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
  45 #include <asm/mmu.h>
  46 #include <asm/processor.h>
  47 #include <asm/io.h>
  48 #include <asm/pgtable.h>
  49 #include <asm/prom.h>
  50 #include <asm/rtas.h>
  51 #include <asm/pci-bridge.h>
  52 #include <asm/iommu.h>
  53 #include <asm/dma.h>
  54 #include <asm/machdep.h>
  55 #include <asm/irq.h>
  56 #include <asm/time.h>
  57 #include <asm/nvram.h>
  58 #include <asm/pmc.h>
  59 #include <asm/xics.h>
  60 #include <asm/xive.h>
  61 #include <asm/ppc-pci.h>
  62 #include <asm/i8259.h>
  63 #include <asm/udbg.h>
  64 #include <asm/smp.h>
  65 #include <asm/firmware.h>
  66 #include <asm/eeh.h>
  67 #include <asm/reg.h>
  68 #include <asm/plpar_wrappers.h>
  69 #include <asm/kexec.h>
  70 #include <asm/isa-bridge.h>
  71 #include <asm/security_features.h>
  72
  73 #include "pseries.h"
  74
  75 int CMO_PrPSP = -1;
  76 int CMO_SecPSP = -1;
  77 unsigned long CMO_PageSize = (ASM_CONST(1) << IOMMU_PAGE_SHIFT_4K);
  78 EXPORT_SYMBOL(CMO_PageSize);
  79
  80 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
  81
  82 static void pSeries_show_cpuinfo(struct seq_file *m)
  83 {
  84         struct device_node *root;
  85         const char *model = "";
  86
  87         root = of_find_node_by_path("/");
  88         if (root)
  89                 model = of_get_property(root, "model", NULL);
  90         seq_printf(m, "machine\t\t: CHRP %s\n", model);
  91         of_node_put(root);
  92         if (radix_enabled())
  93                 seq_printf(m, "MMU\t\t: Radix\n");
  94         else
  95                 seq_printf(m, "MMU\t\t: Hash\n");
  96 }
  97
  98 /* Initialize firmware assisted non-maskable interrupts if
  99  * the firmware supports this feature.
 100  */
 101 static void __init fwnmi_init(void)
 102 {
 103         unsigned long system_reset_addr, machine_check_addr;
 104
 105         int ibm_nmi_register = rtas_token("ibm,nmi-register");
 106         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
 107                 return;
 108
 109         /* If the kernel's not linked at zero we point the firmware at low
 110          * addresses anyway, and use a trampoline to get to the real code. */
 111         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
 112         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
 113
 114         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
 115                                 machine_check_addr))
 116                 fwnmi_active = 1;
 117 }
 118
 119 static void pseries_8259_cascade(struct irq_desc *desc)
 120 {
 121         struct irq_chip *chip = irq_desc_get_chip(desc);
 122         unsigned int cascade_irq = i8259_irq();
 123
 124         if (cascade_irq)
 125                 generic_handle_irq(cascade_irq);
 126
 127         chip->irq_eoi(&desc->irq_data);
 128 }
 129
 130 static void __init pseries_setup_i8259_cascade(void)
 131 {
 132         struct device_node *np, *old, *found = NULL;
 133         unsigned int cascade;
 134         const u32 *addrp;
 135         unsigned long intack = 0;
 136         int naddr;
 137
 138         for_each_node_by_type(np, "interrupt-controller") {
 139                 if (of_device_is_compatible(np, "chrp,iic")) {
 140                         found = np;
 141                         break;
