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 static phys_addr_t first_memblock_size;
  72
  73 static int __init early_parse_mem(char *p)
  74 {
  75         if (!p)
  76                 return 1;
  77
  78         memory_limit = PAGE_ALIGN(memparse(p, &p));
  79         DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit);
  80
  81         return 0;
  82 }
  83 early_param("mem", early_parse_mem);
  84
  85 /*
  86  * overlaps_initrd - check for overlap with page aligned extension of
  87  * initrd.
  88  */
  89 static inline int overlaps_initrd(unsigned long start, unsigned long size)
  90 {
  91 #ifdef CONFIG_BLK_DEV_INITRD
  92         if (!initrd_start)
  93                 return 0;
  94
  95         return  (start + size) > _ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
  96                         start <= _ALIGN_UP(initrd_end, PAGE_SIZE);
  97 #else
  98         return 0;
  99 #endif
 100 }
 101
 102 /**
 103  * move_device_tree - move tree to an unused area, if needed.
 104  *
 105  * The device tree may be allocated beyond our memory limit, or inside the
 106  * crash kernel region for kdump, or within the page aligned range of initrd.
 107  * If so, move it out of the way.
 108  */
 109 static void __init move_device_tree(void)
 110 {
 111         unsigned long start, size;
 112         void *p;
 113
 114         DBG("-> move_device_tree\n");
 115
 116         start = __pa(initial_boot_params);
 117         size = be32_to_cpu(initial_boot_params->totalsize);
 118
 119         if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
 120                         overlaps_crashkernel(start, size) ||
 121                         overlaps_initrd(start, size)) {
 122                 p = __va(memblock_alloc(size, PAGE_SIZE));
 123                 memcpy(p, initial_boot_params, size);
 124                 initial_boot_params = (struct boot_param_header *)p;
 125                 DBG("Moved device tree to 0x%p\n", p);
 126         }
 127
 128         DBG("<- move_device_tree\n");
 129 }
 130
 131 /*
 132  * ibm,pa-features is a per-cpu property that contains a string of
 133  * attribute descriptors, each of which has a 2 byte header plus up
 134  * to 254 bytes worth of processor attribute bits.  First header
 135  * byte specifies the number of bytes following the header.
 136  * Second header byte is an "attribute-specifier" type, of which
 137  * zero is the only currently-defined value.
 138  * Implementation:  Pass in the byte and bit offset for the feature
 139  * that we are interested in.  The function will return -1 if the
 140  * pa-features property is missing, or a 1/0 to indicate if the feature
 141  * is supported/not supported.  Note that the bit numbers are
 142  * big-endian to match the definition in PAPR.
 143  */
 144 static struct ibm_pa_feature {
 145         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 146         unsigned long   mmu_features;   /* MMU_FTR_xxx bit */
 147         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 148         unsigned char   pabyte;         /* byte number in ibm,pa-features */
 149         unsigned char   pabit;          /* bit number (big-endian) */
 150         unsigned char   invert;         /* if 1, pa bit set => clear feature */
 151 } ibm_pa_features[] __initdata = {
 152         {0, 0, PPC_FEATURE_HAS_MMU,     0, 0, 0},
 153         {0, 0, PPC_FEATURE_HAS_FPU,     0, 1, 0},
 154         {0, MMU_FTR_SLB, 0,             0, 2, 0},
 155         {CPU_FTR_CTRL, 0, 0,            0, 3, 0},
 156         {CPU_FTR_NOEXECUTE, 0, 0,       0, 6, 0},
 157         {CPU_FTR_NODSISRALIGN, 0, 0,    1, 1, 1},
 158         {0, MMU_FTR_CI_LARGE_PAGE, 0,   1, 2, 0},
 159         {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
 160 };
 161
 162 static void __init scan_features(unsigned long node, unsigned char *ftrs,
 163                                  unsigned long tablelen,
 164                                  struct ibm_pa_feature *fp,
 165                                  unsigned long ft_size)
 166 {
 167         unsigned long i, len, bit;
 168
 169         /* find descriptor with type == 0 */
 170         for (;;) {
 171                 if (tablelen < 3)
 172                         return;
 173                 len = 2 + ftrs[0];
 174                 if (tablelen < len)
 175                         return;         /* descriptor 0 not found */
 176                 if (ftrs[1] == 0)
 177                         break;
 178                 tablelen -= len;
 179                 ftrs += len;
 180         }
 181
 182         /* loop over bits we know about */
 183         for (i = 0; i < ft_size; ++i, ++fp) {
 184                 if (fp->pabyte >= ftrs[0])
 185                         continue;
 186                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 187                 if (bit ^ fp->invert) {
