arch/powerpc/kernel/prom.c

   1 /*
   2  * Procedures for creating, accessing and interpreting the device tree.
   3  *
   4  * Paul Mackerras       August 1996.
   5  * Copyright (C) 1996-2005 Paul Mackerras.
   6  *
   7  *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
   8  *    {engebret|bergner}@us.ibm.com
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  */
  15
  16 #undef DEBUG
  17
  18 #include <stdarg.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/init.h>
  22 #include <linux/threads.h>
  23 #include <linux/spinlock.h>
  24 #include <linux/types.h>
  25 #include <linux/pci.h>
  26 #include <linux/stringify.h>
  27 #include <linux/delay.h>
  28 #include <linux/initrd.h>
  29 #include <linux/bitops.h>
  30 #include <linux/export.h>
  31 #include <linux/kexec.h>
  32 #include <linux/irq.h>
  33 #include <linux/memblock.h>
  34 #include <linux/of.h>
  35 #include <linux/of_fdt.h>
  36 #include <linux/libfdt.h>
  37
  38 #include <asm/prom.h>
  39 #include <asm/rtas.h>
  40 #include <asm/page.h>
  41 #include <asm/processor.h>
  42 #include <asm/irq.h>
  43 #include <asm/io.h>
  44 #include <asm/kdump.h>
  45 #include <asm/smp.h>
  46 #include <asm/mmu.h>
  47 #include <asm/paca.h>
  48 #include <asm/pgtable.h>
  49 #include <asm/pci.h>
  50 #include <asm/iommu.h>
  51 #include <asm/btext.h>
  52 #include <asm/sections.h>
  53 #include <asm/machdep.h>
  54 #include <asm/pci-bridge.h>
  55 #include <asm/kexec.h>
  56 #include <asm/opal.h>
  57 #include <asm/fadump.h>
  58 #include <asm/debug.h>
  59
  60 #include <mm/mmu_decl.h>
  61
  62 #ifdef DEBUG
  63 #define DBG(fmt...) printk(KERN_ERR fmt)
  64 #else
  65 #define DBG(fmt...)
  66 #endif
  67
  68 #ifdef CONFIG_PPC64
  69 int __initdata iommu_is_off;
  70 int __initdata iommu_force_on;
  71 unsigned long tce_alloc_start, tce_alloc_end;
  72 u64 ppc64_rma_size;
  73 #endif
  74 static phys_addr_t first_memblock_size;
  75 static int __initdata boot_cpu_count;
  76
  77 static int __init early_parse_mem(char *p)
  78 {
  79         if (!p)
  80                 return 1;
  81
  82         memory_limit = PAGE_ALIGN(memparse(p, &p));
  83         DBG("memory limit = 0x%llx\n", memory_limit);
  84
  85         return 0;
  86 }
  87 early_param("mem", early_parse_mem);
  88
  89 /*
  90  * overlaps_initrd - check for overlap with page aligned extension of
  91  * initrd.
  92  */
  93 static inline int overlaps_initrd(unsigned long start, unsigned long size)
  94 {
  95 #ifdef CONFIG_BLK_DEV_INITRD
  96         if (!initrd_start)
  97                 return 0;
  98
  99         return  (start + size) > _ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
 100                         start <= _ALIGN_UP(initrd_end, PAGE_SIZE);
 101 #else
 102         return 0;
 103 #endif
 104 }
 105
 106 /**
 107  * move_device_tree - move tree to an unused area, if needed.
 108  *
 109  * The device tree may be allocated beyond our memory limit, or inside the
 110  * crash kernel region for kdump, or within the page aligned range of initrd.
 111  * If so, move it out of the way.
 112  */
 113 static void __init move_device_tree(void)
 114 {
 115         unsigned long start, size;
 116         void *p;
 117
 118         DBG("-> move_device_tree\n");
 119
 120         start = __pa(initial_boot_params);
 121         size = fdt_totalsize(initial_boot_params);
 122
 123         if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
 124                         overlaps_crashkernel(start, size) ||
 125                         overlaps_initrd(start, size)) {
 126                 p = __va(memblock_alloc(size, PAGE_SIZE));
 127                 memcpy(p, initial_boot_params, size);
 128                 initial_boot_params = p;
 129                 DBG("Moved device tree to 0x%p\n", p);
 130         }
 131
 132         DBG("<- move_device_tree\n");
 133 }
 134
 135 /*
 136  * ibm,pa-features is a per-cpu property that contains a string of
 137  * attribute descriptors, each of which has a 2 byte header plus up
 138  * to 254 bytes worth of processor attribute bits.  First header
 139  * byte specifies the number of bytes following the header.