 142                 }
 143         }
 144
 145         if (found == NULL) {
 146                 printk(KERN_DEBUG "pic: no ISA interrupt controller\n");
 147                 return;
 148         }
 149
 150         cascade = irq_of_parse_and_map(found, 0);
 151         if (!cascade) {
 152                 printk(KERN_ERR "pic: failed to map cascade interrupt");
 153                 return;
 154         }
 155         pr_debug("pic: cascade mapped to irq %d\n", cascade);
 156
 157         for (old = of_node_get(found); old != NULL ; old = np) {
 158                 np = of_get_parent(old);
 159                 of_node_put(old);
 160                 if (np == NULL)
 161                         break;
 162                 if (strcmp(np->name, "pci") != 0)
 163                         continue;
 164                 addrp = of_get_property(np, "8259-interrupt-acknowledge", NULL);
 165                 if (addrp == NULL)
 166                         continue;
 167                 naddr = of_n_addr_cells(np);
 168                 intack = addrp[naddr-1];
 169                 if (naddr > 1)
 170                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
 171         }
 172         if (intack)
 173                 printk(KERN_DEBUG "pic: PCI 8259 intack at 0x%016lx\n", intack);
 174         i8259_init(found, intack);
 175         of_node_put(found);
 176         irq_set_chained_handler(cascade, pseries_8259_cascade);
 177 }
 178
 179 static void __init pseries_init_irq(void)
 180 {
 181         /* Try using a XIVE if available, otherwise use a XICS */
 182         if (!xive_spapr_init()) {
 183                 xics_init();
 184                 pseries_setup_i8259_cascade();
 185         }
 186 }
 187
 188 static void pseries_lpar_enable_pmcs(void)
 189 {
 190         unsigned long set, reset;
 191
 192         set = 1UL << 63;
 193         reset = 0;
 194         plpar_hcall_norets(H_PERFMON, set, reset);
 195 }
 196
 197 static int pci_dn_reconfig_notifier(struct notifier_block *nb, unsigned long action, void *data)
 198 {
 199         struct of_reconfig_data *rd = data;
 200         struct device_node *parent, *np = rd->dn;
 201         struct pci_dn *pdn;
 202         int err = NOTIFY_OK;
 203
 204         switch (action) {
 205         case OF_RECONFIG_ATTACH_NODE:
 206                 parent = of_get_parent(np);
 207                 pdn = parent ? PCI_DN(parent) : NULL;
 208                 if (pdn)
 209                         pci_add_device_node_info(pdn->phb, np);
 210
 211                 of_node_put(parent);
 212                 break;
 213         case OF_RECONFIG_DETACH_NODE:
 214                 pdn = PCI_DN(np);
 215                 if (pdn)
 216                         list_del(&pdn->list);
 217                 break;
 218         default:
 219                 err = NOTIFY_DONE;
 220                 break;
 221         }
 222         return err;
 223 }
 224
 225 static struct notifier_block pci_dn_reconfig_nb = {
 226         .notifier_call = pci_dn_reconfig_notifier,
 227 };
 228
 229 struct kmem_cache *dtl_cache;
 230
 231 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 232 /*
 233  * Allocate space for the dispatch trace log for all possible cpus
 234  * and register the buffers with the hypervisor.  This is used for
 235  * computing time stolen by the hypervisor.