 188                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 189                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 190                         cur_cpu_spec->mmu_features |= fp->mmu_features;
 191                 } else {
 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                 }
 196         }
 197 }
 198
 199 static void __init check_cpu_pa_features(unsigned long node)
 200 {
 201         unsigned char *pa_ftrs;
 202         unsigned long tablelen;
 203
 204         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
 205         if (pa_ftrs == NULL)
 206                 return;
 207
 208         scan_features(node, pa_ftrs, tablelen,
 209                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 210 }
 211
 212 #ifdef CONFIG_PPC_STD_MMU_64
 213 static void __init check_cpu_slb_size(unsigned long node)
 214 {
 215         u32 *slb_size_ptr;
 216
 217         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
 218         if (slb_size_ptr != NULL) {
 219                 mmu_slb_size = *slb_size_ptr;
 220                 return;
 221         }
 222         slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 223         if (slb_size_ptr != NULL) {
 224                 mmu_slb_size = *slb_size_ptr;
 225         }
 226 }
 227 #else
 228 #define check_cpu_slb_size(node) do { } while(0)
 229 #endif
 230
 231 static struct feature_property {
 232         const char *name;
 233         u32 min_value;
 234         unsigned long cpu_feature;
 235         unsigned long cpu_user_ftr;
 236 } feature_properties[] __initdata = {
 237 #ifdef CONFIG_ALTIVEC
 238         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 239         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 240 #endif /* CONFIG_ALTIVEC */
 241 #ifdef CONFIG_VSX
 242         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 243         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 244 #endif /* CONFIG_VSX */
 245 #ifdef CONFIG_PPC64
 246         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 247         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 248         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 249 #endif /* CONFIG_PPC64 */
 250 };
 251
 252 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 253 static inline void identical_pvr_fixup(unsigned long node)
 254 {
 255         unsigned int pvr;
 256         char *model = of_get_flat_dt_prop(node, "model", NULL);
 257
 258         /*
 259          * Since 440GR(x)/440EP(x) processors have the same pvr,
 260          * we check the node path and set bit 28 in the cur_cpu_spec
 261          * pvr for EP(x) processor version. This bit is always 0 in
 262          * the "real" pvr. Then we call identify_cpu again with
 263          * the new logical pvr to enable FPU support.
 264          */
 265         if (model && strstr(model, "440EP")) {
 266                 pvr = cur_cpu_spec->pvr_value | 0x8;
 267                 identify_cpu(0, pvr);
 268                 DBG("Using logical pvr %x for %s\n", pvr, model);
 269         }
 270 }
 271 #else
 272 #define identical_pvr_fixup(node) do { } while(0)
 273 #endif
 274
 275 static void __init check_cpu_feature_properties(unsigned long node)
 276 {
 277         unsigned long i;
 278         struct feature_property *fp = feature_properties;
 279         const u32 *prop;
 280
 281         for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
 282                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 283                 if (prop && *prop >= fp->min_value) {
 284                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 285                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 286                 }
 287         }
 288 }
 289
 290 static int __init early_init_dt_scan_cpus(unsigned long node,
 291                                           const char *uname, int depth,
 292                                           void *data)
 293 {
 294         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 295         const u32 *prop;
 296         const u32 *intserv;
 297         int i, nthreads;
 298         unsigned long len;
 299         int found = -1;
 300         int found_thread = 0;
 301
 302         /* We are scanning "cpu" nodes only */
 303         if (type == NULL || strcmp(type, "cpu") != 0)
 304                 return 0;
 305
 306         /* Get physical cpuid */
 307         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
 308         if (intserv) {
 309                 nthreads = len / sizeof(int);
 310         } else {
 311                 intserv = of_get_flat_dt_prop(node, "reg", NULL);
 312                 nthreads = 1;
 313         }
 314
 315         /*
 316          * Now see if any of these threads match our boot cpu.
