL);
 355         if (!dtl_cache) {
 356                 pr_warn("Failed to create dispatch trace log buffer cache\n");
 357                 pr_warn("Stolen time statistics will be unreliable\n");
 358                 return 0;
 359         }
 360
 361         return alloc_dispatch_logs();
 362 }
 363 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
 364
 365 static void pseries_lpar_idle(void)
 366 {
 367         /*
 368          * Default handler to go into low thread priority and possibly
 369          * low power mode by cedeing processor to hypervisor
 370          */
 371
 372         if (!prep_irq_for_idle())
 373                 return;
 374
 375         /* Indicate to hypervisor that we are idle. */
 376         get_lppaca()->idle = 1;
 377
 378         /*
 379          * Yield the processor to the hypervisor.  We return if
 380          * an external interrupt occurs (which are driven prior
 381          * to returning here) or if a prod occurs from another
 382          * processor. When returning here, external interrupts
 383          * are enabled.
 384          */
 385         cede_processor();
 386
 387         get_lppaca()->idle = 0;
 388 }
 389
 390 /*
 391  * Enable relocation on during exceptions. This has partition wide scope and
 392  * may take a while to complete, if it takes longer than one second we will
 393  * just give up rather than wasting any more time on this - if that turns out
 394  * to ever be a problem in practice we can move this into a kernel thread to
 395  * finish off the process later in boot.
 396  */
 397 long pSeries_enable_reloc_on_exc(void)
 398 {
 399         long rc;
 400         unsigned int delay, total_delay = 0;
 401
 402         while (1) {
 403                 rc = enable_reloc_on_exceptions();
 404                 if (!H_IS_LONG_BUSY(rc))
 405                         return rc;
 406
 407                 delay = get_longbusy_msecs(rc);
 408                 total_delay += delay;
 409                 if (total_delay > 1000) {
 410                         pr_warn("Warning: Giving up waiting to enable "
 411                                 "relocation on exceptions (%u msec)!\n",
 412                                 total_delay);
 413                         return rc;
 414                 }
 415
 416                 mdelay(delay);
 417         }
 418 }
 419 EXPORT_SYMBOL(pSeries_enable_reloc_on_exc);
 420
 421 long pSeries_disable_reloc_on_exc(void)
 422 {
 423         long rc;
 424
 425         while (1) {
 426                 rc = disable_reloc_on_exceptions();
 427                 if (!H_IS_LONG_BUSY(rc))
 428                         return rc;
 429                 mdelay(get_longbusy_msecs(rc));
 430         }
 431 }
 432 EXPORT_SYMBOL(pSeries_disable_reloc_on_exc);
 433
 434 #ifdef CONFIG_KEXEC
 435 static void pSeries_machine_kexec(struct kimage *image)
 436 {
 437         long rc;
 438
 439         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 440                 rc = pSeries_disable_reloc_on_exc();
 441                 if (rc != H_SUCCESS)
 442                         pr_warning("Warning: Failed to disable relocation on "
 443                                    "exceptions: %ld\n", rc);
 444         }
 445
 446         default_machine_kexec(image);
 447 }
 448 #endif
 449
 450 #ifdef __LITTLE_ENDIAN__
 451 long pseries_big_endian_exceptions(void)
 452 {
 453         long rc;
 454
 455         while (1) {
 456                 rc = enable_big_endian_exceptions();
 457                 if (!H_IS_LONG_BUSY(rc))
 458                         return rc;
 459                 mdelay(get_longbusy_msecs(rc));
 460         }
 461 }
 462
 463 static long pseries_little_endian_exceptions(void)
 464 {
 465         long rc;
 466
 467         while (1) {
 468                 rc = enable_little_endian_exceptions();
 469                 if (!H_IS_LONG_BUSY(rc))
 470                         return rc;
 471                 mdelay(get_longbusy_msecs(rc));
 472         }
 473 }
 474 #endif
 475
 476 static void __init find_and_init_phbs(void)
 477 {
 478         struct device_node *node;
 479         struct pci_controller *phb;
 480         struct device_node *root = of_find_node_by_path("/");
 481
 482         for_each_child_of_node(root, node) {
 483                 if (node->type == NULL || (strcmp(node->type, "pci") != 0 &&
 484                                            strcmp(node->type, "pciex") != 0))
 485                         continue;
 486
 487                 phb = pcibios_alloc_controller(node);
 488                 if (!phb)
 489                         continue;
 490                 rtas_setup_phb(phb);
 491                 pci_process_bridge_OF_ranges(phb, node, 0);
 492                 isa_bridge_find_early(phb);
 493                 phb->controller_ops = pseries_pci_controller_ops;
 494         }
 495
 496         of_node_put(root);
 497         pci_devs_phb_init();
 498
 499         /*
 500          * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
 501          * in chosen.