 236  */
 237 static int alloc_dispatch_logs(void)
 238 {
 239         int cpu, ret;
 240         struct paca_struct *pp;
 241         struct dtl_entry *dtl;
 242
 243         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 244                 return 0;
 245
 246         if (!dtl_cache)
 247                 return 0;
 248
 249         for_each_possible_cpu(cpu) {
 250                 pp = &paca[cpu];
 251                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
 252                 if (!dtl) {
 253                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
 254                                 cpu);
 255                         pr_warn("Stolen time statistics will be unreliable\n");
 256                         break;
 257                 }
 258
 259                 pp->dtl_ridx = 0;
 260                 pp->dispatch_log = dtl;
 261                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
 262                 pp->dtl_curr = dtl;
 263         }
 264
 265         /* Register the DTL for the current (boot) cpu */
 266         dtl = get_paca()->dispatch_log;
 267         get_paca()->dtl_ridx = 0;
 268         get_paca()->dtl_curr = dtl;
 269         get_paca()->lppaca_ptr->dtl_idx = 0;
 270
 271         /* hypervisor reads buffer length from this field */
 272         dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 273         ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
 274         if (ret)
 275                 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
 276                        "with %d\n", smp_processor_id(),
 277                        hard_smp_processor_id(), ret);
 278         get_paca()->lppaca_ptr->dtl_enable_mask = 2;
 279
 280         return 0;
 281 }
 282 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 283 static inline int alloc_dispatch_logs(void)
 284 {
 285         return 0;
 286 }
 287 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 288
 289 static int alloc_dispatch_log_kmem_cache(void)
 290 {
 291         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
 292                                                 DISPATCH_LOG_BYTES, 0, NULL);
 293         if (!dtl_cache) {
 294                 pr_warn("Failed to create dispatch trace log buffer cache\n");
 295                 pr_warn("Stolen time statistics will be unreliable\n");
 296                 return 0;
 297         }
 298
 299         return alloc_dispatch_logs();
 300 }
 301 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
 302
 303 static void pseries_lpar_idle(void)
 304 {
 305         /*
 306          * Default handler to go into low thread priority and possibly
 307          * low power mode by ceding processor to hypervisor
 308          */
 309
 310         if (!prep_irq_for_idle())
 311                 return;
 312
 313         /* Indicate to hypervisor that we are idle. */
 314         get_lppaca()->idle = 1;
 315
 316         /*
 317          * Yield the processor to the hypervisor.  We return if
 318          * an external interrupt occurs (which are driven prior
 319          * to returning here) or if a prod occurs from another
 320          * processor. When returning here, external interrupts
 321          * are enabled.
 322          */
 323         cede_processor();
 324
 325         get_lppaca()->idle = 0;
 326 }
 327
 328 /*
 329  * Enable relocation on during exceptions. This has partition wide scope and
 330  * may take a while to complete, if it takes longer than one second we will
 331  * just give up rather than wasting any more time on this - if that turns out
 332  * to ever be a problem in practice we can move this into a kernel thread to
 333  * finish off the process later in boot.
 334  */
 335 void pseries_enable_reloc_on_exc(void)
 336 {
 337         long rc;
 338         unsigned int delay, total_delay = 0;
 339
 340         while (1) {
 341                 rc = enable_reloc_on_exceptions();
 342                 if (!H_IS_LONG_BUSY(rc)) {
 343                         if (rc == H_P2) {
 344                                 pr_info("Relocation on exceptions not"
 345                                         " supported\n");
 346                         } else if (rc != H_SUCCESS) {
 347                                 pr_warn("Unable to enable relocation"
 348                                         " on exceptions: %ld\n", rc);
 349                         }
 350                         break;
 351                 }
 352
 353                 delay = get_longbusy_msecs(rc);
 354                 total_delay += delay;
 355                 if (total_delay > 1000) {
 356                         pr_warn("Warning: Giving up waiting to enable "
 357                                 "relocation on exceptions (%u msec)!\n",
 358                                 total_delay);
 359                         return;
 360                 }
 361
 362                 mdelay(delay);
 363         }
 364 }
 365 EXPORT_SYMBOL(pseries_enable_reloc_on_exc);
 366
 367 void pseries_disable_reloc_on_exc(void)
 368 {
 369         long rc;
 370
 371         while (1) {
 372                 rc = disable_reloc_on_exceptions();
 373                 if (!H_IS_LONG_BUSY(rc))
 374                         break;
 375                 mdelay(get_longbusy_msecs(rc));
 376         }
 377         if (rc != H_SUCCESS)
 378                 pr_warning("Warning: Failed to disable relocation on "
 379                            "exceptions: %ld\n", rc);
 380 }
 381 EXPORT_SYMBOL(pseries_disable_reloc_on_exc);
 382
 383 #ifdef CONFIG_KEXEC_CORE
 384 static void pSeries_machine_kexec(struct kimage *image)
 385 {
 386         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 387                 pseries_disable_reloc_on_exc();
 388
 389         default_machine_kexec(image);
 390 }
 391 #endif
 392
 393 #ifdef __LITTLE_ENDIAN__
 394 void pseries_big_endian_exceptions(void)
 395 {
 396         long rc;
 397
 398         while (1) {
 399                 rc = enable_big_endian_exceptions();
 400                 if (!H_IS_LONG_BUSY(rc))
 401                         break;
 402                 mdelay(get_longbusy_msecs(rc));
 403         }
 404
 405         /*
 406          * At this point it is unlikely panic() will get anything
 407          * out to the user, since this is called very late in kexec
 408          * but at least this will stop us from continuing on further
 409          * and creating an even more difficult to debug situation.
