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  * move_device_tree - move tree to an unused area, if needed.
  87  *
  88  * The device tree may be allocated beyond our memory limit, or inside the
  89  * crash kernel region for kdump. If so, move it out of the way.
  90  */
  91 static void __init move_device_tree(void)
  92 {
  93         unsigned long start, size;
  94         void *p;
  95
  96         DBG("-> move_device_tree\n");
  97
  98         start = __pa(initial_boot_params);
  99         size = be32_to_cpu(initial_boot_params->totalsize);
 100
 101         if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
 102                         overlaps_crashkernel(start, size)) {
 103                 p = __va(memblock_alloc(size, PAGE_SIZE));
 104                 memcpy(p, initial_boot_params, size);
 105                 initial_boot_params = (struct boot_param_header *)p;
 106                 DBG("Moved device tree to 0x%p\n", p);
 107         }
 108
 109         DBG("<- move_device_tree\n");
 110 }
 111
 112 /*
 113  * ibm,pa-features is a per-cpu property that contains a string of
 114  * attribute descriptors, each of which has a 2 byte header plus up
 115  * to 254 bytes worth of processor attribute bits.  First header
 116  * byte specifies the number of bytes following the header.
 117  * Second header byte is an "attribute-specifier" type, of which
 118  * zero is the only currently-defined value.
 119  * Implementation:  Pass in the byte and bit offset for the feature
 120  * that we are interested in.  The function will return -1 if the
 121  * pa-features property is missing, or a 1/0 to indicate if the feature
 122  * is supported/not supported.  Note that the bit numbers are
 123  * big-endian to match the definition in PAPR.
 124  */
 125 static struct ibm_pa_feature {
 126         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 127         unsigned long   mmu_features;   /* MMU_FTR_xxx bit */
 128         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 129         unsigned char   pabyte;         /* byte number in ibm,pa-features */
 130         unsigned char   pabit;          /* bit number (big-endian) */
 131         unsigned char   invert;         /* if 1, pa bit set => clear feature */
 132 } ibm_pa_features[] __initdata = {
 133         {0, 0, PPC_FEATURE_HAS_MMU,     0, 0, 0},
 134         {0, 0, PPC_FEATURE_HAS_FPU,     0, 1, 0},
 135         {0, MMU_FTR_SLB, 0,             0, 2, 0},
 136         {CPU_FTR_CTRL, 0, 0,            0, 3, 0},
 137         {CPU_FTR_NOEXECUTE, 0, 0,       0, 6, 0},
 138         {CPU_FTR_NODSISRALIGN, 0, 0,    1, 1, 1},
 139         {0, MMU_FTR_CI_LARGE_PAGE, 0,   1, 2, 0},
 140         {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
 141 };
 142
 143 static void __init scan_features(unsigned long node, unsigned char *ftrs,
 144                                  unsigned long tablelen,
 145                                  struct ibm_pa_feature *fp,
 146                                  unsigned long ft_size)
 147 {
 148         unsigned long i, len, bit;
 149
 150         /* find descriptor with type == 0 */
 151         for (;;) {
 152                 if (tablelen < 3)
 153                         return;
 154                 len = 2 + ftrs[0];
 155                 if (tablelen < len)
 156                         return;         /* descriptor 0 not found */
 157                 if (ftrs[1] == 0)
 158                         break;
 159                 tablelen -= len;
 160                 ftrs += len;
 161         }
 162
 163         /* loop over bits we know about */
 164         for (i = 0; i < ft_size; ++i, ++fp) {
 165                 if (fp->pabyte >= ftrs[0])
 166                         continue;
 167                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 168                 if (bit ^ fp->invert) {
 169                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 170                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 171                         cur_cpu_spec->mmu_features |= fp->mmu_features;
 172                 } else {
 173                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
 174                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
 175                         cur_cpu_spec->mmu_features &= ~fp->mmu_features;
 176                 }
 177         }
 178 }
 179
 180 static void __init check_cpu_pa_features(unsigned long node)
 181 {
 182         unsigned char *pa_ftrs;
 183         unsigned long tablelen;
 184
 185         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
 186         if (pa_ftrs == NULL)
 187                 return;
 188
 189         scan_features(node, pa_ftrs, tablelen,
 190                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 191 }
 192
 193 #ifdef CONFIG_PPC_STD_MMU_64