 502          */
 503         of_pci_check_probe_only();
 504 }
 505
 506 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
 507 {
 508         /*
 509          * The features below are disabled by default, so we instead look to see
 510          * if firmware has *enabled* them, and set them if so.
 511          */
 512         if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
 513                 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
 514
 515         if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
 516                 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
 517
 518         if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
 519                 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
 520
 521         if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
 522                 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
 523
 524         if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
 525                 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
 526
 527         if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
 528                 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
 529
 530         if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
 531                 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
 532
 533         if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
 534                 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
 535
 536         /*
 537          * The features below are enabled by default, so we instead look to see
 538          * if firmware has *disabled* them, and clear them if so.
 539          */
 540         if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
 541                 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
 542
 543         if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
 544                 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
 545
 546         if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
 547                 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
 548 }
 549
 550 void pseries_setup_rfi_flush(void)
 551 {
 552         struct h_cpu_char_result result;
 553         enum l1d_flush_type types;
 554         bool enable;
 555         long rc;
 556
 557         /*
 558          * Set features to the defaults assumed by init_cpu_char_feature_flags()
 559          * so it can set/clear again any features that might have changed after
 560          * migration, and in case the hypercall fails and it is not even called.
 561          */
 562         powerpc_security_features = SEC_FTR_DEFAULT;
 563
 564         rc = plpar_get_cpu_characteristics(&result);
 565         if (rc == H_SUCCESS)
 566                 init_cpu_char_feature_flags(&result);
 567
 568         /*
 569          * We're the guest so this doesn't apply to us, clear it to simplify
 570          * handling of it elsewhere.
 571          */
 572         security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
 573
 574         types = L1D_FLUSH_FALLBACK;
 575
 576         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
 577                 types |= L1D_FLUSH_MTTRIG;
 578
 579         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
 580                 types |= L1D_FLUSH_ORI;
 581
 582         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
 583                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
 584
 585         setup_rfi_flush(types, enable);
 586         setup_count_cache_flush();
 587
 588         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 589                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
 590         setup_entry_flush(enable);
 591
 592         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 593                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
 594         setup_uaccess_flush(enable);
 595 }
 596
 597 static void __init pSeries_setup_arch(void)
 598 {
 599         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
 600
 601         /* Discover PIC type and setup ppc_md accordingly */
 602         pseries_discover_pic();
 603
 604         /* openpic global configuration register (64-bit format). */
 605         /* openpic Interrupt Source Unit pointer (64-bit format). */
 606         /* python0 facility area (mmio) (64-bit format) REAL address. */
 607
 608         /* init to some ~sane value until calibrate_delay() runs */
 609         loops_per_jiffy = 50000000;
 610
 611         fwnmi_init();
 612
 613         pseries_setup_rfi_flush();
 614         setup_stf_barrier();
 615
 616         /* By default, only probe PCI (can be overridden by rtas_pci) */
 617         pci_add_flags(PCI_PROBE_ONLY);
 618
 619         /* Find and initialize PCI host bridges */
 620         init_pci_config_tokens();
 621         find_and_init_phbs();
 622         of_reconfig_notifier_register(&pci_dn_reconfig_nb);
 623