 410          *
 411          * There is a known problem when kdump'ing, if cpus are offline
 412          * the above call will fail. Rather than panicking again, keep
 413          * going and hope the kdump kernel is also little endian, which
 414          * it usually is.
 415          */
 416         if (rc && !kdump_in_progress())
 417                 panic("Could not enable big endian exceptions");
 418 }
 419
 420 void pseries_little_endian_exceptions(void)
 421 {
 422         long rc;
 423
 424         while (1) {
 425                 rc = enable_little_endian_exceptions();
 426                 if (!H_IS_LONG_BUSY(rc))
 427                         break;
 428                 mdelay(get_longbusy_msecs(rc));
 429         }
 430         if (rc) {
 431                 ppc_md.progress("H_SET_MODE LE exception fail", 0);
 432                 panic("Could not enable little endian exceptions");
 433         }
 434 }
 435 #endif
 436
 437 static void __init find_and_init_phbs(void)
 438 {
 439         struct device_node *node;
 440         struct pci_controller *phb;
 441         struct device_node *root = of_find_node_by_path("/");
 442
 443         for_each_child_of_node(root, node) {
 444                 if (node->type == NULL || (strcmp(node->type, "pci") != 0 &&
 445                                            strcmp(node->type, "pciex") != 0))
 446                         continue;
 447
 448                 phb = pcibios_alloc_controller(node);
 449                 if (!phb)
 450                         continue;
 451                 rtas_setup_phb(phb);
 452                 pci_process_bridge_OF_ranges(phb, node, 0);
 453                 isa_bridge_find_early(phb);
 454                 phb->controller_ops = pseries_pci_controller_ops;
 455         }
 456
 457         of_node_put(root);
 458
 459         /*
 460          * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
 461          * in chosen.
 462          */
 463         of_pci_check_probe_only();
 464 }
 465
 466 static void init_cpu_char_feature_flags(struct h_cpu_char_result *result)
 467 {
 468         /*
 469          * The features below are disabled by default, so we instead look to see
 470          * if firmware has *enabled* them, and set them if so.
 471          */
 472         if (result->character & H_CPU_CHAR_SPEC_BAR_ORI31)
 473                 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
 474
 475         if (result->character & H_CPU_CHAR_BCCTRL_SERIALISED)
 476                 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
 477
 478         if (result->character & H_CPU_CHAR_L1D_FLUSH_ORI30)
 479                 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
 480
 481         if (result->character & H_CPU_CHAR_L1D_FLUSH_TRIG2)
 482                 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
 483
 484         if (result->character & H_CPU_CHAR_L1D_THREAD_PRIV)
 485                 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
 486
 487         if (result->character & H_CPU_CHAR_COUNT_CACHE_DISABLED)
 488                 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
 489
 490         if (result->character & H_CPU_CHAR_BCCTR_FLUSH_ASSIST)
 491                 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
 492
 493         if (result->behaviour & H_CPU_BEHAV_FLUSH_COUNT_CACHE)
 494                 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
 495
 496         /*
 497          * The features below are enabled by default, so we instead look to see
 498          * if firmware has *disabled* them, and clear them if so.