 140  * Second header byte is an "attribute-specifier" type, of which
 141  * zero is the only currently-defined value.
 142  * Implementation:  Pass in the byte and bit offset for the feature
 143  * that we are interested in.  The function will return -1 if the
 144  * pa-features property is missing, or a 1/0 to indicate if the feature
 145  * is supported/not supported.  Note that the bit numbers are
 146  * big-endian to match the definition in PAPR.
 147  */
 148 static struct ibm_pa_feature {
 149         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 150         unsigned long   mmu_features;   /* MMU_FTR_xxx bit */
 151         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 152         unsigned char   pabyte;         /* byte number in ibm,pa-features */
 153         unsigned char   pabit;          /* bit number (big-endian) */
 154         unsigned char   invert;         /* if 1, pa bit set => clear feature */
 155 } ibm_pa_features[] __initdata = {
 156         {0, 0, PPC_FEATURE_HAS_MMU,     0, 0, 0},
 157         {0, 0, PPC_FEATURE_HAS_FPU,     0, 1, 0},
 158         {0, MMU_FTR_SLB, 0,             0, 2, 0},
 159         {CPU_FTR_CTRL, 0, 0,            0, 3, 0},
 160         {CPU_FTR_NOEXECUTE, 0, 0,       0, 6, 0},
 161         {CPU_FTR_NODSISRALIGN, 0, 0,    1, 1, 1},
 162         {0, MMU_FTR_CI_LARGE_PAGE, 0,   1, 2, 0},
 163         {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
 164 };
 165
 166 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
 167                                  unsigned long tablelen,
 168                                  struct ibm_pa_feature *fp,
 169                                  unsigned long ft_size)
 170 {
 171         unsigned long i, len, bit;
 172
 173         /* find descriptor with type == 0 */
 174         for (;;) {
 175                 if (tablelen < 3)
 176                         return;
 177                 len = 2 + ftrs[0];
 178                 if (tablelen < len)
 179                         return;         /* descriptor 0 not found */
 180                 if (ftrs[1] == 0)
 181                         break;
 182                 tablelen -= len;
 183                 ftrs += len;
 184         }
 185
 186         /* loop over bits we know about */
 187         for (i = 0; i < ft_size; ++i, ++fp) {
 188                 if (fp->pabyte >= ftrs[0])
 189                         continue;
 190                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 191                 if (bit ^ fp->invert) {
 192                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 193                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 194                         cur_cpu_spec->mmu_features |= fp->mmu_features;
 195                 } else {
 196                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
 197                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
 198                         cur_cpu_spec->mmu_features &= ~fp->mmu_features;
 199                 }
 200         }
 201 }
 202
 203 static void __init check_cpu_pa_features(unsigned long node)
 204 {
 205         const unsigned char *pa_ftrs;
 206         int tablelen;
 207
 208         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
 209         if (pa_ftrs == NULL)
 210                 return;
 211
 212         scan_features(node, pa_ftrs, tablelen,
 213                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 214 }
 215
 216 #ifdef CONFIG_PPC_STD_MMU_64
 217 static void __init check_cpu_slb_size(unsigned long node)
 218 {
 219         const __be32 *slb_size_ptr;
 220
 221         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
 222         if (slb_size_ptr != NULL) {
 223                 mmu_slb_size = be32_to_cpup(slb_size_ptr);
 224                 return;
 225         }
 226         slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 227         if (slb_size_ptr != NULL) {
 228                 mmu_slb_size = be32_to_cpup(slb_size_ptr);
 229         }
 230 }
 231 #else
 232 #define check_cpu_slb_size(node) do { } while(0)
 233 #endif
 234
 235 static struct feature_property {
 236         const char *name;
 237         u32 min_value;
 238         unsigned long cpu_feature;
 239         unsigned long cpu_user_ftr;
 240 } feature_properties[] __initdata = {
 241 #ifdef CONFIG_ALTIVEC
 242         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 243         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 244 #endif /* CONFIG_ALTIVEC */
 245 #ifdef CONFIG_VSX
 246         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 247         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 248 #endif /* CONFIG_VSX */
 249 #ifdef CONFIG_PPC64
 250         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 251         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 252         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 253 #endif /* CONFIG_PPC64 */
 254 };
 255
 256 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 257 static inline void identical_pvr_fixup(unsigned long node)
 258 {
 259         unsigned int pvr;
 260         const char *model = of_get_flat_dt_prop(node, "model", NULL);
 261
 262         /*
 263          * Since 440GR(x)/440EP(x) processors have the same pvr,
 264          * we check the node path and set bit 28 in the cur_cpu_spec
 265          * pvr for EP(x) processor version. This bit is always 0 in
 266          * the "real" pvr. Then we call identify_cpu again with
 267          * the new logical pvr to enable FPU support.
