arch/powerpc/kernel/prom.c

   1 /*
   2  * Procedures for creating, accessing and interpreting the device tree.
   3  *
   4  * Paul Mackerras       August 1996.
   5  * Copyright (C) 1996-2005 Paul Mackerras.
   6  *
   7  *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
   8  *    {engebret|bergner}@us.ibm.com
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  */
  15
  16 #undef DEBUG
  17
  18 #include <stdarg.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/init.h>
  22 #include <linux/threads.h>
  23 #include <linux/spinlock.h>
  24 #include <linux/types.h>
  25 #include <linux/pci.h>
  26 #include <linux/stringify.h>
  27 #include <linux/delay.h>
  28 #include <linux/initrd.h>
  29 #include <linux/bitops.h>
  30 #include <linux/module.h>
  31 #include <linux/kexec.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/irq.h>
  34 #include <linux/memblock.h>
  35
  36 #include <asm/prom.h>
  37 #include <asm/rtas.h>
  38 #include <asm/page.h>
  39 #include <asm/processor.h>
  40 #include <asm/irq.h>
  41 #include <asm/io.h>
  42 #include <asm/kdump.h>
  43 #include <asm/smp.h>
  44 #include <asm/system.h>
  45 #include <asm/mmu.h>
  46 #include <asm/paca.h>
  47 #include <asm/pgtable.h>
  48 #include <asm/pci.h>
  49 #include <asm/iommu.h>
  50 #include <asm/btext.h>
  51 #include <asm/sections.h>
  52 #include <asm/machdep.h>
  53 #include <asm/pSeries_reconfig.h>
  54 #include <asm/pci-bridge.h>
  55 #include <asm/phyp_dump.h>
  56 #include <asm/kexec.h>
  57 #include <mm/mmu_decl.h>
  58
  59 #ifdef DEBUG
  60 #define DBG(fmt...) printk(KERN_ERR fmt)
  61 #else
  62 #define DBG(fmt...)
  63 #endif
  64
  65 #ifdef CONFIG_PPC64
  66 int __initdata iommu_is_off;
  67 int __initdata iommu_force_on;
  68 unsigned long tce_alloc_start, tce_alloc_end;
  69 u64 ppc64_rma_size;
  70 #endif
  71
  72 static int __init early_parse_mem(char *p)
  73 {
  74         if (!p)
  75                 return 1;
  76
  77         memory_limit = PAGE_ALIGN(memparse(p, &p));
  78         DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit);
  79
  80         return 0;
  81 }
  82 early_param("mem", early_parse_mem);
  83
  84 /**
  85  * move_device_tree - move tree to an unused area, if needed.
  86  *
  87  * The device tree may be allocated beyond our memory limit, or inside the
  88  * crash kernel region for kdump. If so, move it out of the way.
  89  */
  90 static void __init move_device_tree(void)
  91 {
  92         unsigned long start, size;
  93         void *p;
  94
  95         DBG("-> move_device_tree\n");
  96
  97         start = __pa(initial_boot_params);
  98         size = be32_to_cpu(initial_boot_params->totalsize);
  99
 100         if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
 101                         overlaps_crashkernel(start, size)) {
 102                 p = __va(memblock_alloc(size, PAGE_SIZE));
 103                 memcpy(p, initial_boot_params, size);
 104                 initial_boot_params = (struct boot_param_header *)p;
 105                 DBG("Moved device tree to 0x%p\n", p);
 106         }
 107
 108         DBG("<- move_device_tree\n");
 109 }
 110
 111 /*
 112  * ibm,pa-features is a per-cpu property that contains a string of
 113  * attribute descriptors, each of which has a 2 byte header plus up
 114  * to 254 bytes worth of processor attribute bits.  First header
 115  * byte specifies the number of bytes following the header.
 116  * Second header byte is an "attribute-specifier" type, of which
 117  * zero is the only currently-defined value.
 118  * Implementation:  Pass in the byte and bit offset for the feature
 119  * that we are interested in.  The function will return -1 if the
 120  * pa-features property is missing, or a 1/0 to indicate if the feature
 121  * is supported/not supported.  Note that the bit numbers are
 122  * big-endian to match the definition in PAPR.