 499          */
 500         if (!(result->behaviour & H_CPU_BEHAV_FAVOUR_SECURITY))
 501                 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
 502
 503         if (!(result->behaviour & H_CPU_BEHAV_L1D_FLUSH_PR))
 504                 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
 505
 506         if (!(result->behaviour & H_CPU_BEHAV_BNDS_CHK_SPEC_BAR))
 507                 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
 508 }
 509
 510 void pseries_setup_rfi_flush(void)
 511 {
 512         struct h_cpu_char_result result;
 513         enum l1d_flush_type types;
 514         bool enable;
 515         long rc;
 516
 517         /*
 518          * Set features to the defaults assumed by init_cpu_char_feature_flags()
 519          * so it can set/clear again any features that might have changed after
 520          * migration, and in case the hypercall fails and it is not even called.
 521          */
 522         powerpc_security_features = SEC_FTR_DEFAULT;
 523
 524         rc = plpar_get_cpu_characteristics(&result);
 525         if (rc == H_SUCCESS)
 526                 init_cpu_char_feature_flags(&result);
 527
 528         /*
 529          * We're the guest so this doesn't apply to us, clear it to simplify
 530          * handling of it elsewhere.
 531          */
 532         security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
 533
 534         types = L1D_FLUSH_FALLBACK;
 535
 536         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
 537                 types |= L1D_FLUSH_MTTRIG;
 538
 539         if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
 540                 types |= L1D_FLUSH_ORI;
 541
 542         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
 543                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR);
 544
 545         setup_rfi_flush(types, enable);
 546         setup_count_cache_flush();
 547
 548         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 549                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
 550         setup_entry_flush(enable);
 551
 552         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 553                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
 554         setup_uaccess_flush(enable);
 555 }
 556
 557 static void __init pSeries_setup_arch(void)
 558 {
 559         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
 560
 561         /* Discover PIC type and setup ppc_md accordingly */
 562         smp_init_pseries();
 563
 564
 565         /* openpic global configuration register (64-bit format). */
 566         /* openpic Interrupt Source Unit pointer (64-bit format). */
 567         /* python0 facility area (mmio) (64-bit format) REAL address. */
 568
 569         /* init to some ~sane value until calibrate_delay() runs */
 570         loops_per_jiffy = 50000000;
 571
 572         fwnmi_init();
 573
 574         pseries_setup_rfi_flush();
 575         setup_stf_barrier();
 576
 577         /* By default, only probe PCI (can be overridden by rtas_pci) */
 578         pci_add_flags(PCI_PROBE_ONLY);
 579
 580         /* Find and initialize PCI host bridges */
 581         init_pci_config_tokens();
 582         find_and_init_phbs();
 583         of_reconfig_notifier_register(&pci_dn_reconfig_nb);
 584
 585         pSeries_nvram_init();
 586
 587         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 588                 vpa_init(boot_cpuid);
 589                 ppc_md.power_save = pseries_lpar_idle;
 590                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
 591         } else {
 592                 /* No special idle routine */
 593                 ppc_md.enable_pmcs = power4_enable_pmcs;
 594         }
 595
 596         ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
 597 }
 598
 599 static int __init pSeries_init_panel(void)
 600 {
 601         /* Manually leave the kernel version on the panel. */
 602 #ifdef __BIG_ENDIAN__
 603         ppc_md.progress("Linux ppc64\n", 0);
 604 #else
 605         ppc_md.progress("Linux ppc64le\n", 0);
 606 #endif
 607         ppc_md.progress(init_utsname()->version, 0);
 608
 609         return 0;
 610 }
 611 machine_arch_initcall(pseries, pSeries_init_panel);
 612
 613 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
 614 {
 615         return plpar_hcall_norets(H_SET_DABR, dabr);
 616 }
 617
 618 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
 619 {
 620         /* Have to set at least one bit in the DABRX according to PAPR */
 621         if (dabrx == 0 && dabr == 0)
 622                 dabrx = DABRX_USER;
 623         /* PAPR says we can only set kernel and user bits */
 624         dabrx &= DABRX_KERNEL | DABRX_USER;
 625
 626         return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
 627 }
 628
 629 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
 630 {
 631         /* PAPR says we can't set HYP */
