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/lmb.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/pgtable.h>
  47 #include <asm/pci.h>
  48 #include <asm/iommu.h>
  49 #include <asm/btext.h>
  50 #include <asm/sections.h>
  51 #include <asm/machdep.h>
  52 #include <asm/pSeries_reconfig.h>
  53 #include <asm/pci-bridge.h>
  54 #include <asm/phyp_dump.h>
  55 #include <asm/kexec.h>
  56 #include <mm/mmu_decl.h>
  57
  58 #ifdef DEBUG
  59 #define DBG(fmt...) printk(KERN_ERR fmt)
  60 #else
  61 #define DBG(fmt...)
  62 #endif
  63
  64 #ifdef CONFIG_PPC64
  65 int __initdata iommu_is_off;
  66 int __initdata iommu_force_on;
  67 unsigned long tce_alloc_start, tce_alloc_end;
  68 #endif
  69
  70 extern rwlock_t devtree_lock;   /* temporary while merging */
  71
  72 /* export that to outside world */
  73 struct device_node *of_chosen;
  74
  75 static int __init early_parse_mem(char *p)
  76 {
  77         if (!p)
  78                 return 1;
  79
  80         memory_limit = PAGE_ALIGN(memparse(p, &p));
  81         DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit);
  82
  83         return 0;
  84 }
  85 early_param("mem", early_parse_mem);
  86
  87 /**
  88  * move_device_tree - move tree to an unused area, if needed.
  89  *
  90  * The device tree may be allocated beyond our memory limit, or inside the
  91  * crash kernel region for kdump. If so, move it out of the way.
  92  */
  93 static void __init move_device_tree(void)
  94 {
  95         unsigned long start, size;
  96         void *p;
  97
  98         DBG("-> move_device_tree\n");
  99
 100         start = __pa(initial_boot_params);
 101         size = initial_boot_params->totalsize;
 102
 103         if ((memory_limit && (start + size) > memory_limit) ||
 104                         overlaps_crashkernel(start, size)) {
 105                 p = __va(lmb_alloc_base(size, PAGE_SIZE, lmb.rmo_size));
 106                 memcpy(p, initial_boot_params, size);
 107                 initial_boot_params = (struct boot_param_header *)p;
 108                 DBG("Moved device tree to 0x%p\n", p);
 109         }
 110
 111         DBG("<- move_device_tree\n");
 112 }
 113
 114 /*
 115  * ibm,pa-features is a per-cpu property that contains a string of
 116  * attribute descriptors, each of which has a 2 byte header plus up
 117  * to 254 bytes worth of processor attribute bits.  First header
 118  * byte specifies the number of bytes following the header.
 119  * Second header byte is an "attribute-specifier" type, of which
 120  * zero is the only currently-defined value.
 121  * Implementation:  Pass in the byte and bit offset for the feature
 122  * that we are interested in.  The function will return -1 if the
 123  * pa-features property is missing, or a 1/0 to indicate if the feature
 124  * is supported/not supported.  Note that the bit numbers are
 125  * big-endian to match the definition in PAPR.
