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/module.h>
  31 #include <linux/kexec.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/irq.h>
  34 #include <linux/memblock.h>
  35
  36 #include <asm/prom.h>
  37 #include <asm/rtas.h>
  38 #include <asm/page.h>
  39 #include <asm/processor.h>
  40 #include <asm/irq.h>
  41 #include <asm/io.h>
  42 #include <asm/kdump.h>
  43 #include <asm/smp.h>
  44 #include <asm/system.h>
  45 #include <asm/mmu.h>
  46 #include <asm/paca.h>
  47 #include <asm/pgtable.h>
  48 #include <asm/pci.h>
  49 #include <asm/iommu.h>
  50 #include <asm/btext.h>
  51 #include <asm/sections.h>
  52 #include <asm/machdep.h>
  53 #include <asm/pSeries_reconfig.h>
  54 #include <asm/pci-bridge.h>
  55 #include <asm/phyp_dump.h>
  56 #include <asm/kexec.h>
  57 #include <mm/mmu_decl.h>
  58
  59 #ifdef DEBUG
  60 #define DBG(fmt...) printk(KERN_ERR fmt)
  61 #else
  62 #define DBG(fmt...)
  63 #endif
  64
  65 #ifdef CONFIG_PPC64
  66 int __initdata iommu_is_off;
  67 int __initdata iommu_force_on;
  68 unsigned long tce_alloc_start, tce_alloc_end;
  69 u64 ppc64_rma_size;
  70 #endif
  71
  72 static int __init early_parse_mem(char *p)
  73 {
  74         if (!p)
  75                 return 1;
  76
  77         memory_limit = PAGE_ALIGN(memparse(p, &p));
  78         DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit);
  79
  80         return 0;
  81 }
  82 early_param("mem", early_parse_mem);
  83
  84 /**
  85  * move_device_tree - move tree to an unused area, if needed.
  86  *
  87  * The device tree may be allocated beyond our memory limit, or inside the
  88  * crash kernel region for kdump. If so, move it out of the way.
  89  */
  90 static void __init move_device_tree(void)
  91 {
  92         unsigned long start, size;
  93         void *p;
  94
  95         DBG("-> move_device_tree\n");
  96
  97         start = __pa(initial_boot_params);
  98         size = be32_to_cpu(initial_boot_params->totalsize);
  99
 100         if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
 101                         overlaps_crashkernel(start, size)) {
 102                 p = __va(memblock_alloc(size, PAGE_SIZE));
 103                 memcpy(p, initial_boot_params, size);
 104                 initial_boot_params = (struct boot_param_header *)p;
 105                 DBG("Moved device tree to 0x%p\n", p);
 106         }
 107
 108         DBG("<- move_device_tree\n");
 109 }
 110
 111 /*
 112  * ibm,pa-features is a per-cpu property that contains a string of
 113  * attribute descriptors, each of which has a 2 byte header plus up
 114  * to 254 bytes worth of processor attribute bits.  First header
 115  * byte specifies the number of bytes following the header.
 116  * Second header byte is an "attribute-specifier" type, of which
 117  * zero is the only currently-defined value.
 118  * Implementation:  Pass in the byte and bit offset for the feature
 119  * that we are interested in.  The function will return -1 if the
 120  * pa-features property is missing, or a 1/0 to indicate if the feature
 121  * is supported/not supported.  Note that the bit numbers are
 122  * big-endian to match the definition in PAPR.