 317          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
 318          */
 319         for (i = 0; i < nthreads; i++) {
 320                 /*
 321                  * version 2 of the kexec param format adds the phys cpuid of
 322                  * booted proc.
 323                  */
 324                 if (initial_boot_params->version >= 2) {
 325                         if (intserv[i] == initial_boot_params->boot_cpuid_phys) {
 326                                 found = boot_cpu_count;
 327                                 found_thread = i;
 328                         }
 329                 } else {
 330                         /*
 331                          * Check if it's the boot-cpu, set it's hw index now,
 332                          * unfortunately this format did not support booting
 333                          * off secondary threads.
 334                          */
 335                         if (of_get_flat_dt_prop(node,
 336                                         "linux,boot-cpu", NULL) != NULL)
 337                                 found = boot_cpu_count;
 338                 }
 339 #ifdef CONFIG_SMP
 340                 /* logical cpu id is always 0 on UP kernels */
 341                 boot_cpu_count++;
 342 #endif
 343         }
 344
 345         if (found >= 0) {
 346                 DBG("boot cpu: logical %d physical %d\n", found,
 347                         intserv[found_thread]);
 348                 boot_cpuid = found;
 349                 set_hard_smp_processor_id(found, intserv[found_thread]);
 350
 351                 /*
 352                  * PAPR defines "logical" PVR values for cpus that
 353                  * meet various levels of the architecture:
 354                  * 0x0f000001   Architecture version 2.04
 355                  * 0x0f000002   Architecture version 2.05
 356                  * If the cpu-version property in the cpu node contains
 357                  * such a value, we call identify_cpu again with the
 358                  * logical PVR value in order to use the cpu feature
 359                  * bits appropriate for the architecture level.
 360                  *
 361                  * A POWER6 partition in "POWER6 architected" mode
 362                  * uses the 0x0f000002 PVR value; in POWER5+ mode
 363                  * it uses 0x0f000001.
 364                  */
 365                 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
 366                 if (prop && (*prop & 0xff000000) == 0x0f000000)
 367                         identify_cpu(0, *prop);
 368
 369                 identical_pvr_fixup(node);
 370         }
 371
 372         check_cpu_feature_properties(node);
 373         check_cpu_pa_features(node);
 374         check_cpu_slb_size(node);
 375
 376 #ifdef CONFIG_PPC_PSERIES
 377         if (nthreads > 1)
 378                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
 379         else
 380                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
 381 #endif
 382
 383         return 0;
 384 }
 385
 386 int __init early_init_dt_scan_chosen_ppc(unsigned long node, const char *uname,
 387                                          int depth, void *data)
 388 {
 389         unsigned long *lprop;
 390
 391         /* Use common scan routine to determine if this is the chosen node */
 392         if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
 393                 return 0;
 394
 395 #ifdef CONFIG_PPC64
 396         /* check if iommu is forced on or off */
 397         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
 398                 iommu_is_off = 1;
 399         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
 400                 iommu_force_on = 1;
 401 #endif
 402
 403         /* mem=x on the command line is the preferred mechanism */
 404         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
 405         if (lprop)
 406                 memory_limit = *lprop;
 407
 408 #ifdef CONFIG_PPC64
 409         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
 410         if (lprop)
 411                 tce_alloc_start = *lprop;
 412         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
 413         if (lprop)
 414                 tce_alloc_end = *lprop;
 415 #endif
 416
 417 #ifdef CONFIG_KEXEC
 418         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
 419         if (lprop)
 420                 crashk_res.start = *lprop;
 421
 422         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
 423         if (lprop)
 424                 crashk_res.end = crashk_res.start + *lprop - 1;
 425 #endif
 426
 427         /* break now */
 428         return 1;
 429 }
 430
 431 #ifdef CONFIG_PPC_PSERIES
 432 /*
 433  * Interpret the ibm,dynamic-memory property in the
 434  * /ibm,dynamic-reconfiguration-memory node.
 435  * This contains a list of memory blocks along with NUMA affinity
 436  * information.