 123  */
 124 static struct ibm_pa_feature {
 125         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 126         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 127         unsigned char   pabyte;         /* byte number in ibm,pa-features */
 128         unsigned char   pabit;          /* bit number (big-endian) */
 129         unsigned char   invert;         /* if 1, pa bit set => clear feature */
 130 } ibm_pa_features[] __initdata = {
 131         {0, PPC_FEATURE_HAS_MMU,        0, 0, 0},
 132         {0, PPC_FEATURE_HAS_FPU,        0, 1, 0},
 133         {CPU_FTR_SLB, 0,                0, 2, 0},
 134         {CPU_FTR_CTRL, 0,               0, 3, 0},
 135         {CPU_FTR_NOEXECUTE, 0,          0, 6, 0},
 136         {CPU_FTR_NODSISRALIGN, 0,       1, 1, 1},
 137         {CPU_FTR_CI_LARGE_PAGE, 0,      1, 2, 0},
 138         {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
 139 };
 140
 141 static void __init scan_features(unsigned long node, unsigned char *ftrs,
 142                                  unsigned long tablelen,
 143                                  struct ibm_pa_feature *fp,
 144                                  unsigned long ft_size)
 145 {
 146         unsigned long i, len, bit;
 147
 148         /* find descriptor with type == 0 */
 149         for (;;) {
 150                 if (tablelen < 3)
 151                         return;
 152                 len = 2 + ftrs[0];
 153                 if (tablelen < len)
 154                         return;         /* descriptor 0 not found */
 155                 if (ftrs[1] == 0)
 156                         break;
 157                 tablelen -= len;
 158                 ftrs += len;
 159         }
 160
 161         /* loop over bits we know about */
 162         for (i = 0; i < ft_size; ++i, ++fp) {
 163                 if (fp->pabyte >= ftrs[0])
 164                         continue;
 165                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 166                 if (bit ^ fp->invert) {
 167                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 168                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 169                 } else {
 170                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
 171                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
 172                 }
 173         }
 174 }
 175
 176 static void __init check_cpu_pa_features(unsigned long node)
 177 {
 178         unsigned char *pa_ftrs;
 179         unsigned long tablelen;
 180
 181         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
 182         if (pa_ftrs == NULL)
 183                 return;
 184
 185         scan_features(node, pa_ftrs, tablelen,
 186                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 187 }
 188
 189 #ifdef CONFIG_PPC_STD_MMU_64
 190 static void __init check_cpu_slb_size(unsigned long node)
 191 {
 192         u32 *slb_size_ptr;
 193
 194         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
 195         if (slb_size_ptr != NULL) {
 196                 mmu_slb_size = *slb_size_ptr;
 197                 return;
 198         }
 199         slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 200         if (slb_size_ptr != NULL) {
 201                 mmu_slb_size = *slb_size_ptr;
 202         }
 203 }
 204 #else
 205 #define check_cpu_slb_size(node) do { } while(0)
 206 #endif
 207
 208 static struct feature_property {
 209         const char *name;
 210         u32 min_value;
 211         unsigned long cpu_feature;
 212         unsigned long cpu_user_ftr;
 213 } feature_properties[] __initdata = {
 214 #ifdef CONFIG_ALTIVEC
 215         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 216         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 217 #endif /* CONFIG_ALTIVEC */
 218 #ifdef CONFIG_VSX
 219         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 220         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 221 #endif /* CONFIG_VSX */
 222 #ifdef CONFIG_PPC64
 223         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 224         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 225         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 226 #endif /* CONFIG_PPC64 */
 227 };
 228
 229 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 230 static inline void identical_pvr_fixup(unsigned long node)
 231 {
 232         unsigned int pvr;
 233         char *model = of_get_flat_dt_prop(node, "model", NULL);
 234
 235         /*
 236          * Since 440GR(x)/440EP(x) processors have the same pvr,
 237          * we check the node path and set bit 28 in the cur_cpu_spec
 238          * pvr for EP(x) processor version. This bit is always 0 in
 239          * the "real" pvr. Then we call identify_cpu again with
 240          * the new logical pvr to enable FPU support.