 123  */
 124 static struct ibm_pa_feature {
 125         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 126         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 127         unsigned char   pabyte;         /* byte number in ibm,pa-features */
 128         unsigned char   pabit;          /* bit number (big-endian) */
 129         unsigned char   invert;         /* if 1, pa bit set => clear feature */
 130 } ibm_pa_features[] __initdata = {
 131         {0, PPC_FEATURE_HAS_MMU,        0, 0, 0},
 132         {0, PPC_FEATURE_HAS_FPU,        0, 1, 0},
 133         {CPU_FTR_SLB, 0,                0, 2, 0},
 134         {CPU_FTR_CTRL, 0,               0, 3, 0},
 135         {CPU_FTR_NOEXECUTE, 0,          0, 6, 0},
 136         {CPU_FTR_NODSISRALIGN, 0,       1, 1, 1},
 137         {CPU_FTR_CI_LARGE_PAGE, 0,      1, 2, 0},
 138         {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
 139 };
 140
 141 static void __init scan_features(unsigned long node, unsigned char *ftrs,
 142                                  unsigned long tablelen,
 143                                  struct ibm_pa_feature *fp,
 144                                  unsigned long ft_size)
 145 {
 146         unsigned long i, len, bit;
 147
 148         /* find descriptor with type == 0 */
 149         for (;;) {
 150                 if (tablelen < 3)
 151                         return;
 152                 len = 2 + ftrs[0];
 153                 if (tablelen < len)
 154                         return;         /* descriptor 0 not found */
 155                 if (ftrs[1] == 0)
 156                         break;
 157                 tablelen -= len;
 158                 ftrs += len;
 159         }
 160
 161         /* loop over bits we know about */
 162         for (i = 0; i < ft_size; ++i, ++fp) {
 163                 if (fp->pabyte >= ftrs[0])
 164                         continue;
 165                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 166                 if (bit ^ fp->invert) {
 167                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 168                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 169                 } else {
 170                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
 171                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
 172                 }
 173         }
 174 }
 175
 176 static void __init check_cpu_pa_features(unsigned long node)
 177 {
 178         unsigned char *pa_ftrs;
 179         unsigned long tablelen;
 180
 181         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
 182         if (pa_ftrs == NULL)
 183                 return;
 184
 185         scan_features(node, pa_ftrs, tablelen,
 186                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 187 }
 188
 189 #ifdef CONFIG_PPC_STD_MMU_64
 190 static void __init check_cpu_slb_size(unsigned long node)
 191 {
 192         u32 *slb_size_ptr;
 193
 194         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
 195         if (slb_size_ptr != NULL) {
 196                 mmu_slb_size = *slb_size_ptr;
 197                 return;
 198         }
 199         slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 200         if (slb_size_ptr != NULL) {
 201                 mmu_slb_size = *slb_size_ptr;
 202         }
 203 }
 204 #else
 205 #define check_cpu_slb_size(node) do { } while(0)
 206 #endif
 207
 208 static struct feature_property {
 209         const char *name;
 210         u32 min_value;
 211         unsigned long cpu_feature;
 212         unsigned long cpu_user_ftr;
 213 } feature_properties[] __initdata = {
 214 #ifdef CONFIG_ALTIVEC
 215         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 216         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 217 #endif /* CONFIG_ALTIVEC */
 218 #ifdef CONFIG_VSX
 219         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 220         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 221 #endif /* CONFIG_VSX */
 222 #ifdef CONFIG_PPC64
 223         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 224         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 225         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 226 #endif /* CONFIG_PPC64 */
 227 };
 228
 229 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 230 static inline void identical_pvr_fixup(unsigned long node)
 231 {
 232         unsigned int pvr;
 233         char *model = of_get_flat_dt_prop(node, "model", NULL);
 234
 235         /*
 236          * Since 440GR(x)/440EP(x) processors have the same pvr,
 237          * we check the node path and set bit 28 in the cur_cpu_spec
 238          * pvr for EP(x) processor version. This bit is always 0 in
 239          * the "real" pvr. Then we call identify_cpu again with
 240          * the new logical pvr to enable FPU support.
 241          */
 242         if (model && strstr(model, "440EP")) {
 243                 pvr = cur_cpu_spec->pvr_value | 0x8;
 244                 identify_cpu(0, pvr);
 245                 DBG("Using logical pvr %x for %s\n", pvr, model);
 246         }
 247 }
 248 #else
 249 #define identical_pvr_fixup(node) do { } while(0)
 250 #endif
 251
 252 static void __init check_cpu_feature_properties(unsigned long node)
 253 {
 254         unsigned long i;
 255         struct feature_property *fp = feature_properties;
 256         const u32 *prop;
 257
 258         for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
 259                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 260                 if (prop && *prop >= fp->min_value) {
 261                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 262                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 263                 }
 264         }
 265 }
 266
 267 static int __init early_init_dt_scan_cpus(unsigned long node,
 268                                           const char *uname, int depth,
 269                                           void *data)
 270 {
 271         static int logical_cpuid = 0;
 272         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 273         const u32 *prop;
 274         const u32 *intserv;
 275         int i, nthreads;
 276         unsigned long len;
 277         int found = 0;
 278
 279         /* We are scanning "cpu" nodes only */
 280         if (type == NULL || strcmp(type, "cpu") != 0)
 281                 return 0;
 282
 283         /* Get physical cpuid */
 284         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
 285         if (intserv) {
 286                 nthreads = len / sizeof(int);
 287         } else {
 288                 intserv = of_get_flat_dt_prop(node, "reg", NULL);
 289                 nthreads = 1;
 290         }
 291
 292         /*
 293          * Now see if any of these threads match our boot cpu.