 126  */
 127 static struct ibm_pa_feature {
 128         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 129         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 130         unsigned char   pabyte;         /* byte number in ibm,pa-features */
 131         unsigned char   pabit;          /* bit number (big-endian) */
 132         unsigned char   invert;         /* if 1, pa bit set => clear feature */
 133 } ibm_pa_features[] __initdata = {
 134         {0, PPC_FEATURE_HAS_MMU,        0, 0, 0},
 135         {0, PPC_FEATURE_HAS_FPU,        0, 1, 0},
 136         {CPU_FTR_SLB, 0,                0, 2, 0},
 137         {CPU_FTR_CTRL, 0,               0, 3, 0},
 138         {CPU_FTR_NOEXECUTE, 0,          0, 6, 0},
 139         {CPU_FTR_NODSISRALIGN, 0,       1, 1, 1},
 140         {CPU_FTR_CI_LARGE_PAGE, 0,      1, 2, 0},
 141         {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
 142 };
 143
 144 static void __init scan_features(unsigned long node, unsigned char *ftrs,
 145                                  unsigned long tablelen,
 146                                  struct ibm_pa_feature *fp,
 147                                  unsigned long ft_size)
 148 {
 149         unsigned long i, len, bit;
 150
 151         /* find descriptor with type == 0 */
 152         for (;;) {
 153                 if (tablelen < 3)
 154                         return;
 155                 len = 2 + ftrs[0];
 156                 if (tablelen < len)
 157                         return;         /* descriptor 0 not found */
 158                 if (ftrs[1] == 0)
 159                         break;
 160                 tablelen -= len;
 161                 ftrs += len;
 162         }
 163
 164         /* loop over bits we know about */
 165         for (i = 0; i < ft_size; ++i, ++fp) {
 166                 if (fp->pabyte >= ftrs[0])
 167                         continue;
 168                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 169                 if (bit ^ fp->invert) {
 170                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 171                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 172                 } else {
 173                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
 174                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
 175                 }
 176         }
 177 }
 178
 179 static void __init check_cpu_pa_features(unsigned long node)
 180 {
 181         unsigned char *pa_ftrs;
 182         unsigned long tablelen;
 183
 184         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
 185         if (pa_ftrs == NULL)
 186                 return;
 187
 188         scan_features(node, pa_ftrs, tablelen,
 189                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 190 }
 191
 192 #ifdef CONFIG_PPC_STD_MMU_64
 193 static void __init check_cpu_slb_size(unsigned long node)
 194 {
 195         u32 *slb_size_ptr;
 196
 197         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
 198         if (slb_size_ptr != NULL) {
 199                 mmu_slb_size = *slb_size_ptr;
 200                 return;
 201         }
 202         slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 203         if (slb_size_ptr != NULL) {
 204                 mmu_slb_size = *slb_size_ptr;
 205         }
 206 }
 207 #else
 208 #define check_cpu_slb_size(node) do { } while(0)
 209 #endif
 210
 211 static struct feature_property {
 212         const char *name;
 213         u32 min_value;
 214         unsigned long cpu_feature;
 215         unsigned long cpu_user_ftr;
 216 } feature_properties[] __initdata = {
 217 #ifdef CONFIG_ALTIVEC
 218         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 219         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 220 #endif /* CONFIG_ALTIVEC */
 221 #ifdef CONFIG_VSX
 222         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 223         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 224 #endif /* CONFIG_VSX */
 225 #ifdef CONFIG_PPC64
 226         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 227         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 228         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 229 #endif /* CONFIG_PPC64 */
 230 };
 231
 232 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 233 static inline void identical_pvr_fixup(unsigned long node)
 234 {
 235         unsigned int pvr;
 236         char *model = of_get_flat_dt_prop(node, "model", NULL);
 237
 238         /*
 239          * Since 440GR(x)/440EP(x) processors have the same pvr,
 240          * we check the node path and set bit 28 in the cur_cpu_spec
 241          * pvr for EP(x) processor version. This bit is always 0 in
 242          * the "real" pvr. Then we call identify_cpu again with
 243          * the new logical pvr to enable FPU support.
 244          */
 245         if (model && strstr(model, "440EP")) {
 246                 pvr = cur_cpu_spec->pvr_value | 0x8;
 247                 identify_cpu(0, pvr);
 248                 DBG("Using logical pvr %x for %s\n", pvr, model);
 249         }
 250 }
 251 #else
 252 #define identical_pvr_fixup(node) do { } while(0)
 253 #endif
 254
 255 static void __init check_cpu_feature_properties(unsigned long node)
 256 {
 257         unsigned long i;
 258         struct feature_property *fp = feature_properties;
 259         const u32 *prop;
 260
 261         for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
 262                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 263                 if (prop && *prop >= fp->min_value) {
 264                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 265                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 266                 }
 267         }
 268 }
 269
 270 static int __init early_init_dt_scan_cpus(unsigned long node,
 271                                           const char *uname, int depth,
 272                                           void *data)
 273 {
 274         static int logical_cpuid = 0;
 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 = 0;
 281
 282         /* We are scanning "cpu" nodes only */
 283         if (type == NULL || strcmp(type, "cpu") != 0)
 284                 return 0;
 285
 286         /* Get physical cpuid */
 287         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
 288         if (intserv) {
 289                 nthreads = len / sizeof(int);
 290         } else {
 291                 intserv = of_get_flat_dt_prop(node, "reg", NULL);
 292                 nthreads = 1;
 293         }
 294
 295         /*
 296          * Now see if any of these threads match our boot cpu.