 437  */
 438 static int __init early_init_dt_scan_drconf_memory(unsigned long node)
 439 {
 440         __be32 *dm, *ls, *usm;
 441         unsigned long l, n, flags;
 442         u64 base, size, memblock_size;
 443         unsigned int is_kexec_kdump = 0, rngs;
 444
 445         ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
 446         if (ls == NULL || l < dt_root_size_cells * sizeof(__be32))
 447                 return 0;
 448         memblock_size = dt_mem_next_cell(dt_root_size_cells, &ls);
 449
 450         dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
 451         if (dm == NULL || l < sizeof(__be32))
 452                 return 0;
 453
 454         n = *dm++;      /* number of entries */
 455         if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(__be32))
 456                 return 0;
 457
 458         /* check if this is a kexec/kdump kernel. */
 459         usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
 460                                                  &l);
 461         if (usm != NULL)
 462                 is_kexec_kdump = 1;
 463
 464         for (; n != 0; --n) {
 465                 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
 466                 flags = dm[3];
 467                 /* skip DRC index, pad, assoc. list index, flags */
 468                 dm += 4;
 469                 /* skip this block if the reserved bit is set in flags (0x80)
 470                    or if the block is not assigned to this partition (0x8) */
 471                 if ((flags & 0x80) || !(flags & 0x8))
 472                         continue;
 473                 size = memblock_size;
 474                 rngs = 1;
 475                 if (is_kexec_kdump) {
 476                         /*
 477                          * For each memblock in ibm,dynamic-memory, a corresponding
 478                          * entry in linux,drconf-usable-memory property contains
 479                          * a counter 'p' followed by 'p' (base, size) duple.
 480                          * Now read the counter from
 481                          * linux,drconf-usable-memory property
 482                          */
 483                         rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
 484                         if (!rngs) /* there are no (base, size) duple */
 485                                 continue;
 486                 }
 487                 do {
 488                         if (is_kexec_kdump) {
 489                                 base = dt_mem_next_cell(dt_root_addr_cells,
 490                                                          &usm);
 491                                 size = dt_mem_next_cell(dt_root_size_cells,
 492                                                          &usm);
 493                         }
 494                         if (iommu_is_off) {
 495                                 if (base >= 0x80000000ul)
 496                                         continue;
 497                                 if ((base + size) > 0x80000000ul)
 498                                         size = 0x80000000ul - base;
 499                         }
 500                         memblock_add(base, size);
 501                 } while (--rngs);
 502         }
 503         memblock_dump_all();
 504         return 0;
 505 }
 506 #else
 507 #define early_init_dt_scan_drconf_memory(node)  0
 508 #endif /* CONFIG_PPC_PSERIES */
 509
 510 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
 511                                                 const char *uname,
 512                                                 int depth, void *data)
 513 {
 514         if (depth == 1 &&
 515             strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
 516                 return early_init_dt_scan_drconf_memory(node);
 517
 518         return early_init_dt_scan_memory(node, uname, depth, data);
 519 }
 520
 521 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
 522 {
 523 #ifdef CONFIG_PPC64
 524         if (iommu_is_off) {
 525                 if (base >= 0x80000000ul)
 526                         return;
 527                 if ((base + size) > 0x80000000ul)
 528                         size = 0x80000000ul - base;
 529         }
 530 #endif
 531         /* Keep track of the beginning of memory -and- the size of
 532          * the very first block in the device-tree as it represents
 533          * the RMA on ppc64 server
 534          */