 268          */
 269         if (model && strstr(model, "440EP")) {
 270                 pvr = cur_cpu_spec->pvr_value | 0x8;
 271                 identify_cpu(0, pvr);
 272                 DBG("Using logical pvr %x for %s\n", pvr, model);
 273         }
 274 }
 275 #else
 276 #define identical_pvr_fixup(node) do { } while(0)
 277 #endif
 278
 279 static void __init check_cpu_feature_properties(unsigned long node)
 280 {
 281         unsigned long i;
 282         struct feature_property *fp = feature_properties;
 283         const __be32 *prop;
 284
 285         for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
 286                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 287                 if (prop && be32_to_cpup(prop) >= fp->min_value) {
 288                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 289                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 290                 }
 291         }
 292 }
 293
 294 static int __init early_init_dt_scan_cpus(unsigned long node,
 295                                           const char *uname, int depth,
 296                                           void *data)
 297 {
 298         const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 299         const __be32 *prop;
 300         const __be32 *intserv;
 301         int i, nthreads;
 302         int len;
 303         int found = -1;
 304         int found_thread = 0;
 305
 306         /* We are scanning "cpu" nodes only */
 307         if (type == NULL || strcmp(type, "cpu") != 0)
 308                 return 0;
 309
 310         /* Get physical cpuid */
 311         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
 312         if (intserv) {
 313                 nthreads = len / sizeof(int);
 314         } else {
 315                 intserv = of_get_flat_dt_prop(node, "reg", NULL);
 316                 nthreads = 1;
 317         }
 318
 319         /*
 320          * Now see if any of these threads match our boot cpu.
 321          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
 322          */
 323         for (i = 0; i < nthreads; i++) {
 324                 /*
 325                  * version 2 of the kexec param format adds the phys cpuid of
 326                  * booted proc.
 327                  */
 328                 if (fdt_version(initial_boot_params) >= 2) {
 329                         if (be32_to_cpu(intserv[i]) ==
 330                             fdt_boot_cpuid_phys(initial_boot_params)) {
 331                                 found = boot_cpu_count;
 332                                 found_thread = i;
 333                         }
 334                 } else {
 335                         /*
 336                          * Check if it's the boot-cpu, set it's hw index now,
 337                          * unfortunately this format did not support booting
 338                          * off secondary threads.
 339                          */
 340                         if (of_get_flat_dt_prop(node,
 341                                         "linux,boot-cpu", NULL) != NULL)
 342                                 found = boot_cpu_count;
 343                 }
 344 #ifdef CONFIG_SMP
 345                 /* logical cpu id is always 0 on UP kernels */
 346                 boot_cpu_count++;
 347 #endif
 348         }
 349
 350         /* Not the boot CPU */
 351         if (found < 0)
 352                 return 0;
 353
 354         DBG("boot cpu: logical %d physical %d\n", found,
 355             be32_to_cpu(intserv[found_thread]));
 356         boot_cpuid = found;
 357         set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
 358
 359         /*
 360          * PAPR defines "logical" PVR values for cpus that
 361          * meet various levels of the architecture:
 362          * 0x0f000001   Architecture version 2.04
 363          * 0x0f000002   Architecture version 2.05
 364          * If the cpu-version property in the cpu node contains
 365          * such a value, we call identify_cpu again with the
 366          * logical PVR value in order to use the cpu feature
 367          * bits appropriate for the architecture level.
 368          *
 369          * A POWER6 partition in "POWER6 architected" mode
 370          * uses the 0x0f000002 PVR value; in POWER5+ mode
 371          * it uses 0x0f000001.