 297          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
 298          */
 299         for (i = 0; i < nthreads; i++) {
 300                 /*
 301                  * version 2 of the kexec param format adds the phys cpuid of
 302                  * booted proc.
 303                  */
 304                 if (initial_boot_params && initial_boot_params->version >= 2) {
 305                         if (intserv[i] ==
 306                                         initial_boot_params->boot_cpuid_phys) {
 307                                 found = 1;
 308                                 break;
 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 = 1;
 319                                 break;
 320                         }
 321                 }
 322
 323 #ifdef CONFIG_SMP
 324                 /* logical cpu id is always 0 on UP kernels */
 325                 logical_cpuid++;
 326 #endif
 327         }
 328
 329         if (found) {
 330                 DBG("boot cpu: logical %d physical %d\n", logical_cpuid,
 331                         intserv[i]);
 332                 boot_cpuid = logical_cpuid;
 333                 set_hard_smp_processor_id(boot_cpuid, intserv[i]);
 334
 335                 /*
 336                  * PAPR defines "logical" PVR values for cpus that
 337                  * meet various levels of the architecture:
 338                  * 0x0f000001   Architecture version 2.04
 339                  * 0x0f000002   Architecture version 2.05
 340                  * If the cpu-version property in the cpu node contains
 341                  * such a value, we call identify_cpu again with the
 342                  * logical PVR value in order to use the cpu feature
 343                  * bits appropriate for the architecture level.
 344                  *
 345                  * A POWER6 partition in "POWER6 architected" mode
 346                  * uses the 0x0f000002 PVR value; in POWER5+ mode
 347                  * it uses 0x0f000001.
 348                  */
 349                 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
 350                 if (prop && (*prop & 0xff000000) == 0x0f000000)
 351                         identify_cpu(0, *prop);
 352
 353                 identical_pvr_fixup(node);
 354         }
 355
 356         check_cpu_feature_properties(node);
 357         check_cpu_pa_features(node);
 358         check_cpu_slb_size(node);
 359
 360 #ifdef CONFIG_PPC_PSERIES
 361         if (nthreads > 1)
 362                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
 363         else
 364                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
 365 #endif
 366
 367         return 0;
 368 }
 369
 370 void __init early_init_dt_scan_chosen_arch(unsigned long node)
 371 {
 372         unsigned long *lprop;
 373
 374 #ifdef CONFIG_PPC64
 375         /* check if iommu is forced on or off */
 376         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
 377                 iommu_is_off = 1;
 378         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
 379                 iommu_force_on = 1;
 380 #endif
 381
 382         /* mem=x on the command line is the preferred mechanism */
 383         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
 384         if (lprop)
 385                 memory_limit = *lprop;
 386
 387 #ifdef CONFIG_PPC64
 388         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
 389         if (lprop)
 390                 tce_alloc_start = *lprop;
 391         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
 392         if (lprop)
 393                 tce_alloc_end = *lprop;
 394 #endif
 395
 396 #ifdef CONFIG_KEXEC
 397         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
 398         if (lprop)
 399                 crashk_res.start = *lprop;
 400
 401         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
 402         if (lprop)
 403                 crashk_res.end = crashk_res.start + *lprop - 1;
 404 #endif
 405 }
 406
 407 #ifdef CONFIG_PPC_PSERIES
 408 /*
 409  * Interpret the ibm,dynamic-memory property in the
 410  * /ibm,dynamic-reconfiguration-memory node.
 411  * This contains a list of memory blocks along with NUMA affinity
 412  * information.
 413  */
 414 static int __init early_init_dt_scan_drconf_memory(unsigned long node)
 415 {
 416         __be32 *dm, *ls, *usm;
 417         unsigned long l, n, flags;
 418         u64 base, size, lmb_size;
 419         unsigned int is_kexec_kdump = 0, rngs;
 420
 421         ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
 422         if (ls == NULL || l < dt_root_size_cells * sizeof(__be32))
 423                 return 0;
 424         lmb_size = dt_mem_next_cell(dt_root_size_cells, &ls);
 425
 426         dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
 427         if (dm == NULL || l < sizeof(__be32))
 428                 return 0;
 429
 430         n = *dm++;      /* number of entries */
 431         if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(__be32))
 432                 return 0;
 433
 434         /* check if this is a kexec/kdump kernel. */
 435         usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
 436                                                  &l);
 437         if (usm != NULL)
 438                 is_kexec_kdump = 1;
 439
 440         for (; n != 0; --n) {
 441                 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
 442                 flags = dm[3];
 443                 /* skip DRC index, pad, assoc. list index, flags */
 444                 dm += 4;
 445                 /* skip this block if the reserved bit is set in flags (0x80)
 446                    or if the block is not assigned to this partition (0x8) */
 447                 if ((flags & 0x80) || !(flags & 0x8))
 448                         continue;
 449                 size = lmb_size;
 450                 rngs = 1;
 451                 if (is_kexec_kdump) {
 452                         /*
 453                          * For each lmb in ibm,dynamic-memory, a corresponding
 454                          * entry in linux,drconf-usable-memory property contains
 455                          * a counter 'p' followed by 'p' (base, size) duple.