 294          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
 295          */
 296         for (i = 0; i < nthreads; i++) {
 297                 /*
 298                  * version 2 of the kexec param format adds the phys cpuid of
 299                  * booted proc.
 300                  */
 301                 if (initial_boot_params && initial_boot_params->version >= 2) {
 302                         if (intserv[i] ==
 303                                         initial_boot_params->boot_cpuid_phys) {
 304                                 found = 1;
 305                                 break;
 306                         }
 307                 } else {
 308                         /*
 309                          * Check if it's the boot-cpu, set it's hw index now,
 310                          * unfortunately this format did not support booting
 311                          * off secondary threads.
 312                          */
 313                         if (of_get_flat_dt_prop(node,
 314                                         "linux,boot-cpu", NULL) != NULL) {
 315                                 found = 1;
 316                                 break;
 317                         }
 318                 }
 319
 320 #ifdef CONFIG_SMP
 321                 /* logical cpu id is always 0 on UP kernels */
 322                 logical_cpuid++;
 323 #endif
 324         }
 325
 326         if (found) {
 327                 DBG("boot cpu: logical %d physical %d\n", logical_cpuid,
 328                         intserv[i]);
 329                 boot_cpuid = logical_cpuid;
 330                 set_hard_smp_processor_id(boot_cpuid, intserv[i]);
 331
 332                 /*
 333                  * PAPR defines "logical" PVR values for cpus that
 334                  * meet various levels of the architecture:
 335                  * 0x0f000001   Architecture version 2.04
 336                  * 0x0f000002   Architecture version 2.05
 337                  * If the cpu-version property in the cpu node contains
 338                  * such a value, we call identify_cpu again with the
 339                  * logical PVR value in order to use the cpu feature
 340                  * bits appropriate for the architecture level.
 341                  *
 342                  * A POWER6 partition in "POWER6 architected" mode
 343                  * uses the 0x0f000002 PVR value; in POWER5+ mode
 344                  * it uses 0x0f000001.
 345                  */
 346                 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
 347                 if (prop && (*prop & 0xff000000) == 0x0f000000)
 348                         identify_cpu(0, *prop);
 349
 350                 identical_pvr_fixup(node);
 351         }
 352
 353         check_cpu_feature_properties(node);
 354         check_cpu_pa_features(node);
 355         check_cpu_slb_size(node);
 356
 357 #ifdef CONFIG_PPC_PSERIES
 358         if (nthreads > 1)
 359                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
 360         else
 361                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
 362 #endif
 363
 364         return 0;
 365 }
 366
 367 int __init early_init_dt_scan_chosen_ppc(unsigned long node, const char *uname,
 368                                          int depth, void *data)
 369 {
 370         unsigned long *lprop;
 371
 372         /* Use common scan routine to determine if this is the chosen node */
 373         if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
 374                 return 0;
 375
 376 #ifdef CONFIG_PPC64
 377         /* check if iommu is forced on or off */
 378         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
 379                 iommu_is_off = 1;
 380         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
 381                 iommu_force_on = 1;
 382 #endif
 383
 384         /* mem=x on the command line is the preferred mechanism */
 385         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
 386         if (lprop)
 387                 memory_limit = *lprop;
 388
 389 #ifdef CONFIG_PPC64
 390         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
 391         if (lprop)
 392                 tce_alloc_start = *lprop;
 393         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
 394         if (lprop)
 395                 tce_alloc_end = *lprop;
 396 #endif
 397
 398 #ifdef CONFIG_KEXEC
 399         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
 400         if (lprop)
 401                 crashk_res.start = *lprop;
 402
 403         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
 404         if (lprop)
 405                 crashk_res.end = crashk_res.start + *lprop - 1;
 406 #endif
 407
 408         /* break now */
 409         return 1;
 410 }
 411
 412 #ifdef CONFIG_PPC_PSERIES
 413 /*
 414  * Interpret the ibm,dynamic-memory property in the
 415  * /ibm,dynamic-reconfiguration-memory node.