 372          */
 373         prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
 374         if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
 375                 identify_cpu(0, be32_to_cpup(prop));
 376
 377         identical_pvr_fixup(node);
 378
 379         check_cpu_feature_properties(node);
 380         check_cpu_pa_features(node);
 381         check_cpu_slb_size(node);
 382
 383 #ifdef CONFIG_PPC64
 384         if (nthreads > 1)
 385                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
 386         else
 387                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
 388 #endif
 389         return 0;
 390 }
 391
 392 int __init early_init_dt_scan_chosen_ppc(unsigned long node, const char *uname,
 393                                          int depth, void *data)
 394 {
 395         const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
 396
 397         /* Use common scan routine to determine if this is the chosen node */
 398         if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
 399                 return 0;
 400
 401 #ifdef CONFIG_PPC64
 402         /* check if iommu is forced on or off */
 403         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
 404                 iommu_is_off = 1;
 405         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
 406                 iommu_force_on = 1;
 407 #endif
 408
 409         /* mem=x on the command line is the preferred mechanism */
 410         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
 411         if (lprop)
 412                 memory_limit = *lprop;
 413
 414 #ifdef CONFIG_PPC64
 415         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
 416         if (lprop)
 417                 tce_alloc_start = *lprop;
 418         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
 419         if (lprop)
 420                 tce_alloc_end = *lprop;
 421 #endif
 422
 423 #ifdef CONFIG_KEXEC
 424         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
 425         if (lprop)
 426                 crashk_res.start = *lprop;
 427
 428         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
 429         if (lprop)
 430                 crashk_res.end = crashk_res.start + *lprop - 1;
 431 #endif
 432
 433         /* break now */
 434         return 1;
 435 }
 436
 437 #ifdef CONFIG_PPC_PSERIES
 438 /*
 439  * Interpret the ibm,dynamic-memory property in the
 440  * /ibm,dynamic-reconfiguration-memory node.
 441  * This contains a list of memory blocks along with NUMA affinity
 442  * information.
 443  */
 444 static int __init early_init_dt_scan_drconf_memory(unsigned long node)
 445 {
 446         const __be32 *dm, *ls, *usm;
 447         int l;
 448         unsigned long n, flags;
 449         u64 base, size, memblock_size;
 450         unsigned int is_kexec_kdump = 0, rngs;
 451
 452         ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
 453         if (ls == NULL || l < dt_root_size_cells * sizeof(__be32))
 454                 return 0;
 455         memblock_size = dt_mem_next_cell(dt_root_size_cells, &ls);
 456
 457         dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
 458         if (dm == NULL || l < sizeof(__be32))
 459                 return 0;
 460
 461         n = of_read_number(dm++, 1);    /* number of entries */
 462         if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(__be32))
 463                 return 0;
 464
 465         /* check if this is a kexec/kdump kernel. */
 466         usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
 467                                                  &l);
 468         if (usm != NULL)
 469                 is_kexec_kdump = 1;
 470
 471         for (; n != 0; --n) {
 472                 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
 473                 flags = of_read_number(&dm[3], 1);
 474                 /* skip DRC index, pad, assoc. list index, flags */
 475                 dm += 4;
 476                 /* skip this block if the reserved bit is set in flags (0x80)
 477                    or if the block is not assigned to this partition (0x8) */
 478                 if ((flags & 0x80) || !(flags & 0x8))
 479                         continue;
 480                 size = memblock_size;
 481                 rngs = 1;
 482                 if (is_kexec_kdump) {
 483                         /*
 484                          * For each memblock in ibm,dynamic-memory, a corresponding
 485                          * entry in linux,drconf-usable-memory property contains
 486                          * a counter 'p' followed by 'p' (base, size) duple.