 456                          * Now read the counter from
 457                          * linux,drconf-usable-memory property
 458                          */
 459                         rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
 460                         if (!rngs) /* there are no (base, size) duple */
 461                                 continue;
 462                 }
 463                 do {
 464                         if (is_kexec_kdump) {
 465                                 base = dt_mem_next_cell(dt_root_addr_cells,
 466                                                          &usm);
 467                                 size = dt_mem_next_cell(dt_root_size_cells,
 468                                                          &usm);
 469                         }
 470                         if (iommu_is_off) {
 471                                 if (base >= 0x80000000ul)
 472                                         continue;
 473                                 if ((base + size) > 0x80000000ul)
 474                                         size = 0x80000000ul - base;
 475                         }
 476                         lmb_add(base, size);
 477                 } while (--rngs);
 478         }
 479         lmb_dump_all();
 480         return 0;
 481 }
 482 #else
 483 #define early_init_dt_scan_drconf_memory(node)  0
 484 #endif /* CONFIG_PPC_PSERIES */
 485
 486 static int __init early_init_dt_scan_memory(unsigned long node,
 487                                             const char *uname, int depth, void *data)
 488 {
 489         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 490         __be32 *reg, *endp;
 491         unsigned long l;
 492
 493         /* Look for the ibm,dynamic-reconfiguration-memory node */
 494         if (depth == 1 &&
 495             strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
 496                 return early_init_dt_scan_drconf_memory(node);
 497
 498         /* We are scanning "memory" nodes only */
 499         if (type == NULL) {
 500                 /*
 501                  * The longtrail doesn't have a device_type on the
 502                  * /memory node, so look for the node called /memory@0.
 503                  */
 504                 if (depth != 1 || strcmp(uname, "memory@0") != 0)
 505                         return 0;
 506         } else if (strcmp(type, "memory") != 0)
 507                 return 0;
 508
 509         reg = of_get_flat_dt_prop(node, "linux,usable-memory", &l);
 510         if (reg == NULL)
 511                 reg = of_get_flat_dt_prop(node, "reg", &l);
 512         if (reg == NULL)
 513                 return 0;
 514
 515         endp = reg + (l / sizeof(__be32));
 516
 517         DBG("memory scan node %s, reg size %ld, data: %x %x %x %x,\n",
 518             uname, l, reg[0], reg[1], reg[2], reg[3]);
 519
 520         while ((endp - reg) >= (dt_root_addr_cells + dt_root_size_cells)) {
 521                 u64 base, size;
 522
 523                 base = dt_mem_next_cell(dt_root_addr_cells, &reg);
 524                 size = dt_mem_next_cell(dt_root_size_cells, &reg);
 525
 526                 if (size == 0)
 527                         continue;
 528                 DBG(" - %llx ,  %llx\n", (unsigned long long)base,
 529                     (unsigned long long)size);
 530 #ifdef CONFIG_PPC64
 531                 if (iommu_is_off) {
 532                         if (base >= 0x80000000ul)
 533                                 continue;
 534                         if ((base + size) > 0x80000000ul)
 535                                 size = 0x80000000ul - base;
 536                 }
 537 #endif
 538                 lmb_add(base, size);
 539