 535         if (base < memstart_addr) {
 536                 memstart_addr = base;
 537                 first_memblock_size = size;
 538         }
 539
 540         /* Add the chunk to the MEMBLOCK list */
 541         memblock_add(base, size);
 542 }
 543
 544 void * __init early_init_dt_alloc_memory_arch(u64 size, u64 align)
 545 {
 546         return __va(memblock_alloc(size, align));
 547 }
 548
 549 #ifdef CONFIG_BLK_DEV_INITRD
 550 void __init early_init_dt_setup_initrd_arch(unsigned long start,
 551                 unsigned long end)
 552 {
 553         initrd_start = (unsigned long)__va(start);
 554         initrd_end = (unsigned long)__va(end);
 555         initrd_below_start_ok = 1;
 556 }
 557 #endif
 558
 559 static void __init early_reserve_mem(void)
 560 {
 561         u64 base, size;
 562         u64 *reserve_map;
 563         unsigned long self_base;
 564         unsigned long self_size;
 565
 566         reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
 567                                         initial_boot_params->off_mem_rsvmap);
 568
 569         /* before we do anything, lets reserve the dt blob */
 570         self_base = __pa((unsigned long)initial_boot_params);
 571         self_size = initial_boot_params->totalsize;
 572         memblock_reserve(self_base, self_size);
 573
 574 #ifdef CONFIG_BLK_DEV_INITRD
 575         /* then reserve the initrd, if any */
 576         if (initrd_start && (initrd_end > initrd_start))
 577                 memblock_reserve(_ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
 578                         _ALIGN_UP(initrd_end, PAGE_SIZE) -
 579                         _ALIGN_DOWN(initrd_start, PAGE_SIZE));
 580 #endif /* CONFIG_BLK_DEV_INITRD */
 581
 582 #ifdef CONFIG_PPC32
 583         /*
 584          * Handle the case where we might be booting from an old kexec
 585          * image that setup the mem_rsvmap as pairs of 32-bit values
 586          */
 587         if (*reserve_map > 0xffffffffull) {
 588                 u32 base_32, size_32;
 589                 u32 *reserve_map_32 = (u32 *)reserve_map;
 590
 591                 while (1) {
 592                         base_32 = *(reserve_map_32++);
 593                         size_32 = *(reserve_map_32++);
 594                         if (size_32 == 0)
 595                                 break;
 596                         /* skip if the reservation is for the blob */
 597                         if (base_32 == self_base && size_32 == self_size)
 598                                 continue;
 599                         DBG("reserving: %x -> %x\n", base_32, size_32);
 600                         memblock_reserve(base_32, size_32);
 601                 }
 602                 return;
 603         }
 604 #endif
 605         while (1) {
 606                 base = *(reserve_map++);
 607                 size = *(reserve_map++);
 608                 if (size == 0)
 609                         break;
 610                 DBG("reserving: %llx -> %llx\n", base, size);
 611                 memblock_reserve(base, size);
 612         }
 613 }
 614
 615 #ifdef CONFIG_PHYP_DUMP
 616 /**
 617  * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
 618  *
 619  * Function to find the largest size we need to reserve
 620  * during early boot process.
 621  *
 622  * It either looks for boot param and returns that OR
 623  * returns larger of 256 or 5% rounded down to multiples of 256MB.
 624  *
 625  */
 626 static inline unsigned long phyp_dump_calculate_reserve_size(void)
 627 {
 628         unsigned long tmp;
 629
 630         if (phyp_dump_info->reserve_bootvar)
 631                 return phyp_dump_info->reserve_bootvar;
 632
 633         /* divide by 20 to get 5% of value */
 634         tmp = memblock_end_of_DRAM();
 635         do_div(tmp, 20);
 636
 637         /* round it down in multiples of 256 */
 638         tmp = tmp & ~0x0FFFFFFFUL;
 639
 640         return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
 641 }
 642
 643 /**
 644  * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
 645  *
 646  * This routine may reserve memory regions in the kernel only
 647  * if the system is supported and a dump was taken in last
 648  * boot instance or if the hardware is supported and the
 649  * scratch area needs to be setup. In other instances it returns
 650  * without reserving anything. The memory in case of dump being
 651  * active is freed when the dump is collected (by userland tools).