 416  * This contains a list of memory blocks along with NUMA affinity
 417  * information.
 418  */
 419 static int __init early_init_dt_scan_drconf_memory(unsigned long node)
 420 {
 421         __be32 *dm, *ls, *usm;
 422         unsigned long l, n, flags;
 423         u64 base, size, memblock_size;
 424         unsigned int is_kexec_kdump = 0, rngs;
 425
 426         ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
 427         if (ls == NULL || l < dt_root_size_cells * sizeof(__be32))
 428                 return 0;
 429         memblock_size = dt_mem_next_cell(dt_root_size_cells, &ls);
 430
 431         dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
 432         if (dm == NULL || l < sizeof(__be32))
 433                 return 0;
 434
 435         n = *dm++;      /* number of entries */
 436         if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(__be32))
 437                 return 0;
 438
 439         /* check if this is a kexec/kdump kernel. */
 440         usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
 441                                                  &l);
 442         if (usm != NULL)
 443                 is_kexec_kdump = 1;
 444
 445         for (; n != 0; --n) {
 446                 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
 447                 flags = dm[3];
 448                 /* skip DRC index, pad, assoc. list index, flags */
 449                 dm += 4;
 450                 /* skip this block if the reserved bit is set in flags (0x80)
 451                    or if the block is not assigned to this partition (0x8) */
 452                 if ((flags & 0x80) || !(flags & 0x8))
 453                         continue;
 454                 size = memblock_size;
 455                 rngs = 1;
 456                 if (is_kexec_kdump) {
 457                         /*
 458                          * For each memblock in ibm,dynamic-memory, a corresponding
 459                          * entry in linux,drconf-usable-memory property contains
 460                          * a counter 'p' followed by 'p' (base, size) duple.
 461                          * Now read the counter from
 462                          * linux,drconf-usable-memory property
 463                          */
 464                         rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
 465                         if (!rngs) /* there are no (base, size) duple */
 466                                 continue;
 467                 }
 468                 do {
 469                         if (is_kexec_kdump) {
 470                                 base = dt_mem_next_cell(dt_root_addr_cells,
 471                                                          &usm);
 472                                 size = dt_mem_next_cell(dt_root_size_cells,
 473                                                          &usm);
 474                         }
 475                         if (iommu_is_off) {
 476                                 if (base >= 0x80000000ul)
 477                                         continue;
 478                                 if ((base + size) > 0x80000000ul)
 479                                         size = 0x80000000ul - base;
 480                         }
 481                         memblock_add(base, size);
 482                 } while (--rngs);
 483         }
 484         memblock_dump_all();
 485         return 0;
 486 }
 487 #else
 488 #define early_init_dt_scan_drconf_memory(node)  0
 489 #endif /* CONFIG_PPC_PSERIES */
 490
 491 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
 492                                                 const char *uname,
 493                                                 int depth, void *data)
 494 {
 495         if (depth == 1 &&
 496             strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
 497                 return early_init_dt_scan_drconf_memory(node);
 498
 499         return early_init_dt_scan_memory(node, uname, depth, data);
 500 }
 501
 502 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
 503 {
 504 #ifdef CONFIG_PPC64
 505         if (iommu_is_off) {
 506                 if (base >= 0x80000000ul)
 507                         return;
 508                 if ((base + size) > 0x80000000ul)
 509                         size = 0x80000000ul - base;
 510         }
 511 #endif
 512
 513         /* First MEMBLOCK added, do some special initializations */