 540                 memstart_addr = min((u64)memstart_addr, base);
 541         }
 542
 543         return 0;
 544 }
 545
 546 static void __init early_reserve_mem(void)
 547 {
 548         u64 base, size;
 549         u64 *reserve_map;
 550         unsigned long self_base;
 551         unsigned long self_size;
 552
 553         reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
 554                                         initial_boot_params->off_mem_rsvmap);
 555
 556         /* before we do anything, lets reserve the dt blob */
 557         self_base = __pa((unsigned long)initial_boot_params);
 558         self_size = initial_boot_params->totalsize;
 559         lmb_reserve(self_base, self_size);
 560
 561 #ifdef CONFIG_BLK_DEV_INITRD
 562         /* then reserve the initrd, if any */
 563         if (initrd_start && (initrd_end > initrd_start))
 564                 lmb_reserve(__pa(initrd_start), initrd_end - initrd_start);
 565 #endif /* CONFIG_BLK_DEV_INITRD */
 566
 567 #ifdef CONFIG_PPC32
 568         /*
 569          * Handle the case where we might be booting from an old kexec
 570          * image that setup the mem_rsvmap as pairs of 32-bit values
 571          */
 572         if (*reserve_map > 0xffffffffull) {
 573                 u32 base_32, size_32;
 574                 u32 *reserve_map_32 = (u32 *)reserve_map;
 575
 576                 while (1) {
 577                         base_32 = *(reserve_map_32++);
 578                         size_32 = *(reserve_map_32++);
 579                         if (size_32 == 0)
 580                                 break;
 581                         /* skip if the reservation is for the blob */
 582                         if (base_32 == self_base && size_32 == self_size)
 583                                 continue;
 584                         DBG("reserving: %x -> %x\n", base_32, size_32);
 585                         lmb_reserve(base_32, size_32);
 586                 }
 587                 return;
 588         }
 589 #endif
 590         while (1) {
 591                 base = *(reserve_map++);
 592                 size = *(reserve_map++);
 593                 if (size == 0)
 594                         break;
 595                 DBG("reserving: %llx -> %llx\n", base, size);
 596                 lmb_reserve(base, size);
 597         }
 598 }
 599
 600 #ifdef CONFIG_PHYP_DUMP
 601 /**
 602  * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
 603  *
 604  * Function to find the largest size we need to reserve
 605  * during early boot process.
 606  *
 607  * It either looks for boot param and returns that OR
 608  * returns larger of 256 or 5% rounded down to multiples of 256MB.
 609  *
 610  */
 611 static inline unsigned long phyp_dump_calculate_reserve_size(void)
 612 {
 613         unsigned long tmp;
 614
 615         if (phyp_dump_info->reserve_bootvar)
 616                 return phyp_dump_info->reserve_bootvar;
 617
 618         /* divide by 20 to get 5% of value */
 619         tmp = lmb_end_of_DRAM();
 620         do_div(tmp, 20);
 621
 622         /* round it down in multiples of 256 */
 623         tmp = tmp & ~0x0FFFFFFFUL;
 624
 625         return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
 626 }
 627
 628 /**
 629  * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
 630  *
 631  * This routine may reserve memory regions in the kernel only
 632  * if the system is supported and a dump was taken in last
 633  * boot instance or if the hardware is supported and the
 634  * scratch area needs to be setup. In other instances it returns
 635  * without reserving anything. The memory in case of dump being
 636  * active is freed when the dump is collected (by userland tools).
