kernel/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 *np = rd->dn;
 258         struct pci_dn *pci = NULL;
 259         int err = NOTIFY_OK;
 260
 261         switch (action) {
 262         case OF_RECONFIG_ATTACH_NODE:
 263                 pci = np->parent->data;
 264                 if (pci) {
 265                         update_dn_pci_info(np, pci->phb);
 266
 267                         /* Create EEH device for the OF node */
 268                         eeh_dev_init(PCI_DN(np), pci->phb);
 269                 }
 270                 break;
 271         case OF_RECONFIG_DETACH_NODE:
 272                 pci = PCI_DN(np);
 273                 if (pci)
 274                         list_del(&pci->list);
 275                 break;
 276         default:
 277                 err = NOTIFY_DONE;
 278                 break;
 279         }
 280         return err;
 281 }
 282
 283 static struct notifier_block pci_dn_reconfig_nb = {
 284         .notifier_call = pci_dn_reconfig_notifier,
 285 };
 286
 287 struct kmem_cache *dtl_cache;
 288
 289 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 290 /*
 291  * Allocate space for the dispatch trace log for all possible cpus
 292  * and register the buffers with the hypervisor.  This is used for
 293  * computing time stolen by the hypervisor.
 294  */
 295 static int alloc_dispatch_logs(void)
 296 {
 297         int cpu, ret;
 298         struct paca_struct *pp;
 299         struct dtl_entry *dtl;
 300
 301         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 302                 return 0;
 303
 304         if (!dtl_cache)
 305                 return 0;
 306
 307         for_each_possible_cpu(cpu) {
 308                 pp = &paca[cpu];
 309                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
 310                 if (!dtl) {
 311                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
 312                                 cpu);
 313                         pr_warn("Stolen time statistics will be unreliable\n");
 314                         break;
 315                 }
 316
 317                 pp->dtl_ridx = 0;
 318                 pp->dispatch_log = dtl;
 319                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
 320                 pp->dtl_curr = dtl;
 321         }
 322
 323         /* Register the DTL for the current (boot) cpu */
 324         dtl = get_paca()->dispatch_log;
 325         get_paca()->dtl_ridx = 0;
 326         get_paca()->dtl_curr = dtl;
 327         get_paca()->lppaca_ptr->dtl_idx = 0;
 328
 329         /* hypervisor reads buffer length from this field */
 330         dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 331         ret = register_dtl(hard_smp_processor_id(), __pa(dtl));
 332         if (ret)
 333                 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed "
 334                        "with %d\n", smp_processor_id(),
 335                        hard_smp_processor_id(), ret);
 336         get_paca()->lppaca_ptr->dtl_enable_mask = 2;
 337
 338         return 0;
 339 }
 340 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 341 static inline int alloc_dispatch_logs(void)
 342 {
 343         return 0;
 344 }
 345 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 346
 347 static int alloc_dispatch_log_kmem_cache(void)
 348 {
 349         dtl_cache = kmem_cache_create("dtl", DISPATCH_LOG_BYTES,
 350                                                 DISPATCH_LOG_BYTES, 0, NULL);
 351         if (!dtl_cache) {
 352                 pr_warn("Failed to create dispatch trace log buffer cache\n");
 353                 pr_warn("Stolen time statistics will be unreliable\n");
 354                 return 0;
 355         }
 356
 357         return alloc_dispatch_logs();
 358 }
 359 machine_early_initcall(pseries, alloc_dispatch_log_kmem_cache);
 360
 361 static void pseries_lpar_idle(void)
 362 {
 363         /*
 364          * Default handler to go into low thread priority and possibly
 365          * low power mode by cedeing processor to hypervisor
 366          */
 367
 368         /* Indicate to hypervisor that we are idle. */
 369         get_lppaca()->idle = 1;
 370
 371         /*
 372          * Yield the processor to the hypervisor.  We return if
 373          * an external interrupt occurs (which are driven prior
 374          * to returning here) or if a prod occurs from another
 375          * processor. When returning here, external interrupts
 376          * are enabled.
 377          */
 378         cede_processor();
 379
 380         get_lppaca()->idle = 0;
 381 }
 382
 383 /*
 384  * Enable relocation on during exceptions. This has partition wide scope and
 385  * may take a while to complete, if it takes longer than one second we will
 386  * just give up rather than wasting any more time on this - if that turns out
 387  * to ever be a problem in practice we can move this into a kernel thread to
 388  * finish off the process later in boot.
