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