 514         if (memstart_addr == ~(phys_addr_t)0)
 515                 setup_initial_memory_limit(base, size);
 516         memstart_addr = min((u64)memstart_addr, base);
 517
 518         /* Add the chunk to the MEMBLOCK list */
 519         memblock_add(base, size);
 520 }
 521
 522 void * __init early_init_dt_alloc_memory_arch(u64 size, u64 align)
 523 {
 524         return __va(memblock_alloc(size, align));
 525 }
 526
 527 #ifdef CONFIG_BLK_DEV_INITRD
 528 void __init early_init_dt_setup_initrd_arch(unsigned long start,
 529                 unsigned long end)
 530 {
 531         initrd_start = (unsigned long)__va(start);
 532         initrd_end = (unsigned long)__va(end);
 533         initrd_below_start_ok = 1;
 534 }
 535 #endif
 536
 537 static void __init early_reserve_mem(void)
 538 {
 539         u64 base, size;
 540         u64 *reserve_map;
 541         unsigned long self_base;
 542         unsigned long self_size;
 543
 544         reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
 545                                         initial_boot_params->off_mem_rsvmap);
 546
 547         /* before we do anything, lets reserve the dt blob */
 548         self_base = __pa((unsigned long)initial_boot_params);
 549         self_size = initial_boot_params->totalsize;
 550         memblock_reserve(self_base, self_size);
 551
 552 #ifdef CONFIG_BLK_DEV_INITRD
 553         /* then reserve the initrd, if any */
 554         if (initrd_start && (initrd_end > initrd_start))
 555                 memblock_reserve(__pa(initrd_start), initrd_end - initrd_start);
 556 #endif /* CONFIG_BLK_DEV_INITRD */
 557
 558 #ifdef CONFIG_PPC32
 559         /*
 560          * Handle the case where we might be booting from an old kexec
 561          * image that setup the mem_rsvmap as pairs of 32-bit values
 562          */
 563         if (*reserve_map > 0xffffffffull) {
 564                 u32 base_32, size_32;
 565                 u32 *reserve_map_32 = (u32 *)reserve_map;
 566
 567                 while (1) {
 568                         base_32 = *(reserve_map_32++);
 569                         size_32 = *(reserve_map_32++);
 570                         if (size_32 == 0)
 571                                 break;
 572                         /* skip if the reservation is for the blob */
 573                         if (base_32 == self_base && size_32 == self_size)
 574                                 continue;
 575                         DBG("reserving: %x -> %x\n", base_32, size_32);
 576                         memblock_reserve(base_32, size_32);
 577                 }
 578                 return;
 579         }
 580 #endif
 581         while (1) {
 582                 base = *(reserve_map++);
 583                 size = *(reserve_map++);
 584                 if (size == 0)
 585                         break;
 586                 DBG("reserving: %llx -> %llx\n", base, size);
 587                 memblock_reserve(base, size);
 588         }
 589 }
 590
 591 #ifdef CONFIG_PHYP_DUMP
 592 /**
 593  * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
 594  *
 595  * Function to find the largest size we need to reserve
 596  * during early boot process.
 597  *
 598  * It either looks for boot param and returns that OR
 599  * returns larger of 256 or 5% rounded down to multiples of 256MB.
 600  *
 601  */
 602 static inline unsigned long phyp_dump_calculate_reserve_size(void)
 603 {
 604         unsigned long tmp;
 605
 606         if (phyp_dump_info->reserve_bootvar)
 607                 return phyp_dump_info->reserve_bootvar;
 608
 609         /* divide by 20 to get 5% of value */
 610         tmp = memblock_end_of_DRAM();
 611         do_div(tmp, 20);
 612
 613         /* round it down in multiples of 256 */
 614         tmp = tmp & ~0x0FFFFFFFUL;
 615
 616         return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
 617 }
 618
 619 /**
 620  * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
 621  *
 622  * This routine may reserve memory regions in the kernel only
 623  * if the system is supported and a dump was taken in last
 624  * boot instance or if the hardware is supported and the
 625  * scratch area needs to be setup. In other instances it returns
 626  * without reserving anything. The memory in case of dump being
 627  * active is freed when the dump is collected (by userland tools).