 194 static void __init check_cpu_slb_size(unsigned long node)
 195 {
 196         u32 *slb_size_ptr;
 197
 198         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
 199         if (slb_size_ptr != NULL) {
 200                 mmu_slb_size = *slb_size_ptr;
 201                 return;
 202         }
 203         slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 204         if (slb_size_ptr != NULL) {
 205                 mmu_slb_size = *slb_size_ptr;
 206         }
 207 }
 208 #else
 209 #define check_cpu_slb_size(node) do { } while(0)
 210 #endif
 211
 212 static struct feature_property {
 213         const char *name;
 214         u32 min_value;
 215         unsigned long cpu_feature;
 216         unsigned long cpu_user_ftr;
 217 } feature_properties[] __initdata = {
 218 #ifdef CONFIG_ALTIVEC
 219         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 220         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 221 #endif /* CONFIG_ALTIVEC */
 222 #ifdef CONFIG_VSX
 223         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 224         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 225 #endif /* CONFIG_VSX */
 226 #ifdef CONFIG_PPC64
 227         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 228         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 229         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 230 #endif /* CONFIG_PPC64 */
 231 };
 232
 233 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 234 static inline void identical_pvr_fixup(unsigned long node)
 235 {
 236         unsigned int pvr;
 237         char *model = of_get_flat_dt_prop(node, "model", NULL);
 238
 239         /*
 240          * Since 440GR(x)/440EP(x) processors have the same pvr,
 241          * we check the node path and set bit 28 in the cur_cpu_spec
 242          * pvr for EP(x) processor version. This bit is always 0 in
 243          * the "real" pvr. Then we call identify_cpu again with
 244          * the new logical pvr to enable FPU support.
 245          */
 246         if (model && strstr(model, "440EP")) {
 247                 pvr = cur_cpu_spec->pvr_value | 0x8;
 248                 identify_cpu(0, pvr);
 249                 DBG("Using logical pvr %x for %s\n", pvr, model);
 250         }
 251 }
 252 #else
 253 #define identical_pvr_fixup(node) do { } while(0)
 254 #endif
 255
 256 static void __init check_cpu_feature_properties(unsigned long node)
 257 {
 258         unsigned long i;
 259         struct feature_property *fp = feature_properties;
 260         const u32 *prop;
 261
 262         for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
 263                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 264                 if (prop && *prop >= fp->min_value) {
 265                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 266                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 267                 }
 268         }
 269 }
 270
 271 static int __init early_init_dt_scan_cpus(unsigned long node,
 272                                           const char *uname, int depth,
 273                                           void *data)
 274 {
 275         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 276         const u32 *prop;
 277         const u32 *intserv;
 278         int i, nthreads;
 279         unsigned long len;
 280         int found = -1;
 281         int found_thread = 0;
 282
 283         /* We are scanning "cpu" nodes only */
 284         if (type == NULL || strcmp(type, "cpu") != 0)
 285                 return 0;
 286
 287         /* Get physical cpuid */
 288         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
 289         if (intserv) {
 290                 nthreads = len / sizeof(int);
 291         } else {
 292                 intserv = of_get_flat_dt_prop(node, "reg", NULL);
 293                 nthreads = 1;
 294         }
 295
 296         /*
 297          * Now see if any of these threads match our boot cpu.
 298          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
 299          */
 300         for (i = 0; i < nthreads; i++) {
 301                 /*
 302                  * version 2 of the kexec param format adds the phys cpuid of
 303                  * booted proc.
 304                  */
 305                 if (initial_boot_params->version >= 2) {
 306                         if (intserv[i] == initial_boot_params->boot_cpuid_phys) {
 307                                 found = boot_cpu_count;
 308                                 found_thread = i;
 309                         }
 310                 } else {
 311                         /*
 312                          * Check if it's the boot-cpu, set it's hw index now,
 313                          * unfortunately this format did not support booting
 314                          * off secondary threads.