 632         dawrx &= ~DAWRX_HYP;
 633
 634         return  plapr_set_watchpoint0(dawr, dawrx);
 635 }
 636
 637 #define CMO_CHARACTERISTICS_TOKEN 44
 638 #define CMO_MAXLENGTH 1026
 639
 640 void pSeries_coalesce_init(void)
 641 {
 642         struct hvcall_mpp_x_data mpp_x_data;
 643
 644         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
 645                 powerpc_firmware_features |= FW_FEATURE_XCMO;
 646         else
 647                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
 648 }
 649
 650 /**
 651  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
 652  * handle that here. (Stolen from parse_system_parameter_string)
 653  */
 654 static void pSeries_cmo_feature_init(void)
 655 {
 656         char *ptr, *key, *value, *end;
 657         int call_status;
 658         int page_order = IOMMU_PAGE_SHIFT_4K;
 659
 660         pr_debug(" -> fw_cmo_feature_init()\n");
 661         spin_lock(&rtas_data_buf_lock);
 662         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
 663         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 664                                 NULL,
 665                                 CMO_CHARACTERISTICS_TOKEN,
 666                                 __pa(rtas_data_buf),
 667                                 RTAS_DATA_BUF_SIZE);
 668
 669         if (call_status != 0) {
 670                 spin_unlock(&rtas_data_buf_lock);
 671                 pr_debug("CMO not available\n");
 672                 pr_debug(" <- fw_cmo_feature_init()\n");
 673                 return;
 674         }
 675
 676         end = rtas_data_buf + CMO_MAXLENGTH - 2;
 677         ptr = rtas_data_buf + 2;        /* step over strlen value */
 678         key = value = ptr;
 679
 680         while (*ptr && (ptr <= end)) {
 681                 /* Separate the key and value by replacing '=' with '\0' and
 682                  * point the value at the string after the '='
 683                  */
 684                 if (ptr[0] == '=') {
 685                         ptr[0] = '\0';
 686                         value = ptr + 1;
 687                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
 688                         /* Terminate the string containing the key/value pair */
 689                         ptr[0] = '\0';
 690
 691                         if (key == value) {
 692                                 pr_debug("Malformed key/value pair\n");
 693                                 /* Never found a '=', end processing */
 694                                 break;
 695                         }
 696
 697                         if (0 == strcmp(key, "CMOPageSize"))
 698                                 page_order = simple_strtol(value, NULL, 10);
 699                         else if (0 == strcmp(key, "PrPSP"))
 700                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
 701                         else if (0 == strcmp(key, "SecPSP"))
 702                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
 703                         value = key = ptr + 1;
 704                 }
 705                 ptr++;
 706         }
 707
 708         /* Page size is returned as the power of 2 of the page size,
 709          * convert to the page size in bytes before returning
 710          */
 711         CMO_PageSize = 1 << page_order;
 712         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
 713
 714         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
 715                 pr_info("CMO enabled\n");
 716                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 717                          CMO_SecPSP);
 718                 powerpc_firmware_features |= FW_FEATURE_CMO;
 719                 pSeries_coalesce_init();
 720         } else
 721                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 722                          CMO_SecPSP);
 723         spin_unlock(&rtas_data_buf_lock);
 724         pr_debug(" <- fw_cmo_feature_init()\n");
 725 }
 726
 727 /*
 728  * Early initialization.  Relocation is on but do not reference unbolted pages
 729  */
 730 static void __init pseries_init(void)
 731 {
 732         pr_debug(" -> pseries_init()\n");
 733
 734 #ifdef CONFIG_HVC_CONSOLE
 735         if (firmware_has_feature(FW_FEATURE_LPAR))
 736                 hvc_vio_init_early();
 737 #endif
 738         if (firmware_has_feature(FW_FEATURE_XDABR))
 739                 ppc_md.set_dabr = pseries_set_xdabr;
 740         else if (firmware_has_feature(FW_FEATURE_DABR))
 741                 ppc_md.set_dabr = pseries_set_dabr;
 742
 743         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 744                 ppc_md.set_dawr = pseries_set_dawr;
 745
 746         pSeries_cmo_feature_init();
 747         iommu_init_early_pSeries();
 748
 749         pr_debug(" <- pseries_init()\n");
 750 }
 751
 752 /**
 753  * pseries_power_off - tell firmware about how to power off the system.