 624         pSeries_nvram_init();
 625
 626         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 627                 vpa_init(boot_cpuid);
 628                 ppc_md.power_save = pseries_lpar_idle;
 629                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
 630         } else {
 631                 /* No special idle routine */
 632                 ppc_md.enable_pmcs = power4_enable_pmcs;
 633         }
 634
 635         ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
 636
 637         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 638                 long rc;
 639
 640                 rc = pSeries_enable_reloc_on_exc();
 641                 if (rc == H_P2) {
 642                         pr_info("Relocation on exceptions not supported\n");
 643                 } else if (rc != H_SUCCESS) {
 644                         pr_warn("Unable to enable relocation on exceptions: "
 645                                 "%ld\n", rc);
 646                 }
 647         }
 648 }
 649
 650 static int __init pSeries_init_panel(void)
 651 {
 652         /* Manually leave the kernel version on the panel. */
 653 #ifdef __BIG_ENDIAN__
 654         ppc_md.progress("Linux ppc64\n", 0);
 655 #else
 656         ppc_md.progress("Linux ppc64le\n", 0);
 657 #endif
 658         ppc_md.progress(init_utsname()->version, 0);
 659
 660         return 0;
 661 }
 662 machine_arch_initcall(pseries, pSeries_init_panel);
 663
 664 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
 665 {
 666         return plpar_hcall_norets(H_SET_DABR, dabr);
 667 }
 668
 669 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
 670 {
 671         /* Have to set at least one bit in the DABRX according to PAPR */
 672         if (dabrx == 0 && dabr == 0)
 673                 dabrx = DABRX_USER;
 674         /* PAPR says we can only set kernel and user bits */
 675         dabrx &= DABRX_KERNEL | DABRX_USER;
 676
 677         return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
 678 }
 679
 680 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
 681 {
 682         /* PAPR says we can't set HYP */
 683         dawrx &= ~DAWRX_HYP;
 684
 685         return  plapr_set_watchpoint0(dawr, dawrx);
 686 }
 687
 688 #define CMO_CHARACTERISTICS_TOKEN 44
 689 #define CMO_MAXLENGTH 1026
 690
 691 void pSeries_coalesce_init(void)
 692 {
 693         struct hvcall_mpp_x_data mpp_x_data;
 694
 695         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
 696                 powerpc_firmware_features |= FW_FEATURE_XCMO;
 697         else
 698                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
 699 }
 700
 701 /**
 702  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
 703  * handle that here. (Stolen from parse_system_parameter_string)
 704  */
 705 static void pSeries_cmo_feature_init(void)
 706 {
 707         char *ptr, *key, *value, *end;
 708         int call_status;
 709         int page_order = IOMMU_PAGE_SHIFT_4K;
 710
 711         pr_debug(" -> fw_cmo_feature_init()\n");
 712         spin_lock(&rtas_data_buf_lock);
 713         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
 714         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 715                                 NULL,
 716                                 CMO_CHARACTERISTICS_TOKEN,
 717                                 __pa(rtas_data_buf),
 718                                 RTAS_DATA_BUF_SIZE);
 719
 720         if (call_status != 0) {
 721                 spin_unlock(&rtas_data_buf_lock);
 722                 pr_debug("CMO not available\n");
 723                 pr_debug(" <- fw_cmo_feature_init()\n");
 724                 return;
 725         }
 726
 727         end = rtas_data_buf + CMO_MAXLENGTH - 2;
 728         ptr = rtas_data_buf + 2;        /* step over strlen value */
 729         key = value = ptr;
 730
 731         while (*ptr && (ptr <= end)) {
 732                 /* Separate the key and value by replacing '=' with '\0' and
 733                  * point the value at the string after the '='
 734                  */
 735                 if (ptr[0] == '=') {
 736                         ptr[0] = '\0';
 737                         value = ptr + 1;
 738                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
 739                         /* Terminate the string containing the key/value pair */
 740                         ptr[0] = '\0';
 741
 742                         if (key == value) {
 743                                 pr_debug("Malformed key/value pair\n");
 744                                 /* Never found a '=', end processing */
 745                                 break;
 746                         }
 747
 748                         if (0 == strcmp(key, "CMOPageSize"))
 749                                 page_order = simple_strtol(value, NULL, 10);
 750                         else if (0 == strcmp(key, "PrPSP"))
 751                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