 487                          * Now read the counter from
 488                          * linux,drconf-usable-memory property
 489                          */
 490                         rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
 491                         if (!rngs) /* there are no (base, size) duple */
 492                                 continue;
 493                 }
 494                 do {
 495                         if (is_kexec_kdump) {
 496                                 base = dt_mem_next_cell(dt_root_addr_cells,
 497                                                          &usm);
 498                                 size = dt_mem_next_cell(dt_root_size_cells,
 499                                                          &usm);
 500                         }
 501                         if (iommu_is_off) {
 502                                 if (base >= 0x80000000ul)
 503                                         continue;
 504                                 if ((base + size) > 0x80000000ul)
 505                                         size = 0x80000000ul - base;
 506                         }
 507                         memblock_add(base, size);
 508                 } while (--rngs);
 509         }
 510         memblock_dump_all();
 511         return 0;
 512 }
 513 #else
 514 #define early_init_dt_scan_drconf_memory(node)  0
 515 #endif /* CONFIG_PPC_PSERIES */
 516
 517 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
 518                                                 const char *uname,
 519                                                 int depth, void *data)
 520 {
 521         if (depth == 1 &&
 522             strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
 523                 return early_init_dt_scan_drconf_memory(node);
 524
 525         return early_init_dt_scan_memory(node, uname, depth, data);
 526 }
 527
 528 /*
 529  * For a relocatable kernel, we need to get the memstart_addr first,
 530  * then use it to calculate the virtual kernel start address. This has
 531  * to happen at a very early stage (before machine_init). In this case,
 532  * we just want to get the memstart_address and would not like to mess the
 533  * memblock at this stage. So introduce a variable to skip the memblock_add()
 534  * for this reason.
 535  */
 536 #ifdef CONFIG_RELOCATABLE
 537 static int add_mem_to_memblock = 1;
 538 #else
 539 #define add_mem_to_memblock 1
 540 #endif
 541
 542 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
 543 {
 544 #ifdef CONFIG_PPC64
 545         if (iommu_is_off) {
 546                 if (base >= 0x80000000ul)
 547                         return;
 548                 if ((base + size) > 0x80000000ul)
 549                         size = 0x80000000ul - base;
 550         }
 551 #endif
 552         /* Keep track of the beginning of memory -and- the size of
 553          * the very first block in the device-tree as it represents
 554          * the RMA on ppc64 server
 555          */
 556         if (base < memstart_addr) {
 557                 memstart_addr = base;
 558                 first_memblock_size = size;
 559         }
 560
 561         /* Add the chunk to the MEMBLOCK list */
 562         if (add_mem_to_memblock)
 563                 memblock_add(base, size);
 564 }
 565
 566 static void __init early_reserve_mem_dt(void)
 567 {
 568         unsigned long i, dt_root;
 569         int len;
 570         const __be32 *prop;
 571
 572         early_init_fdt_scan_reserved_mem();
 573
 574         dt_root = of_get_flat_dt_root();
 575
 576         prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
 577
 578         if (!prop)
 579                 return;
 580
 581         DBG("Found new-style reserved-ranges\n");
 582
 583         /* Each reserved range is an (address,size) pair, 2 cells each,
 584          * totalling 4 cells per range. */
 585         for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
 586                 u64 base, size;
 587
 588                 base = of_read_number(prop + (i * 4) + 0, 2);
 589                 size = of_read_number(prop + (i * 4) + 2, 2);
 590
 591                 if (size) {
 592                         DBG("reserving: %llx -> %llx\n", base, size);
 593                         memblock_reserve(base, size);
 594                 }
 595         }
 596 }
 597
 598 static void __init early_reserve_mem(void)
 599 {
 600         __be64 *reserve_map;
 601
 602         reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
 603                         fdt_off_mem_rsvmap(initial_boot_params));
 604
 605         /* Look for the new "reserved-regions" property in the DT */
 606         early_reserve_mem_dt();
 607
 608 #ifdef CONFIG_BLK_DEV_INITRD
 609         /* Then reserve the initrd, if any */
 610         if (initrd_start && (initrd_end > initrd_start)) {
 611                 memblock_reserve(_ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
 612                         _ALIGN_UP(initrd_end, PAGE_SIZE) -
 613                         _ALIGN_DOWN(initrd_start, PAGE_SIZE));
 614         }
 615 #endif /* CONFIG_BLK_DEV_INITRD */
 616
 617 #ifdef CONFIG_PPC32
 618         /*
 619          * Handle the case where we might be booting from an old kexec