 754  *
 755  * This function calls either the power-off rtas token in normal cases
 756  * or the ibm,power-off-ups token (if present & requested) in case of
 757  * a power failure. If power-off token is used, power on will only be
 758  * possible with power button press. If ibm,power-off-ups token is used
 759  * it will allow auto poweron after power is restored.
 760  */
 761 static void pseries_power_off(void)
 762 {
 763         int rc;
 764         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
 765
 766         if (rtas_flash_term_hook)
 767                 rtas_flash_term_hook(SYS_POWER_OFF);
 768
 769         if (rtas_poweron_auto == 0 ||
 770                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
 771                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
 772                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
 773         } else {
 774                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
 775                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
 776         }
 777         for (;;);
 778 }
 779
 780 static int __init pSeries_probe(void)
 781 {
 782         const char *dtype = of_get_property(of_root, "device_type", NULL);
 783
 784         if (dtype == NULL)
 785                 return 0;
 786         if (strcmp(dtype, "chrp"))
 787                 return 0;
 788
 789         /* Cell blades firmware claims to be chrp while it's not. Until this
 790          * is fixed, we need to avoid those here.
 791          */
 792         if (of_machine_is_compatible("IBM,CPBW-1.0") ||
 793             of_machine_is_compatible("IBM,CBEA"))
 794                 return 0;
 795
 796         pm_power_off = pseries_power_off;
 797
 798         pr_debug("Machine is%s LPAR !\n",
 799                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
 800
 801         pseries_init();
 802
 803         return 1;
 804 }
 805
 806 static int pSeries_pci_probe_mode(struct pci_bus *bus)
 807 {
 808         if (firmware_has_feature(FW_FEATURE_LPAR))
 809                 return PCI_PROBE_DEVTREE;
 810         return PCI_PROBE_NORMAL;
 811 }
 812
 813 struct pci_controller_ops pseries_pci_controller_ops = {
 814         .probe_mode             = pSeries_pci_probe_mode,
 815 };
 816
 817 define_machine(pseries) {
 818         .name                   = "pSeries",
 819         .probe                  = pSeries_probe,
 820         .setup_arch             = pSeries_setup_arch,
 821         .init_IRQ               = pseries_init_irq,
 822         .show_cpuinfo           = pSeries_show_cpuinfo,
 823         .log_error              = pSeries_log_error,
 824         .pcibios_fixup          = pSeries_final_fixup,
 825         .restart                = rtas_restart,
 826         .halt                   = rtas_halt,
 827         .panic                  = rtas_os_term,
 828         .get_boot_time          = rtas_get_boot_time,
 829         .get_rtc_time           = rtas_get_rtc_time,
 830         .set_rtc_time           = rtas_set_rtc_time,
 831         .calibrate_decr         = generic_calibrate_decr,
 832         .progress               = rtas_progress,
 833         .system_reset_exception = pSeries_system_reset_exception,
 834         .machine_check_exception = pSeries_machine_check_exception,
 835 #ifdef CONFIG_KEXEC_CORE
 836         .machine_kexec          = pSeries_machine_kexec,
 837         .kexec_cpu_down         = pseries_kexec_cpu_down,
 838 #endif
 839 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 840         .memory_block_size      = pseries_memory_block_size,
 841 #endif
 842 };