 752                         else if (0 == strcmp(key, "SecPSP"))
 753                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
 754                         value = key = ptr + 1;
 755                 }
 756                 ptr++;
 757         }
 758
 759         /* Page size is returned as the power of 2 of the page size,
 760          * convert to the page size in bytes before returning
 761          */
 762         CMO_PageSize = 1 << page_order;
 763         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
 764
 765         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
 766                 pr_info("CMO enabled\n");
 767                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 768                          CMO_SecPSP);
 769                 powerpc_firmware_features |= FW_FEATURE_CMO;
 770                 pSeries_coalesce_init();
 771         } else
 772                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 773                          CMO_SecPSP);
 774         spin_unlock(&rtas_data_buf_lock);
 775         pr_debug(" <- fw_cmo_feature_init()\n");
 776 }
 777
 778 /*
 779  * Early initialization.  Relocation is on but do not reference unbolted pages
 780  */
 781 static void __init pSeries_init_early(void)
 782 {
 783         pr_debug(" -> pSeries_init_early()\n");
 784
 785 #ifdef CONFIG_HVC_CONSOLE
 786         if (firmware_has_feature(FW_FEATURE_LPAR))
 787                 hvc_vio_init_early();
 788 #endif
 789         if (firmware_has_feature(FW_FEATURE_XDABR))
 790                 ppc_md.set_dabr = pseries_set_xdabr;
 791         else if (firmware_has_feature(FW_FEATURE_DABR))
 792                 ppc_md.set_dabr = pseries_set_dabr;
 793
 794         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 795                 ppc_md.set_dawr = pseries_set_dawr;
 796
 797         pSeries_cmo_feature_init();
 798         iommu_init_early_pSeries();
 799
 800         pr_debug(" <- pSeries_init_early()\n");
 801 }
 802
 803 /**
 804  * pseries_power_off - tell firmware about how to power off the system.
 805  *
 806  * This function calls either the power-off rtas token in normal cases
 807  * or the ibm,power-off-ups token (if present & requested) in case of
 808  * a power failure. If power-off token is used, power on will only be
 809  * possible with power button press. If ibm,power-off-ups token is used
 810  * it will allow auto poweron after power is restored.
 811  */
 812 static void pseries_power_off(void)
 813 {
 814         int rc;
 815         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
 816
 817         if (rtas_flash_term_hook)
 818                 rtas_flash_term_hook(SYS_POWER_OFF);
 819
 820         if (rtas_poweron_auto == 0 ||
 821                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
 822                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
 823                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
 824         } else {
 825                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
 826                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
 827         }
 828         for (;;);
 829 }
 830
 831 /*
 832  * Called very early, MMU is off, device-tree isn't unflattened
 833  */
 834
 835 static int __init pseries_probe_fw_features(unsigned long node,
 836                                             const char *uname, int depth,
 837                                             void *data)
 838 {
 839         const char *prop;
 840         int len;
 841         static int hypertas_found;
 842         static int vec5_found;
 843
 844         if (depth != 1)
 845                 return 0;
 846
 847         if (!strcmp(uname, "rtas") || !strcmp(uname, "rtas@0")) {
 848                 prop = of_get_flat_dt_prop(node, "ibm,hypertas-functions",
 849                                            &len);
 850                 if (prop) {
 851                         powerpc_firmware_features |= FW_FEATURE_LPAR;
 852                         fw_hypertas_feature_init(prop, len);
 853                 }
 854
 855                 hypertas_found = 1;
 856         }
 857
 858         if (!strcmp(uname, "chosen")) {
 859                 prop = of_get_flat_dt_prop(node, "ibm,architecture-vec-5",
 860                                            &len);
 861                 if (prop)
 862                         fw_vec5_feature_init(prop, len);
 863
 864                 vec5_found = 1;
 865         }
 866
 867         return hypertas_found && vec5_found;
 868 }
 869
 870 static int __init pSeries_probe(void)
 871 {
 872         unsigned long root = of_get_flat_dt_root();
 873         const char *dtype = of_get_flat_dt_prop(root, "device_type", NULL);
 874
 875         if (dtype == NULL)
 876                 return 0;
 877         if (strcmp(dtype, "chrp"))
 878                 return 0;
 879
 880         /* Cell blades firmware claims to be chrp while it's not. Until this
 881          * is fixed, we need to avoid those here.