 637  */
 638 static void __init phyp_dump_reserve_mem(void)
 639 {
 640         unsigned long base, size;
 641         unsigned long variable_reserve_size;
 642
 643         if (!phyp_dump_info->phyp_dump_configured) {
 644                 printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
 645                 return;
 646         }
 647
 648         if (!phyp_dump_info->phyp_dump_at_boot) {
 649                 printk(KERN_INFO "Phyp-dump disabled at boot time\n");
 650                 return;
 651         }
 652
 653         variable_reserve_size = phyp_dump_calculate_reserve_size();
 654
 655         if (phyp_dump_info->phyp_dump_is_active) {
 656                 /* Reserve *everything* above RMR.Area freed by userland tools*/
 657                 base = variable_reserve_size;
 658                 size = lmb_end_of_DRAM() - base;
 659
 660                 /* XXX crashed_ram_end is wrong, since it may be beyond
 661                  * the memory_limit, it will need to be adjusted. */
 662                 lmb_reserve(base, size);
 663
 664                 phyp_dump_info->init_reserve_start = base;
 665                 phyp_dump_info->init_reserve_size = size;
 666         } else {
 667                 size = phyp_dump_info->cpu_state_size +
 668                         phyp_dump_info->hpte_region_size +
 669                         variable_reserve_size;
 670                 base = lmb_end_of_DRAM() - size;
 671                 lmb_reserve(base, size);
 672                 phyp_dump_info->init_reserve_start = base;
 673                 phyp_dump_info->init_reserve_size = size;
 674         }
 675 }
 676 #else
 677 static inline void __init phyp_dump_reserve_mem(void) {}
 678 #endif /* CONFIG_PHYP_DUMP  && CONFIG_PPC_RTAS */
 679
 680
 681 void __init early_init_devtree(void *params)
 682 {
 683         phys_addr_t limit;
 684
 685         DBG(" -> early_init_devtree(%p)\n", params);
 686
 687         /* Setup flat device-tree pointer */
 688         initial_boot_params = params;
 689
 690 #ifdef CONFIG_PPC_RTAS
 691         /* Some machines might need RTAS info for debugging, grab it now. */
 692         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 693 #endif
 694
 695 #ifdef CONFIG_PHYP_DUMP
 696         /* scan tree to see if dump occured during last boot */
 697         of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
 698 #endif
 699
 700         /* Retrieve various informations from the /chosen node of the
 701          * device-tree, including the platform type, initrd location and
 702          * size, TCE reserve, and more ...
 703          */
 704         of_scan_flat_dt(early_init_dt_scan_chosen, NULL);
 705
 706         /* Scan memory nodes and rebuild LMBs */
 707         lmb_init();
 708         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 709         of_scan_flat_dt(early_init_dt_scan_memory, NULL);
 710
 711         /* Save command line for /proc/cmdline and then parse parameters */
 712         strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
 713         parse_early_param();
 714
 715         /* Reserve LMB regions used by kernel, initrd, dt, etc... */
 716         lmb_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
 717         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 718         if (PHYSICAL_START > MEMORY_START)
 719                 lmb_reserve(MEMORY_START, 0x8000);
 720         reserve_kdump_trampoline();
 721         reserve_crashkernel();
 722         early_reserve_mem();
 723         phyp_dump_reserve_mem();
 724
 725         limit = memory_limit;
 726         if (! limit) {
 727                 phys_addr_t memsize;
 728
 729                 /* Ensure that total memory size is page-aligned, because
 730                  * otherwise mark_bootmem() gets upset. */
 731                 lmb_analyze();
 732                 memsize = lmb_phys_mem_size();
 733                 if ((memsize & PAGE_MASK) != memsize)
 734                         limit = memsize & PAGE_MASK;
 735         }
 736         lmb_enforce_memory_limit(limit);
 737
 738         lmb_analyze();
 739         lmb_dump_all();
 740
 741         DBG("Phys. mem: %llx\n", lmb_phys_mem_size());
 742
 743         /* We may need to relocate the flat tree, do it now.
 744          * FIXME .. and the initrd too? */
 745         move_device_tree();
 746
 747         DBG("Scanning CPUs ...\n");
 748
 749         /* Retreive CPU related informations from the flat tree
 750          * (altivec support, boot CPU ID, ...)
 751          */
 752         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 753
 754         DBG(" <- early_init_devtree()\n");
 755 }
 756
 757 /*******
 758  *
 759  * New implementation of the OF "find" APIs, return a refcounted
 760  * object, call of_node_put() when done.  The device tree and list
 761  * are protected by a rw_lock.
 762  *
 763  * Note that property management will need some locking as well,
 764  * this isn't dealt with yet.
 765  *
 766  *******/
 767
 768 /**
 769  *      of_find_next_cache_node - Find a node's subsidiary cache
 770  *      @np:    node of type "cpu" or "cache"
 771  *
 772  *      Returns a node pointer with refcount incremented, use
 773  *      of_node_put() on it when done.  Caller should hold a reference
 774  *      to np.
 775  */
 776 struct device_node *of_find_next_cache_node(struct device_node *np)
 777 {
 778         struct device_node *child;
 779         const phandle *handle;
 780
 781         handle = of_get_property(np, "l2-cache", NULL);
 782         if (!handle)
 783                 handle = of_get_property(np, "next-level-cache", NULL);
 784
 785         if (handle)
 786                 return of_find_node_by_phandle(*handle);
 787
 788         /* OF on pmac has nodes instead of properties named "l2-cache"
 789          * beneath CPU nodes.