 389  */
 390 long pSeries_enable_reloc_on_exc(void)
 391 {
 392         long rc;
 393         unsigned int delay, total_delay = 0;
 394
 395         while (1) {
 396                 rc = enable_reloc_on_exceptions();
 397                 if (!H_IS_LONG_BUSY(rc))
 398                         return rc;
 399
 400                 delay = get_longbusy_msecs(rc);
 401                 total_delay += delay;
 402                 if (total_delay > 1000) {
 403                         pr_warn("Warning: Giving up waiting to enable "
 404                                 "relocation on exceptions (%u msec)!\n",
 405                                 total_delay);
 406                         return rc;
 407                 }
 408
 409                 mdelay(delay);
 410         }
 411 }
 412 EXPORT_SYMBOL(pSeries_enable_reloc_on_exc);
 413
 414 long pSeries_disable_reloc_on_exc(void)
 415 {
 416         long rc;
 417
 418         while (1) {
 419                 rc = disable_reloc_on_exceptions();
 420                 if (!H_IS_LONG_BUSY(rc))
 421                         return rc;
 422                 mdelay(get_longbusy_msecs(rc));
 423         }
 424 }
 425 EXPORT_SYMBOL(pSeries_disable_reloc_on_exc);
 426
 427 #ifdef CONFIG_KEXEC
 428 static void pSeries_machine_kexec(struct kimage *image)
 429 {
 430         long rc;
 431
 432         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 433                 rc = pSeries_disable_reloc_on_exc();
 434                 if (rc != H_SUCCESS)
 435                         pr_warning("Warning: Failed to disable relocation on "
 436                                    "exceptions: %ld\n", rc);
 437         }
 438
 439         default_machine_kexec(image);
 440 }
 441 #endif
 442
 443 #ifdef __LITTLE_ENDIAN__
 444 long pseries_big_endian_exceptions(void)
 445 {
 446         long rc;
 447
 448         while (1) {
 449                 rc = enable_big_endian_exceptions();
 450                 if (!H_IS_LONG_BUSY(rc))
 451                         return rc;
 452                 mdelay(get_longbusy_msecs(rc));
 453         }
 454 }
 455
 456 static long pseries_little_endian_exceptions(void)
 457 {
 458         long rc;
 459
 460         while (1) {
 461                 rc = enable_little_endian_exceptions();
 462                 if (!H_IS_LONG_BUSY(rc))
 463                         return rc;
 464                 mdelay(get_longbusy_msecs(rc));
 465         }
 466 }
 467 #endif
 468
 469 static void __init find_and_init_phbs(void)
 470 {
 471         struct device_node *node;
 472         struct pci_controller *phb;
 473         struct device_node *root = of_find_node_by_path("/");
 474
 475         for_each_child_of_node(root, node) {
 476                 if (node->type == NULL || (strcmp(node->type, "pci") != 0 &&
 477                                            strcmp(node->type, "pciex") != 0))
 478                         continue;
 479
 480                 phb = pcibios_alloc_controller(node);
 481                 if (!phb)
 482                         continue;
 483                 rtas_setup_phb(phb);
 484                 pci_process_bridge_OF_ranges(phb, node, 0);
 485                 isa_bridge_find_early(phb);
 486                 phb->controller_ops = pseries_pci_controller_ops;
 487         }
 488
 489         of_node_put(root);
 490         pci_devs_phb_init();
 491
 492         /*
 493          * PCI_PROBE_ONLY and PCI_REASSIGN_ALL_BUS can be set via properties
 494          * in chosen.