 882          */
 883         if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
 884             of_flat_dt_is_compatible(root, "IBM,CBEA"))
 885                 return 0;
 886
 887         pr_debug("pSeries detected, looking for LPAR capability...\n");
 888
 889         /* Now try to figure out if we are running on LPAR */
 890         of_scan_flat_dt(pseries_probe_fw_features, NULL);
 891
 892 #ifdef __LITTLE_ENDIAN__
 893         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 894                 long rc;
 895                 /*
 896                  * Tell the hypervisor that we want our exceptions to
 897                  * be taken in little endian mode. If this fails we don't
 898                  * want to use BUG() because it will trigger an exception.
 899                  */
 900                 rc = pseries_little_endian_exceptions();
 901                 if (rc) {
 902                         ppc_md.progress("H_SET_MODE LE exception fail", 0);
 903                         panic("Could not enable little endian exceptions");
 904                 }
 905         }
 906 #endif
 907
 908         if (firmware_has_feature(FW_FEATURE_LPAR))
 909                 hpte_init_lpar();
 910         else
 911                 hpte_init_native();
 912
 913         pm_power_off = pseries_power_off;
 914
 915         pr_debug("Machine is%s LPAR !\n",
 916                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
 917
 918         return 1;
 919 }
 920
 921 static int pSeries_pci_probe_mode(struct pci_bus *bus)
 922 {
 923         if (firmware_has_feature(FW_FEATURE_LPAR))
 924                 return PCI_PROBE_DEVTREE;
 925         return PCI_PROBE_NORMAL;
 926 }
 927
 928 struct pci_controller_ops pseries_pci_controller_ops = {
 929         .probe_mode             = pSeries_pci_probe_mode,
 930 };
 931
 932 define_machine(pseries) {
 933         .name                   = "pSeries",
 934         .probe                  = pSeries_probe,
 935         .setup_arch             = pSeries_setup_arch,
 936         .init_early             = pSeries_init_early,
 937         .show_cpuinfo           = pSeries_show_cpuinfo,
 938         .log_error              = pSeries_log_error,
 939         .pcibios_fixup          = pSeries_final_fixup,
 940         .restart                = rtas_restart,
 941         .halt                   = rtas_halt,
 942         .panic                  = rtas_os_term,
 943         .get_boot_time          = rtas_get_boot_time,
 944         .get_rtc_time           = rtas_get_rtc_time,
 945         .set_rtc_time           = rtas_set_rtc_time,
 946         .calibrate_decr         = generic_calibrate_decr,
 947         .progress               = rtas_progress,
 948         .system_reset_exception = pSeries_system_reset_exception,
 949         .machine_check_exception = pSeries_machine_check_exception,
 950 #ifdef CONFIG_KEXEC
 951         .machine_kexec          = pSeries_machine_kexec,
 952 #endif
 953 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 954         .memory_block_size      = pseries_memory_block_size,
 955 #endif
 956 };