 652  */
 653 static void __init phyp_dump_reserve_mem(void)
 654 {
 655         unsigned long base, size;
 656         unsigned long variable_reserve_size;
 657
 658         if (!phyp_dump_info->phyp_dump_configured) {
 659                 printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
 660                 return;
 661         }
 662
 663         if (!phyp_dump_info->phyp_dump_at_boot) {
 664                 printk(KERN_INFO "Phyp-dump disabled at boot time\n");
 665                 return;
 666         }
 667
 668         variable_reserve_size = phyp_dump_calculate_reserve_size();
 669
 670         if (phyp_dump_info->phyp_dump_is_active) {
 671                 /* Reserve *everything* above RMR.Area freed by userland tools*/
 672                 base = variable_reserve_size;
 673                 size = memblock_end_of_DRAM() - base;
 674
 675                 /* XXX crashed_ram_end is wrong, since it may be beyond
 676                  * the memory_limit, it will need to be adjusted. */
 677                 memblock_reserve(base, size);
 678
 679                 phyp_dump_info->init_reserve_start = base;
 680                 phyp_dump_info->init_reserve_size = size;
 681         } else {
 682                 size = phyp_dump_info->cpu_state_size +
 683                         phyp_dump_info->hpte_region_size +
 684                         variable_reserve_size;
 685                 base = memblock_end_of_DRAM() - size;
 686                 memblock_reserve(base, size);
 687                 phyp_dump_info->init_reserve_start = base;
 688                 phyp_dump_info->init_reserve_size = size;
 689         }
 690 }
 691 #else
 692 static inline void __init phyp_dump_reserve_mem(void) {}
 693 #endif /* CONFIG_PHYP_DUMP  && CONFIG_PPC_RTAS */
 694
 695 void __init early_init_devtree(void *params)
 696 {
 697         phys_addr_t limit;
 698
 699         DBG(" -> early_init_devtree(%p)\n", params);
 700
 701         /* Setup flat device-tree pointer */
 702         initial_boot_params = params;
 703
 704 #ifdef CONFIG_PPC_RTAS
 705         /* Some machines might need RTAS info for debugging, grab it now. */
 706         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 707 #endif
 708
 709 #ifdef CONFIG_PHYP_DUMP
 710         /* scan tree to see if dump occurred during last boot */
 711         of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
 712 #endif
 713
 714         /* Retrieve various informations from the /chosen node of the
 715          * device-tree, including the platform type, initrd location and
 716          * size, TCE reserve, and more ...
 717          */
 718         of_scan_flat_dt(early_init_dt_scan_chosen_ppc, cmd_line);
 719
 720         /* Scan memory nodes and rebuild MEMBLOCKs */
 721         memblock_init();
 722
 723         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 724         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 725         setup_initial_memory_limit(memstart_addr, first_memblock_size);
 726
 727         /* Save command line for /proc/cmdline and then parse parameters */
 728         strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
 729         parse_early_param();
 730
 731         /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
 732         memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
 733         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 734         if (PHYSICAL_START > MEMORY_START)
 735                 memblock_reserve(MEMORY_START, 0x8000);
 736         reserve_kdump_trampoline();
 737         reserve_crashkernel();
 738         early_reserve_mem();
 739         phyp_dump_reserve_mem();
 740
 741         limit = memory_limit;
 742         if (! limit) {
 743                 phys_addr_t memsize;
 744
 745                 /* Ensure that total memory size is page-aligned, because
 746                  * otherwise mark_bootmem() gets upset. */
 747                 memblock_analyze();
 748                 memsize = memblock_phys_mem_size();
 749                 if ((memsize & PAGE_MASK) != memsize)
 750                         limit = memsize & PAGE_MASK;
 751         }
 752         memblock_enforce_memory_limit(limit);
 753
 754         memblock_analyze();
 755         memblock_dump_all();
 756
 757         DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
 758
 759         /* We may need to relocate the flat tree, do it now.
 760          * FIXME .. and the initrd too? */
 761         move_device_tree();
 762
 763         allocate_pacas();
 764
 765         DBG("Scanning CPUs ...\n");
 766
 767         /* Retrieve CPU related informations from the flat tree
 768          * (altivec support, boot CPU ID, ...)
 769          */
 770         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 771
 772         DBG(" <- early_init_devtree()\n");
 773 }
 774
 775 /*******
 776  *
 777  * New implementation of the OF "find" APIs, return a refcounted
 778  * object, call of_node_put() when done.  The device tree and list
 779  * are protected by a rw_lock.
 780  *
 781  * Note that property management will need some locking as well,
 782  * this isn't dealt with yet.
 783  *
 784  *******/
 785
 786 /**
 787  *      of_find_next_cache_node - Find a node's subsidiary cache
 788  *      @np:    node of type "cpu" or "cache"
 789  *
 790  *      Returns a node pointer with refcount incremented, use
 791  *      of_node_put() on it when done.  Caller should hold a reference
 792  *      to np.
 793  */
 794 struct device_node *of_find_next_cache_node(struct device_node *np)
 795 {
 796         struct device_node *child;
 797         const phandle *handle;
 798
 799         handle = of_get_property(np, "l2-cache", NULL);
 800         if (!handle)
 801                 handle = of_get_property(np, "next-level-cache", NULL);
 802
 803         if (handle)
 804                 return of_find_node_by_phandle(*handle);
 805
 806         /* OF on pmac has nodes instead of properties named "l2-cache"
 807          * beneath CPU nodes.