 495          */
 496         if (of_chosen) {
 497                 const int *prop;
 498
 499                 prop = of_get_property(of_chosen,
 500                                 "linux,pci-probe-only", NULL);
 501                 if (prop) {
 502                         if (*prop)
 503                                 pci_add_flags(PCI_PROBE_ONLY);
 504                         else
 505                                 pci_clear_flags(PCI_PROBE_ONLY);
 506                 }
 507         }
 508 }
 509
 510 static void __init pSeries_setup_arch(void)
 511 {
 512         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
 513
 514         /* Discover PIC type and setup ppc_md accordingly */
 515         pseries_discover_pic();
 516
 517         /* openpic global configuration register (64-bit format). */
 518         /* openpic Interrupt Source Unit pointer (64-bit format). */
 519         /* python0 facility area (mmio) (64-bit format) REAL address. */
 520
 521         /* init to some ~sane value until calibrate_delay() runs */
 522         loops_per_jiffy = 50000000;
 523
 524         fwnmi_init();
 525
 526         /* By default, only probe PCI (can be overriden by rtas_pci) */
 527         pci_add_flags(PCI_PROBE_ONLY);
 528
 529         /* Find and initialize PCI host bridges */
 530         init_pci_config_tokens();
 531         find_and_init_phbs();
 532         of_reconfig_notifier_register(&pci_dn_reconfig_nb);
 533
 534         pSeries_nvram_init();
 535
 536         if (firmware_has_feature(FW_FEATURE_LPAR)) {
 537                 vpa_init(boot_cpuid);
 538                 ppc_md.power_save = pseries_lpar_idle;
 539                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
 540         } else {
 541                 /* No special idle routine */
 542                 ppc_md.enable_pmcs = power4_enable_pmcs;
 543         }
 544
 545         ppc_md.pcibios_root_bridge_prepare = pseries_root_bridge_prepare;
 546
 547         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 548                 long rc;
 549
 550                 rc = pSeries_enable_reloc_on_exc();
 551                 if (rc == H_P2) {
 552                         pr_info("Relocation on exceptions not supported\n");
 553                 } else if (rc != H_SUCCESS) {
 554                         pr_warn("Unable to enable relocation on exceptions: "
 555                                 "%ld\n", rc);
 556                 }
 557         }
 558 }
 559
 560 static int __init pSeries_init_panel(void)
 561 {
 562         /* Manually leave the kernel version on the panel. */
 563 #ifdef __BIG_ENDIAN__
 564         ppc_md.progress("Linux ppc64\n", 0);
 565 #else
 566         ppc_md.progress("Linux ppc64le\n", 0);
 567 #endif
 568         ppc_md.progress(init_utsname()->version, 0);
 569
 570         return 0;
 571 }
 572 machine_arch_initcall(pseries, pSeries_init_panel);
 573
 574 static int pseries_set_dabr(unsigned long dabr, unsigned long dabrx)
 575 {
 576         return plpar_hcall_norets(H_SET_DABR, dabr);
 577 }
 578
 579 static int pseries_set_xdabr(unsigned long dabr, unsigned long dabrx)
 580 {
 581         /* Have to set at least one bit in the DABRX according to PAPR */
 582         if (dabrx == 0 && dabr == 0)
 583                 dabrx = DABRX_USER;
 584         /* PAPR says we can only set kernel and user bits */
 585         dabrx &= DABRX_KERNEL | DABRX_USER;
 586
 587         return plpar_hcall_norets(H_SET_XDABR, dabr, dabrx);
 588 }
 589
 590 static int pseries_set_dawr(unsigned long dawr, unsigned long dawrx)
 591 {
 592         /* PAPR says we can't set HYP */
 593         dawrx &= ~DAWRX_HYP;
 594
 595         return  plapr_set_watchpoint0(dawr, dawrx);
 596 }
 597
 598 #define CMO_CHARACTERISTICS_TOKEN 44
 599 #define CMO_MAXLENGTH 1026
 600
 601 void pSeries_coalesce_init(void)
 602 {
 603         struct hvcall_mpp_x_data mpp_x_data;
 604
 605         if (firmware_has_feature(FW_FEATURE_CMO) && !h_get_mpp_x(&mpp_x_data))
 606                 powerpc_firmware_features |= FW_FEATURE_XCMO;
 607         else
 608                 powerpc_firmware_features &= ~FW_FEATURE_XCMO;
 609 }
 610
 611 /**
 612  * fw_cmo_feature_init - FW_FEATURE_CMO is not stored in ibm,hypertas-functions,
 613  * handle that here. (Stolen from parse_system_parameter_string)
 614  */
 615 static void pSeries_cmo_feature_init(void)
 616 {
 617         char *ptr, *key, *value, *end;
 618         int call_status;
 619         int page_order = IOMMU_PAGE_SHIFT_4K;
