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