 315                          */
 316                         if (of_get_flat_dt_prop(node,
 317                                         "linux,boot-cpu", NULL) != NULL)
 318                                 found = boot_cpu_count;
 319                 }
 320 #ifdef CONFIG_SMP
 321                 /* logical cpu id is always 0 on UP kernels */
 322                 boot_cpu_count++;
 323 #endif
 324         }
 325
 326         if (found >= 0) {
 327                 DBG("boot cpu: logical %d physical %d\n", found,
 328                         intserv[found_thread]);
 329                 boot_cpuid = found;
 330                 set_hard_smp_processor_id(found, intserv[found_thread]);
 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         /* Keep track of the beginning of memory -and- the size of
 513          * the very first block in the device-tree as it represents
 514          * the RMA on ppc64 server
 515          */
 516         if (base < memstart_addr) {
 517                 memstart_addr = base;
 518                 first_memblock_size = size;
 519         }
 520
 521         /* Add the chunk to the MEMBLOCK list */
 522         memblock_add(base, size);
 523 }
 524
 525 void * __init early_init_dt_alloc_memory_arch(u64 size, u64 align)
 526 {
 527         return __va(memblock_alloc(size, align));
 528 }
 529
 530 #ifdef CONFIG_BLK_DEV_INITRD
 531 void __init early_init_dt_setup_initrd_arch(unsigned long start,
 532                 unsigned long end)
 533 {
 534         initrd_start = (unsigned long)__va(start);
 535         initrd_end = (unsigned long)__va(end);
 536         initrd_below_start_ok = 1;
 537 }
 538 #endif
 539
 540 static void __init early_reserve_mem(void)
 541 {
 542         u64 base, size;
 543         u64 *reserve_map;
 544         unsigned long self_base;
 545         unsigned long self_size;
 546
 547         reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
 548                                         initial_boot_params->off_mem_rsvmap);
 549
 550         /* before we do anything, lets reserve the dt blob */
 551         self_base = __pa((unsigned long)initial_boot_params);
 552         self_size = initial_boot_params->totalsize;
 553         memblock_reserve(self_base, self_size);
 554
 555 #ifdef CONFIG_BLK_DEV_INITRD
 556         /* then reserve the initrd, if any */
 557         if (initrd_start && (initrd_end > initrd_start))
 558                 memblock_reserve(__pa(initrd_start), initrd_end - initrd_start);
 559 #endif /* CONFIG_BLK_DEV_INITRD */
 560
 561 #ifdef CONFIG_PPC32
 562         /*
 563          * Handle the case where we might be booting from an old kexec
 564          * image that setup the mem_rsvmap as pairs of 32-bit values
 565          */
 566         if (*reserve_map > 0xffffffffull) {
 567                 u32 base_32, size_32;
 568                 u32 *reserve_map_32 = (u32 *)reserve_map;
 569
 570                 while (1) {
 571                         base_32 = *(reserve_map_32++);
 572                         size_32 = *(reserve_map_32++);
 573                         if (size_32 == 0)
 574                                 break;
 575                         /* skip if the reservation is for the blob */
 576                         if (base_32 == self_base && size_32 == self_size)
 577                                 continue;
 578                         DBG("reserving: %x -> %x\n", base_32, size_32);
 579                         memblock_reserve(base_32, size_32);
 580                 }
 581                 return;
 582         }
 583 #endif
 584         while (1) {
 585                 base = *(reserve_map++);
 586                 size = *(reserve_map++);
 587                 if (size == 0)
 588                         break;
 589                 DBG("reserving: %llx -> %llx\n", base, size);
 590                 memblock_reserve(base, size);
 591         }
 592 }
 593
 594 #ifdef CONFIG_PHYP_DUMP
 595 /**
 596  * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
 597  *
 598  * Function to find the largest size we need to reserve
 599  * during early boot process.
 600  *
 601  * It either looks for boot param and returns that OR
 602  * returns larger of 256 or 5% rounded down to multiples of 256MB.
 603  *
 604  */
 605 static inline unsigned long phyp_dump_calculate_reserve_size(void)
 606 {
 607         unsigned long tmp;
 608
 609         if (phyp_dump_info->reserve_bootvar)
 610                 return phyp_dump_info->reserve_bootvar;
 611
 612         /* divide by 20 to get 5% of value */
 613         tmp = memblock_end_of_DRAM();
 614         do_div(tmp, 20);
 615
 616         /* round it down in multiples of 256 */
 617         tmp = tmp & ~0x0FFFFFFFUL;
 618
 619         return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
 620 }
 621
 622 /**
 623  * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
 624  *
 625  * This routine may reserve memory regions in the kernel only
 626  * if the system is supported and a dump was taken in last
 627  * boot instance or if the hardware is supported and the
 628  * scratch area needs to be setup. In other instances it returns
 629  * without reserving anything. The memory in case of dump being
 630  * active is freed when the dump is collected (by userland tools).