 808          */
 809         if (!strcmp(np->type, "cpu"))
 810                 for_each_child_of_node(np, child)
 811                         if (!strcmp(child->type, "cache"))
 812                                 return child;
 813
 814         return NULL;
 815 }
 816
 817 #ifdef CONFIG_PPC_PSERIES
 818 /*
 819  * Fix up the uninitialized fields in a new device node:
 820  * name, type and pci-specific fields
 821  */
 822
 823 static int of_finish_dynamic_node(struct device_node *node)
 824 {
 825         struct device_node *parent = of_get_parent(node);
 826         int err = 0;
 827         const phandle *ibm_phandle;
 828
 829         node->name = of_get_property(node, "name", NULL);
 830         node->type = of_get_property(node, "device_type", NULL);
 831
 832         if (!node->name)
 833                 node->name = "<NULL>";
 834         if (!node->type)
 835                 node->type = "<NULL>";
 836
 837         if (!parent) {
 838                 err = -ENODEV;
 839                 goto out;
 840         }
 841
 842         /* We don't support that function on PowerMac, at least
 843          * not yet
 844          */
 845         if (machine_is(powermac))
 846                 return -ENODEV;
 847
 848         /* fix up new node's phandle field */
 849         if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
 850                 node->phandle = *ibm_phandle;
 851
 852 out:
 853         of_node_put(parent);
 854         return err;
 855 }
 856
 857 static int prom_reconfig_notifier(struct notifier_block *nb,
 858                                   unsigned long action, void *node)
 859 {
 860         int err;
 861
 862         switch (action) {
 863         case PSERIES_RECONFIG_ADD:
 864                 err = of_finish_dynamic_node(node);
 865                 if (err < 0) {
 866                         printk(KERN_ERR "finish_node returned %d\n", err);
 867                         err = NOTIFY_BAD;
 868                 }
 869                 break;
 870         default:
 871                 err = NOTIFY_DONE;
 872                 break;
 873         }
 874         return err;
 875 }
 876
 877 static struct notifier_block prom_reconfig_nb = {
 878         .notifier_call = prom_reconfig_notifier,
 879         .priority = 10, /* This one needs to run first */
 880 };
 881
 882 static int __init prom_reconfig_setup(void)
 883 {
 884         return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
 885 }
 886 __initcall(prom_reconfig_setup);
 887 #endif
 888
 889 /* Find the device node for a given logical cpu number, also returns the cpu
 890  * local thread number (index in ibm,interrupt-server#s) if relevant and
 891  * asked for (non NULL)
 892  */
 893 struct device_node *of_get_cpu_node(int cpu, unsigned int *thread)
 894 {
 895         int hardid;
 896         struct device_node *np;
 897
 898         hardid = get_hard_smp_processor_id(cpu);
 899
 900         for_each_node_by_type(np, "cpu") {
 901                 const u32 *intserv;
 902                 unsigned int plen, t;
 903
 904                 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
 905                  * fallback to "reg" property and assume no threads
 906                  */
 907                 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
 908                                 &plen);
 909                 if (intserv == NULL) {
 910                         const u32 *reg = of_get_property(np, "reg", NULL);
 911                         if (reg == NULL)
 912                                 continue;
 913                         if (*reg == hardid) {
 914                                 if (thread)
 915                                         *thread = 0;
 916                                 return np;
 917                         }
 918                 } else {
 919                         plen /= sizeof(u32);
 920                         for (t = 0; t < plen; t++) {
 921                                 if (hardid == intserv[t]) {
 922                                         if (thread)
 923                                                 *thread = t;
 924                                         return np;
 925                                 }
 926                         }
 927                 }
 928         }
 929         return NULL;
 930 }
 931 EXPORT_SYMBOL(of_get_cpu_node);
 932
 933 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
 934 static struct debugfs_blob_wrapper flat_dt_blob;
 935
 936 static int __init export_flat_device_tree(void)
 937 {
 938         struct dentry *d;
 939
 940         flat_dt_blob.data = initial_boot_params;
 941         flat_dt_blob.size = initial_boot_params->totalsize;
 942
 943         d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
 944                                 powerpc_debugfs_root, &flat_dt_blob);
 945         if (!d)
 946                 return 1;
 947
 948         return 0;
 949 }
 950 __initcall(export_flat_device_tree);
 951 #endif