 620
 621         pr_debug(" -> fw_cmo_feature_init()\n");
 622         spin_lock(&rtas_data_buf_lock);
 623         memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
 624         call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
 625                                 NULL,
 626                                 CMO_CHARACTERISTICS_TOKEN,
 627                                 __pa(rtas_data_buf),
 628                                 RTAS_DATA_BUF_SIZE);
 629
 630         if (call_status != 0) {
 631                 spin_unlock(&rtas_data_buf_lock);
 632                 pr_debug("CMO not available\n");
 633                 pr_debug(" <- fw_cmo_feature_init()\n");
 634                 return;
 635         }
 636
 637         end = rtas_data_buf + CMO_MAXLENGTH - 2;
 638         ptr = rtas_data_buf + 2;        /* step over strlen value */
 639         key = value = ptr;
 640
 641         while (*ptr && (ptr <= end)) {
 642                 /* Separate the key and value by replacing '=' with '\0' and
 643                  * point the value at the string after the '='
 644                  */
 645                 if (ptr[0] == '=') {
 646                         ptr[0] = '\0';
 647                         value = ptr + 1;
 648                 } else if (ptr[0] == '\0' || ptr[0] == ',') {
 649                         /* Terminate the string containing the key/value pair */
 650                         ptr[0] = '\0';
 651
 652                         if (key == value) {
 653                                 pr_debug("Malformed key/value pair\n");
 654                                 /* Never found a '=', end processing */
 655                                 break;
 656                         }
 657
 658                         if (0 == strcmp(key, "CMOPageSize"))
 659                                 page_order = simple_strtol(value, NULL, 10);
 660                         else if (0 == strcmp(key, "PrPSP"))
 661                                 CMO_PrPSP = simple_strtol(value, NULL, 10);
 662                         else if (0 == strcmp(key, "SecPSP"))
 663                                 CMO_SecPSP = simple_strtol(value, NULL, 10);
 664                         value = key = ptr + 1;
 665                 }
 666                 ptr++;
 667         }
 668
 669         /* Page size is returned as the power of 2 of the page size,
 670          * convert to the page size in bytes before returning
 671          */
 672         CMO_PageSize = 1 << page_order;
 673         pr_debug("CMO_PageSize = %lu\n", CMO_PageSize);
 674
 675         if (CMO_PrPSP != -1 || CMO_SecPSP != -1) {
 676                 pr_info("CMO enabled\n");
 677                 pr_debug("CMO enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 678                          CMO_SecPSP);
 679                 powerpc_firmware_features |= FW_FEATURE_CMO;
 680                 pSeries_coalesce_init();
 681         } else
 682                 pr_debug("CMO not enabled, PrPSP=%d, SecPSP=%d\n", CMO_PrPSP,
 683                          CMO_SecPSP);
 684         spin_unlock(&rtas_data_buf_lock);
 685         pr_debug(" <- fw_cmo_feature_init()\n");
 686 }
 687
 688 /*
 689  * Early initialization.  Relocation is on but do not reference unbolted pages
 690  */
 691 static void __init pSeries_init_early(void)
 692 {
 693         pr_debug(" -> pSeries_init_early()\n");
 694
 695 #ifdef CONFIG_HVC_CONSOLE
 696         if (firmware_has_feature(FW_FEATURE_LPAR))
 697                 hvc_vio_init_early();
 698 #endif
 699         if (firmware_has_feature(FW_FEATURE_XDABR))
 700                 ppc_md.set_dabr = pseries_set_xdabr;
 701         else if (firmware_has_feature(FW_FEATURE_DABR))
 702                 ppc_md.set_dabr = pseries_set_dabr;
 703
 704         if (firmware_has_feature(FW_FEATURE_SET_MODE))
 705                 ppc_md.set_dawr = pseries_set_dawr;
 706
 707         pSeries_cmo_feature_init();
 708         iommu_init_early_pSeries();
 709
 710         pr_debug(" <- pSeries_init_early()\n");
 711 }
 712
 713 /**
 714  * pseries_power_off - tell firmware about how to power off the system.
 715  *
 716  * This function calls either the power-off rtas token in normal cases
 717  * or the ibm,power-off-ups token (if present & requested) in case of
 718  * a power failure. If power-off token is used, power on will only be
 719  * possible with power button press. If ibm,power-off-ups token is used
 720  * it will allow auto poweron after power is restored.