 631  */
 632 static void __init phyp_dump_reserve_mem(void)
 633 {
 634         unsigned long base, size;
 635         unsigned long variable_reserve_size;
 636
 637         if (!phyp_dump_info->phyp_dump_configured) {
 638                 printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
 639                 return;
 640         }
 641
 642         if (!phyp_dump_info->phyp_dump_at_boot) {
 643                 printk(KERN_INFO "Phyp-dump disabled at boot time\n");
 644                 return;
 645         }
 646
 647         variable_reserve_size = phyp_dump_calculate_reserve_size();
 648
 649         if (phyp_dump_info->phyp_dump_is_active) {
 650                 /* Reserve *everything* above RMR.Area freed by userland tools*/
 651                 base = variable_reserve_size;
 652                 size = memblock_end_of_DRAM() - base;
 653
 654                 /* XXX crashed_ram_end is wrong, since it may be beyond
 655                  * the memory_limit, it will need to be adjusted. */
 656                 memblock_reserve(base, size);
 657
 658                 phyp_dump_info->init_reserve_start = base;
 659                 phyp_dump_info->init_reserve_size = size;
 660         } else {
 661                 size = phyp_dump_info->cpu_state_size +
 662                         phyp_dump_info->hpte_region_size +
 663                         variable_reserve_size;
 664                 base = memblock_end_of_DRAM() - size;
 665                 memblock_reserve(base, size);
 666                 phyp_dump_info->init_reserve_start = base;
 667                 phyp_dump_info->init_reserve_size = size;
 668         }
 669 }
 670 #else
 671 static inline void __init phyp_dump_reserve_mem(void) {}
 672 #endif /* CONFIG_PHYP_DUMP  && CONFIG_PPC_RTAS */
 673
 674 void __init early_init_devtree(void *params)
 675 {
 676         phys_addr_t limit;
 677
 678         DBG(" -> early_init_devtree(%p)\n", params);
 679
 680         /* Setup flat device-tree pointer */
 681         initial_boot_params = params;
 682
 683 #ifdef CONFIG_PPC_RTAS
 684         /* Some machines might need RTAS info for debugging, grab it now. */
 685         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 686 #endif
 687
 688 #ifdef CONFIG_PHYP_DUMP
 689         /* scan tree to see if dump occurred during last boot */
 690         of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
 691 #endif
 692
 693         /* Retrieve various informations from the /chosen node of the
 694          * device-tree, including the platform type, initrd location and
 695          * size, TCE reserve, and more ...
 696          */
 697         of_scan_flat_dt(early_init_dt_scan_chosen_ppc, NULL);
 698
 699         /* Scan memory nodes and rebuild MEMBLOCKs */
 700         memblock_init();
 701
 702         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 703         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 704         setup_initial_memory_limit(memstart_addr, first_memblock_size);
 705
 706         /* Save command line for /proc/cmdline and then parse parameters */
 707         strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
 708         parse_early_param();
 709
 710         /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
 711         memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
 712         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 713         if (PHYSICAL_START > MEMORY_START)
 714                 memblock_reserve(MEMORY_START, 0x8000);
 715         reserve_kdump_trampoline();
 716         reserve_crashkernel();
 717         early_reserve_mem();
 718         phyp_dump_reserve_mem();
 719
 720         limit = memory_limit;
 721         if (! limit) {
 722                 phys_addr_t memsize;
 723
 724                 /* Ensure that total memory size is page-aligned, because
 725                  * otherwise mark_bootmem() gets upset. */
 726                 memblock_analyze();
 727                 memsize = memblock_phys_mem_size();
 728                 if ((memsize & PAGE_MASK) != memsize)
 729                         limit = memsize & PAGE_MASK;
 730         }
 731         memblock_enforce_memory_limit(limit);
 732
 733         memblock_analyze();
 734         memblock_dump_all();
 735
 736         DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
 737
 738         /* We may need to relocate the flat tree, do it now.
 739          * FIXME .. and the initrd too? */
 740         move_device_tree();
 741
 742         allocate_pacas();
 743
 744         DBG("Scanning CPUs ...\n");
 745
 746         /* Retrieve CPU related informations from the flat tree
 747          * (altivec support, boot CPU ID, ...)
 748          */
 749         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 750
 751         DBG(" <- early_init_devtree()\n");
 752 }
 753
 754 /*******
 755  *
 756  * New implementation of the OF "find" APIs, return a refcounted
 757  * object, call of_node_put() when done.  The device tree and list
 758  * are protected by a rw_lock.
 759  *
 760  * Note that property management will need some locking as well,
 761  * this isn't dealt with yet.
 762  *
 763  *******/
 764
 765 /**
 766  *      of_find_next_cache_node - Find a node's subsidiary cache
 767  *      @np:    node of type "cpu" or "cache"
 768  *
 769  *      Returns a node pointer with refcount incremented, use
 770  *      of_node_put() on it when done.  Caller should hold a reference
 771  *      to np.