 628  */
 629 static void __init phyp_dump_reserve_mem(void)
 630 {
 631         unsigned long base, size;
 632         unsigned long variable_reserve_size;
 633
 634         if (!phyp_dump_info->phyp_dump_configured) {
 635                 printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
 636                 return;
 637         }
 638
 639         if (!phyp_dump_info->phyp_dump_at_boot) {
 640                 printk(KERN_INFO "Phyp-dump disabled at boot time\n");
 641                 return;
 642         }
 643
 644         variable_reserve_size = phyp_dump_calculate_reserve_size();
 645
 646         if (phyp_dump_info->phyp_dump_is_active) {
 647                 /* Reserve *everything* above RMR.Area freed by userland tools*/
 648                 base = variable_reserve_size;
 649                 size = memblock_end_of_DRAM() - base;
 650
 651                 /* XXX crashed_ram_end is wrong, since it may be beyond
 652                  * the memory_limit, it will need to be adjusted. */
 653                 memblock_reserve(base, size);
 654
 655                 phyp_dump_info->init_reserve_start = base;
 656                 phyp_dump_info->init_reserve_size = size;
 657         } else {
 658                 size = phyp_dump_info->cpu_state_size +
 659                         phyp_dump_info->hpte_region_size +
 660                         variable_reserve_size;
 661                 base = memblock_end_of_DRAM() - size;
 662                 memblock_reserve(base, size);
 663                 phyp_dump_info->init_reserve_start = base;
 664                 phyp_dump_info->init_reserve_size = size;
 665         }
 666 }
 667 #else
 668 static inline void __init phyp_dump_reserve_mem(void) {}
 669 #endif /* CONFIG_PHYP_DUMP  && CONFIG_PPC_RTAS */
 670
 671 void __init early_init_devtree(void *params)
 672 {
 673         phys_addr_t limit;
 674
 675         DBG(" -> early_init_devtree(%p)\n", params);
 676
 677         /* Setup flat device-tree pointer */
 678         initial_boot_params = params;
 679
 680 #ifdef CONFIG_PPC_RTAS
 681         /* Some machines might need RTAS info for debugging, grab it now. */
 682         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 683 #endif
 684
 685 #ifdef CONFIG_PHYP_DUMP
 686         /* scan tree to see if dump occured during last boot */
 687         of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
 688 #endif
 689
 690         /* Retrieve various informations from the /chosen node of the
 691          * device-tree, including the platform type, initrd location and
 692          * size, TCE reserve, and more ...
 693          */
 694         of_scan_flat_dt(early_init_dt_scan_chosen_ppc, NULL);
 695
 696         /* Scan memory nodes and rebuild MEMBLOCKs */
 697         memblock_init();
 698
 699         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 700         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 701
 702         /* Save command line for /proc/cmdline and then parse parameters */
 703         strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
 704         parse_early_param();
 705
 706         /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
 707         memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
 708         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 709         if (PHYSICAL_START > MEMORY_START)
 710                 memblock_reserve(MEMORY_START, 0x8000);
 711         reserve_kdump_trampoline();
 712         reserve_crashkernel();
 713         early_reserve_mem();
 714         phyp_dump_reserve_mem();
 715
 716         limit = memory_limit;
 717         if (! limit) {
 718                 phys_addr_t memsize;
 719
 720                 /* Ensure that total memory size is page-aligned, because
 721                  * otherwise mark_bootmem() gets upset. */
 722                 memblock_analyze();
 723                 memsize = memblock_phys_mem_size();
 724                 if ((memsize & PAGE_MASK) != memsize)
 725                         limit = memsize & PAGE_MASK;
 726         }
 727         memblock_enforce_memory_limit(limit);
 728
 729         memblock_analyze();
 730         memblock_dump_all();
 731
 732         DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
 733
 734         /* We may need to relocate the flat tree, do it now.
 735          * FIXME .. and the initrd too? */
 736         move_device_tree();
 737
 738         allocate_pacas();
 739
 740         DBG("Scanning CPUs ...\n");
 741
 742         /* Retreive CPU related informations from the flat tree
 743          * (altivec support, boot CPU ID, ...)
 744          */
 745         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 746
 747         DBG(" <- early_init_devtree()\n");
 748 }
 749
 750 /*******
 751  *
 752  * New implementation of the OF "find" APIs, return a refcounted
 753  * object, call of_node_put() when done.  The device tree and list
 754  * are protected by a rw_lock.
 755  *
 756  * Note that property management will need some locking as well,
 757  * this isn't dealt with yet.
 758  *
 759  *******/
 760
 761 /**
 762  *      of_find_next_cache_node - Find a node's subsidiary cache
 763  *      @np:    node of type "cpu" or "cache"
 764  *
 765  *      Returns a node pointer with refcount incremented, use
 766  *      of_node_put() on it when done.  Caller should hold a reference
 767  *      to np.
 768  */
 769 struct device_node *of_find_next_cache_node(struct device_node *np)
 770 {
 771         struct device_node *child;
 772         const phandle *handle;
 773
 774         handle = of_get_property(np, "l2-cache", NULL);
 775         if (!handle)
 776                 handle = of_get_property(np, "next-level-cache", NULL);
 777
 778         if (handle)
 779                 return of_find_node_by_phandle(*handle);
 780
 781         /* OF on pmac has nodes instead of properties named "l2-cache"
 782          * beneath CPU nodes.