 620          * image that setup the mem_rsvmap as pairs of 32-bit values
 621          */
 622         if (be64_to_cpup(reserve_map) > 0xffffffffull) {
 623                 u32 base_32, size_32;
 624                 __be32 *reserve_map_32 = (__be32 *)reserve_map;
 625
 626                 DBG("Found old 32-bit reserve map\n");
 627
 628                 while (1) {
 629                         base_32 = be32_to_cpup(reserve_map_32++);
 630                         size_32 = be32_to_cpup(reserve_map_32++);
 631                         if (size_32 == 0)
 632                                 break;
 633                         DBG("reserving: %x -> %x\n", base_32, size_32);
 634                         memblock_reserve(base_32, size_32);
 635                 }
 636                 return;
 637         }
 638 #endif
 639 }
 640
 641 void __init early_init_devtree(void *params)
 642 {
 643         phys_addr_t limit;
 644
 645         DBG(" -> early_init_devtree(%p)\n", params);
 646
 647         /* Setup flat device-tree pointer */
 648         initial_boot_params = params;
 649
 650 #ifdef CONFIG_PPC_RTAS
 651         /* Some machines might need RTAS info for debugging, grab it now. */
 652         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 653 #endif
 654
 655 #ifdef CONFIG_PPC_POWERNV
 656         /* Some machines might need OPAL info for debugging, grab it now. */
 657         of_scan_flat_dt(early_init_dt_scan_opal, NULL);
 658 #endif
 659
 660 #ifdef CONFIG_FA_DUMP
 661         /* scan tree to see if dump is active during last boot */
 662         of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
 663 #endif
 664
 665         /* Pre-initialize the cmd_line with the content of boot_commmand_line,
 666          * which will be empty except when the content of the variable has
 667          * been overriden by a bootloading mechanism. This happens typically
 668          * with HAL takeover
 669          */
 670         strlcpy(cmd_line, boot_command_line, COMMAND_LINE_SIZE);
 671
 672         /* Retrieve various informations from the /chosen node of the
 673          * device-tree, including the platform type, initrd location and
 674          * size, TCE reserve, and more ...
 675          */
 676         of_scan_flat_dt(early_init_dt_scan_chosen_ppc, cmd_line);
 677
 678         /* Scan memory nodes and rebuild MEMBLOCKs */
 679         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 680         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 681
 682         /* Save command line for /proc/cmdline and then parse parameters */
 683         strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
 684         parse_early_param();
 685
 686         /* make sure we've parsed cmdline for mem= before this */
 687         if (memory_limit)
 688                 first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
 689         setup_initial_memory_limit(memstart_addr, first_memblock_size);
 690         /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
 691         memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
 692         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 693         if (PHYSICAL_START > MEMORY_START)
 694                 memblock_reserve(MEMORY_START, 0x8000);
 695         reserve_kdump_trampoline();
 696 #ifdef CONFIG_FA_DUMP
 697         /*
 698          * If we fail to reserve memory for firmware-assisted dump then
 699          * fallback to kexec based kdump.
 700          */
 701         if (fadump_reserve_mem() == 0)
 702 #endif
 703                 reserve_crashkernel();
 704         early_reserve_mem();
 705
 706         /*
 707          * Ensure that total memory size is page-aligned, because otherwise
 708          * mark_bootmem() gets upset.
 709          */
 710         limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
 711         memblock_enforce_memory_limit(limit);
 712
 713         memblock_allow_resize();
 714         memblock_dump_all();
 715
 716         DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
 717
 718         /* We may need to relocate the flat tree, do it now.
 719          * FIXME .. and the initrd too? */
 720         move_device_tree();
 721
 722         allocate_pacas();
 723
 724         DBG("Scanning CPUs ...\n");
 725
 726         /* Retrieve CPU related informations from the flat tree
 727          * (altivec support, boot CPU ID, ...)
 728          */
 729         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 730         if (boot_cpuid < 0) {
 731                 printk("Failed to indentify boot CPU !\n");
 732                 BUG();
 733         }
 734
 735 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
 736         /* We'll later wait for secondaries to check in; there are
 737          * NCPUS-1 non-boot CPUs  :-)
 738          */
 739         spinning_secondaries = boot_cpu_count - 1;
 740 #endif
 741
 742 #ifdef CONFIG_PPC_POWERNV
 743         /* Scan and build the list of machine check recoverable ranges */
 744         of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
 745 #endif
 746
 747         DBG(" <- early_init_devtree()\n");
 748 }
 749
 750 #ifdef CONFIG_RELOCATABLE
 751 /*
 752  * This function run before early_init_devtree, so we have to init
 753  * initial_boot_params.