 772  */
 773 struct device_node *of_find_next_cache_node(struct device_node *np)
 774 {
 775         struct device_node *child;
 776         const phandle *handle;
 777
 778         handle = of_get_property(np, "l2-cache", NULL);
 779         if (!handle)
 780                 handle = of_get_property(np, "next-level-cache", NULL);
 781
 782         if (handle)
 783                 return of_find_node_by_phandle(*handle);
 784
 785         /* OF on pmac has nodes instead of properties named "l2-cache"
 786          * beneath CPU nodes.
 787          */
 788         if (!strcmp(np->type, "cpu"))
 789                 for_each_child_of_node(np, child)
 790                         if (!strcmp(child->type, "cache"))
 791                                 return child;
 792
 793         return NULL;
 794 }
 795
 796 #ifdef CONFIG_PPC_PSERIES
 797 /*
 798  * Fix up the uninitialized fields in a new device node:
 799  * name, type and pci-specific fields
 800  */
 801
 802 static int of_finish_dynamic_node(struct device_node *node)
 803 {
 804         struct device_node *parent = of_get_parent(node);
 805         int err = 0;
 806         const phandle *ibm_phandle;
 807
 808         node->name = of_get_property(node, "name", NULL);
 809         node->type = of_get_property(node, "device_type", NULL);
 810
 811         if (!node->name)
 812                 node->name = "<NULL>";
 813         if (!node->type)
 814                 node->type = "<NULL>";
 815
 816         if (!parent) {
 817                 err = -ENODEV;
 818                 goto out;
 819         }
 820
 821         /* We don't support that function on PowerMac, at least
 822          * not yet
 823          */
 824         if (machine_is(powermac))
 825                 return -ENODEV;
 826
 827         /* fix up new node's phandle field */
 828         if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
 829                 node->phandle = *ibm_phandle;
 830
 831 out:
 832         of_node_put(parent);
 833         return err;
 834 }
 835
 836 static int prom_reconfig_notifier(struct notifier_block *nb,
 837                                   unsigned long action, void *node)
 838 {
 839         int err;
 840
 841         switch (action) {
 842         case PSERIES_RECONFIG_ADD:
 843                 err = of_finish_dynamic_node(node);
 844                 if (err < 0) {
 845                         printk(KERN_ERR "finish_node returned %d\n", err);
 846                         err = NOTIFY_BAD;
 847                 }
 848                 break;
 849         default:
 850                 err = NOTIFY_DONE;
 851                 break;
 852         }
 853         return err;
 854 }
 855
 856 static struct notifier_block prom_reconfig_nb = {
 857         .notifier_call = prom_reconfig_notifier,
 858         .priority = 10, /* This one needs to run first */
 859 };
 860
 861 static int __init prom_reconfig_setup(void)
 862 {
 863         return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
 864 }
 865 __initcall(prom_reconfig_setup);
 866 #endif
 867
 868 /* Find the device node for a given logical cpu number, also returns the cpu
 869  * local thread number (index in ibm,interrupt-server#s) if relevant and
 870  * asked for (non NULL)
 871  */
 872 struct device_node *of_get_cpu_node(int cpu, unsigned int *thread)
 873 {
 874         int hardid;
 875         struct device_node *np;
 876
 877         hardid = get_hard_smp_processor_id(cpu);
 878
 879         for_each_node_by_type(np, "cpu") {
 880                 const u32 *intserv;
 881                 unsigned int plen, t;
 882
 883                 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
 884                  * fallback to "reg" property and assume no threads
 885                  */
 886                 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
 887                                 &plen);
 888                 if (intserv == NULL) {
 889                         const u32 *reg = of_get_property(np, "reg", NULL);
 890                         if (reg == NULL)
 891                                 continue;
 892                         if (*reg == hardid) {
 893                                 if (thread)
 894                                         *thread = 0;
 895                                 return np;
 896                         }
 897                 } else {
 898                         plen /= sizeof(u32);
 899                         for (t = 0; t < plen; t++) {
 900                                 if (hardid == intserv[t]) {
 901                                         if (thread)
 902                                                 *thread = t;
 903                                         return np;
 904                                 }
 905                         }
 906                 }
 907         }
 908         return NULL;
 909 }
 910 EXPORT_SYMBOL(of_get_cpu_node);
 911
 912 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
 913 static struct debugfs_blob_wrapper flat_dt_blob;
 914
 915 static int __init export_flat_device_tree(void)
 916 {
 917         struct dentry *d;
 918
 919         flat_dt_blob.data = initial_boot_params;
 920         flat_dt_blob.size = initial_boot_params->totalsize;
 921
 922         d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
 923                                 powerpc_debugfs_root, &flat_dt_blob);
 924         if (!d)
 925                 return 1;
 926
 927         return 0;
 928 }
 929 __initcall(export_flat_device_tree);
 930 #endif