 721  */
 722 static void pseries_power_off(void)
 723 {
 724         int rc;
 725         int rtas_poweroff_ups_token = rtas_token("ibm,power-off-ups");
 726
 727         if (rtas_flash_term_hook)
 728                 rtas_flash_term_hook(SYS_POWER_OFF);
 729
 730         if (rtas_poweron_auto == 0 ||
 731                 rtas_poweroff_ups_token == RTAS_UNKNOWN_SERVICE) {
 732                 rc = rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1);
 733                 printk(KERN_INFO "RTAS power-off returned %d\n", rc);
 734         } else {
 735                 rc = rtas_call(rtas_poweroff_ups_token, 0, 1, NULL);
 736                 printk(KERN_INFO "RTAS ibm,power-off-ups returned %d\n", rc);
 737         }
 738         for (;;);
 739 }
 740
 741 /*
 742  * Called very early, MMU is off, device-tree isn't unflattened
 743  */
 744
 745 static int __init pseries_probe_fw_features(unsigned long node,
 746                                             const char *uname, int depth,
 747                                             void *data)
 748 {
 749         const char *prop;
 750         int len;
 751         static int hypertas_found;
 752         static int vec5_found;
 753
 754         if (depth != 1)
 755                 return 0;
 756
 757         if (!strcmp(uname, "rtas") || !strcmp(uname, "rtas@0")) {
 758                 prop = of_get_flat_dt_prop(node, "ibm,hypertas-functions",
 759                                            &len);
 760                 if (prop) {
 761                         powerpc_firmware_features |= FW_FEATURE_LPAR;
 762                         fw_hypertas_feature_init(prop, len);
 763                 }
 764
 765                 hypertas_found = 1;
 766         }
 767
 768         if (!strcmp(uname, "chosen")) {
 769                 prop = of_get_flat_dt_prop(node, "ibm,architecture-vec-5",
 770                                            &len);
 771                 if (prop)
 772                         fw_vec5_feature_init(prop, len);
 773
 774                 vec5_found = 1;
 775         }
 776
 777         return hypertas_found && vec5_found;
 778 }
 779
 780 static int __init pSeries_probe(void)
 781 {
 782         unsigned long root = of_get_flat_dt_root();
 783         const char *dtype = of_get_flat_dt_prop(root, "device_type", NULL);
 784
 785         if (dtype == NULL)
 786                 return 0;
 787         if (strcmp(dtype, "chrp"))
 788                 return 0;
 789
 790         /* Cell blades firmware claims to be chrp while it's not. Until this
 791          * is fixed, we need to avoid those here.
 792          */
 793         if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
 794             of_flat_dt_is_compatible(root, "IBM,CBEA"))
 795                 return 0;
 796
 797         pr_debug("pSeries detected, looking for LPAR capability...\n");
 798
 799         /* Now try to figure out if we are running on LPAR */
 800         of_scan_flat_dt(pseries_probe_fw_features, NULL);
 801
 802 #ifdef __LITTLE_ENDIAN__
 803         if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
 804                 long rc;
 805                 /*
 806                  * Tell the hypervisor that we want our exceptions to
 807                  * be taken in little endian mode. If this fails we don't
 808                  * want to use BUG() because it will trigger an exception.
 809                  */
 810                 rc = pseries_little_endian_exceptions();
 811                 if (rc) {
 812                         ppc_md.progress("H_SET_MODE LE exception fail", 0);
 813                         panic("Could not enable little endian exceptions");
 814                 }
 815         }
 816 #endif
 817
 818         if (firmware_has_feature(FW_FEATURE_LPAR))
 819                 hpte_init_lpar();
 820         else
 821                 hpte_init_native();
 822
 823         pm_power_off = pseries_power_off;
 824
 825         pr_debug("Machine is%s LPAR !\n",
 826                  (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
 827
 828         return 1;
 829 }
 830
 831 static int pSeries_pci_probe_mode(struct pci_bus *bus)
 832 {
 833         if (firmware_has_feature(FW_FEATURE_LPAR))
 834                 return PCI_PROBE_DEVTREE;
 835         return PCI_PROBE_NORMAL;
 836 }
 837
 838 #ifndef CONFIG_PCI
 839 void pSeries_final_fixup(void) { }
 840 #endif
 841
 842 struct pci_controller_ops pseries_pci_controller_ops = {
 843         .probe_mode             = pSeries_pci_probe_mode,
 844 };
 845
 846 define_machine(pseries) {
 847         .name                   = "pSeries",
 848         .probe                  = pSeries_probe,
 849         .setup_arch             = pSeries_setup_arch,
 850         .init_early             = pSeries_init_early,
 851         .show_cpuinfo           = pSeries_show_cpuinfo,
 852         .log_error              = pSeries_log_error,
 853         .pcibios_fixup          = pSeries_final_fixup,
 854         .restart                = rtas_restart,
 855         .halt                   = rtas_halt,
 856         .panic                  = rtas_os_term,
 857         .get_boot_time          = rtas_get_boot_time,
 858         .get_rtc_time           = rtas_get_rtc_time,
 859         .set_rtc_time           = rtas_set_rtc_time,
 860         .calibrate_decr         = generic_calibrate_decr,
 861         .progress               = rtas_progress,
 862         .system_reset_exception = pSeries_system_reset_exception,
 863         .machine_check_exception = pSeries_machine_check_exception,
 864 #ifdef CONFIG_KEXEC
 865         .machine_kexec          = pSeries_machine_kexec,
 866 #endif
 867 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 868         .memory_block_size      = pseries_memory_block_size,
 869 #endif
 870 };