 754  */
 755 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
 756 {
 757         /* Setup flat device-tree pointer */
 758         initial_boot_params = params;
 759
 760         /*
 761          * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
 762          * mess the memblock.
 763          */
 764         add_mem_to_memblock = 0;
 765         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 766         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 767         add_mem_to_memblock = 1;
 768
 769         if (size)
 770                 *size = first_memblock_size;
 771 }
 772 #endif
 773
 774 /*******
 775  *
 776  * New implementation of the OF "find" APIs, return a refcounted
 777  * object, call of_node_put() when done.  The device tree and list
 778  * are protected by a rw_lock.
 779  *
 780  * Note that property management will need some locking as well,
 781  * this isn't dealt with yet.
 782  *
 783  *******/
 784
 785 /**
 786  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
 787  * @np: device node of the device
 788  *
 789  * This looks for a property "ibm,chip-id" in the node or any
 790  * of its parents and returns its content, or -1 if it cannot
 791  * be found.
 792  */
 793 int of_get_ibm_chip_id(struct device_node *np)
 794 {
 795         of_node_get(np);
 796         while(np) {
 797                 struct device_node *old = np;
 798                 const __be32 *prop;
 799
 800                 prop = of_get_property(np, "ibm,chip-id", NULL);
 801                 if (prop) {
 802                         of_node_put(np);
 803                         return be32_to_cpup(prop);
 804                 }
 805                 np = of_get_parent(np);
 806                 of_node_put(old);
 807         }
 808         return -1;
 809 }
 810
 811 /**
 812  * cpu_to_chip_id - Return the cpus chip-id
 813  * @cpu: The logical cpu number.
 814  *
 815  * Return the value of the ibm,chip-id property corresponding to the given
 816  * logical cpu number. If the chip-id can not be found, returns -1.
 817  */
 818 int cpu_to_chip_id(int cpu)
 819 {
 820         struct device_node *np;
 821
 822         np = of_get_cpu_node(cpu, NULL);
 823         if (!np)
 824                 return -1;
 825
 826         of_node_put(np);
 827         return of_get_ibm_chip_id(np);
 828 }
 829 EXPORT_SYMBOL(cpu_to_chip_id);
 830
 831 #ifdef CONFIG_PPC_PSERIES
 832 /*
 833  * Fix up the uninitialized fields in a new device node:
 834  * name, type and pci-specific fields
 835  */
 836
 837 static int of_finish_dynamic_node(struct device_node *node)
 838 {
 839         struct device_node *parent = of_get_parent(node);
 840         int err = 0;
 841         const phandle *ibm_phandle;
 842
 843         node->name = of_get_property(node, "name", NULL);
 844         node->type = of_get_property(node, "device_type", NULL);
 845
 846         if (!node->name)
 847                 node->name = "<NULL>";
 848         if (!node->type)
 849                 node->type = "<NULL>";
 850
 851         if (!parent) {
 852                 err = -ENODEV;
 853                 goto out;
 854         }
 855
 856         /* We don't support that function on PowerMac, at least
 857          * not yet
 858          */
 859         if (machine_is(powermac))
 860                 return -ENODEV;
 861
 862         /* fix up new node's phandle field */
 863         if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
 864                 node->phandle = *ibm_phandle;
 865
 866 out:
 867         of_node_put(parent);
 868         return err;
 869 }
 870
 871 static int prom_reconfig_notifier(struct notifier_block *nb,
 872                                   unsigned long action, void *node)
 873 {
 874         int err;
 875
 876         switch (action) {
 877         case OF_RECONFIG_ATTACH_NODE:
 878                 err = of_finish_dynamic_node(node);
 879                 if (err < 0)
 880                         printk(KERN_ERR "finish_node returned %d\n", err);
 881                 break;
 882         default:
 883                 err = 0;
 884                 break;
 885         }
 886         return notifier_from_errno(err);
 887 }
 888
 889 static struct notifier_block prom_reconfig_nb = {
 890         .notifier_call = prom_reconfig_notifier,
 891         .priority = 10, /* This one needs to run first */
 892 };
 893
 894 static int __init prom_reconfig_setup(void)
 895 {
 896         return of_reconfig_notifier_register(&prom_reconfig_nb);
 897 }
 898 __initcall(prom_reconfig_setup);
 899 #endif
 900
 901 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
 902 {
 903         return (int)phys_id == get_hard_smp_processor_id(cpu);
 904 }