 790          */
 791         if (!strcmp(np->type, "cpu"))
 792                 for_each_child_of_node(np, child)
 793                         if (!strcmp(child->type, "cache"))
 794                                 return child;
 795
 796         return NULL;
 797 }
 798
 799 #ifdef CONFIG_PPC_PSERIES
 800 /*
 801  * Fix up the uninitialized fields in a new device node:
 802  * name, type and pci-specific fields
 803  */
 804
 805 static int of_finish_dynamic_node(struct device_node *node)
 806 {
 807         struct device_node *parent = of_get_parent(node);
 808         int err = 0;
 809         const phandle *ibm_phandle;
 810
 811         node->name = of_get_property(node, "name", NULL);
 812         node->type = of_get_property(node, "device_type", NULL);
 813
 814         if (!node->name)
 815                 node->name = "<NULL>";
 816         if (!node->type)
 817                 node->type = "<NULL>";
 818
 819         if (!parent) {
 820                 err = -ENODEV;
 821                 goto out;
 822         }
 823
 824         /* We don't support that function on PowerMac, at least
 825          * not yet
 826          */
 827         if (machine_is(powermac))
 828                 return -ENODEV;
 829
 830         /* fix up new node's phandle field */
 831         if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
 832                 node->phandle = *ibm_phandle;
 833
 834 out:
 835         of_node_put(parent);
 836         return err;
 837 }
 838
 839 static int prom_reconfig_notifier(struct notifier_block *nb,
 840                                   unsigned long action, void *node)
 841 {
 842         int err;
 843
 844         switch (action) {
 845         case PSERIES_RECONFIG_ADD:
 846                 err = of_finish_dynamic_node(node);
 847                 if (err < 0) {
 848                         printk(KERN_ERR "finish_node returned %d\n", err);
 849                         err = NOTIFY_BAD;
 850                 }
 851                 break;
 852         default:
 853                 err = NOTIFY_DONE;
 854                 break;
 855         }
 856         return err;
 857 }
 858
 859 static struct notifier_block prom_reconfig_nb = {
 860         .notifier_call = prom_reconfig_notifier,
 861         .priority = 10, /* This one needs to run first */
 862 };
 863
 864 static int __init prom_reconfig_setup(void)
 865 {
 866         return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
 867 }
 868 __initcall(prom_reconfig_setup);
 869 #endif
 870
 871 /* Find the device node for a given logical cpu number, also returns the cpu
 872  * local thread number (index in ibm,interrupt-server#s) if relevant and
 873  * asked for (non NULL)
 874  */
 875 struct device_node *of_get_cpu_node(int cpu, unsigned int *thread)
 876 {
 877         int hardid;
 878         struct device_node *np;
 879
 880         hardid = get_hard_smp_processor_id(cpu);
 881
 882         for_each_node_by_type(np, "cpu") {
 883                 const u32 *intserv;
 884                 unsigned int plen, t;
 885
 886                 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
 887                  * fallback to "reg" property and assume no threads
 888                  */
 889                 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
 890                                 &plen);
 891                 if (intserv == NULL) {
 892                         const u32 *reg = of_get_property(np, "reg", NULL);
 893                         if (reg == NULL)
 894                                 continue;
 895                         if (*reg == hardid) {
 896                                 if (thread)
 897                                         *thread = 0;
 898                                 return np;
 899                         }
 900                 } else {
 901                         plen /= sizeof(u32);
 902                         for (t = 0; t < plen; t++) {
 903                                 if (hardid == intserv[t]) {
 904                                         if (thread)
 905                                                 *thread = t;
 906                                         return np;
 907                                 }
 908                         }
 909                 }
 910         }
 911         return NULL;
 912 }
 913 EXPORT_SYMBOL(of_get_cpu_node);
 914
 915 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
 916 static struct debugfs_blob_wrapper flat_dt_blob;
 917
 918 static int __init export_flat_device_tree(void)
 919 {
 920         struct dentry *d;
 921
 922         flat_dt_blob.data = initial_boot_params;
 923         flat_dt_blob.size = initial_boot_params->totalsize;
 924
 925         d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
 926                                 powerpc_debugfs_root, &flat_dt_blob);
 927         if (!d)
 928                 return 1;
 929
 930         return 0;
 931 }
 932 __initcall(export_flat_device_tree);
 933 #endif