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