 783          */
 784         if (!strcmp(np->type, "cpu"))
 785                 for_each_child_of_node(np, child)
 786                         if (!strcmp(child->type, "cache"))
 787                                 return child;
 788
 789         return NULL;
 790 }
 791
 792 #ifdef CONFIG_PPC_PSERIES
 793 /*
 794  * Fix up the uninitialized fields in a new device node:
 795  * name, type and pci-specific fields
 796  */
 797
 798 static int of_finish_dynamic_node(struct device_node *node)
 799 {
 800         struct device_node *parent = of_get_parent(node);
 801         int err = 0;
 802         const phandle *ibm_phandle;
 803
 804         node->name = of_get_property(node, "name", NULL);
 805         node->type = of_get_property(node, "device_type", NULL);
 806
 807         if (!node->name)
 808                 node->name = "<NULL>";
 809         if (!node->type)
 810                 node->type = "<NULL>";
 811
 812         if (!parent) {
 813                 err = -ENODEV;
 814                 goto out;
 815         }
 816
 817         /* We don't support that function on PowerMac, at least
 818          * not yet
 819          */
 820         if (machine_is(powermac))
 821                 return -ENODEV;
 822
 823         /* fix up new node's phandle field */
 824         if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
 825                 node->phandle = *ibm_phandle;
 826
 827 out:
 828         of_node_put(parent);
 829         return err;
 830 }
 831
 832 static int prom_reconfig_notifier(struct notifier_block *nb,
 833                                   unsigned long action, void *node)
 834 {
 835         int err;
 836
 837         switch (action) {
 838         case PSERIES_RECONFIG_ADD:
 839                 err = of_finish_dynamic_node(node);
 840                 if (err < 0) {
 841                         printk(KERN_ERR "finish_node returned %d\n", err);
 842                         err = NOTIFY_BAD;
 843                 }
 844                 break;
 845         default:
 846                 err = NOTIFY_DONE;
 847                 break;
 848         }
 849         return err;
 850 }
 851
 852 static struct notifier_block prom_reconfig_nb = {
 853         .notifier_call = prom_reconfig_notifier,
 854         .priority = 10, /* This one needs to run first */
 855 };
 856
 857 static int __init prom_reconfig_setup(void)
 858 {
 859         return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
 860 }
 861 __initcall(prom_reconfig_setup);
 862 #endif
 863
 864 /* Find the device node for a given logical cpu number, also returns the cpu
 865  * local thread number (index in ibm,interrupt-server#s) if relevant and
 866  * asked for (non NULL)
 867  */
 868 struct device_node *of_get_cpu_node(int cpu, unsigned int *thread)
 869 {
 870         int hardid;
 871         struct device_node *np;
 872
 873         hardid = get_hard_smp_processor_id(cpu);
 874
 875         for_each_node_by_type(np, "cpu") {
 876                 const u32 *intserv;
 877                 unsigned int plen, t;
 878
 879                 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
 880                  * fallback to "reg" property and assume no threads
 881                  */
 882                 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
 883                                 &plen);
 884                 if (intserv == NULL) {
 885                         const u32 *reg = of_get_property(np, "reg", NULL);
 886                         if (reg == NULL)
 887                                 continue;
 888                         if (*reg == hardid) {
 889                                 if (thread)
 890                                         *thread = 0;
 891                                 return np;
 892                         }
 893                 } else {
 894                         plen /= sizeof(u32);
 895                         for (t = 0; t < plen; t++) {
 896                                 if (hardid == intserv[t]) {
 897                                         if (thread)
 898                                                 *thread = t;
 899                                         return np;
 900                                 }
 901                         }
 902                 }
 903         }
 904         return NULL;
 905 }
 906 EXPORT_SYMBOL(of_get_cpu_node);
 907
 908 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
 909 static struct debugfs_blob_wrapper flat_dt_blob;
 910
 911 static int __init export_flat_device_tree(void)
 912 {
 913         struct dentry *d;
 914
 915         flat_dt_blob.data = initial_boot_params;
 916         flat_dt_blob.size = initial_boot_params->totalsize;
 917
 918         d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
 919                                 powerpc_debugfs_root, &flat_dt_blob);
 920         if (!d)
 921                 return 1;
 922
 923         return 0;
 924 }
 925 __initcall(export_flat_device_tree);
 926 #endif