 241          */
 242         if (model && strstr(model, "440EP")) {
 243                 pvr = cur_cpu_spec->pvr_value | 0x8;
 244                 identify_cpu(0, pvr);
 245                 DBG("Using logical pvr %x for %s\n", pvr, model);
 246         }
 247 }
 248 #else
 249 #define identical_pvr_fixup(node) do { } while(0)
 250 #endif
 251
 252 static void __init check_cpu_feature_properties(unsigned long node)
 253 {
 254         unsigned long i;
 255         struct feature_property *fp = feature_properties;
 256         const u32 *prop;
 257
 258         for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
 259                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 260                 if (prop && *prop >= fp->min_value) {
 261                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 262                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 263                 }
 264         }
 265 }
 266
 267 static int __init early_init_dt_scan_cpus(unsigned long node,
 268                                           const char *uname, int depth,
 269                                           void *data)
 270 {
 271         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 = -1;
 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] == initial_boot_params->boot_cpuid_phys)
 302                                 found = boot_cpu_count;
 303                 } else {
 304                         /*
 305                          * Check if it's the boot-cpu, set it's hw index now,
 306                          * unfortunately this format did not support booting
 307                          * off secondary threads.
 308                          */
 309                         if (of_get_flat_dt_prop(node,
 310                                         "linux,boot-cpu", NULL) != NULL)
 311                                 found = boot_cpu_count;
 312                 }
 313 #ifdef CONFIG_SMP
 314                 /* logical cpu id is always 0 on UP kernels */
 315                 boot_cpu_count++;
 316 #endif
 317         }
 318
 319         if (found >= 0) {
 320                 DBG("boot cpu: logical %d physical %d\n", found,
 321                         intserv[i]);
 322                 boot_cpuid = found;
 323                 set_hard_smp_processor_id(found, intserv[i]);
 324
 325                 /*
 326                  * PAPR defines "logical" PVR values for cpus that
 327                  * meet various levels of the architecture:
 328                  * 0x0f000001   Architecture version 2.04
 329                  * 0x0f000002   Architecture version 2.05
 330                  * If the cpu-version property in the cpu node contains
 331                  * such a value, we call identify_cpu again with the
 332                  * logical PVR value in order to use the cpu feature
 333                  * bits appropriate for the architecture level.
 334                  *
 335                  * A POWER6 partition in "POWER6 architected" mode
 336                  * uses the 0x0f000002 PVR value; in POWER5+ mode
 337                  * it uses 0x0f000001.
 338                  */
 339                 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
 340                 if (prop && (*prop & 0xff000000) == 0x0f000000)
 341                         identify_cpu(0, *prop);
 342
 343                 identical_pvr_fixup(node);
 344         }
 345
 346         check_cpu_feature_properties(node);
 347         check_cpu_pa_features(node);
 348         check_cpu_slb_size(node);
 349
 350 #ifdef CONFIG_PPC_PSERIES
 351         if (nthreads > 1)
 352                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
 353         else
 354                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
 355 #endif
 356
 357         return 0;
 358 }
 359
 360 int __init early_init_dt_scan_chosen_ppc(unsigned long node, const char *uname,
 361                                          int depth, void *data)
 362 {
 363         unsigned long *lprop;
 364
 365         /* Use common scan routine to determine if this is the chosen node */
 366         if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
 367                 return 0;
 368
 369 #ifdef CONFIG_PPC64
 370         /* check if iommu is forced on or off */
 371         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
 372                 iommu_is_off = 1;
 373         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
 374                 iommu_force_on = 1;
 375 #endif
 376
 377         /* mem=x on the command line is the preferred mechanism */
 378         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
 379         if (lprop)
 380                 memory_limit = *lprop;
 381
 382 #ifdef CONFIG_PPC64
 383         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
 384         if (lprop)
 385                 tce_alloc_start = *lprop;
 386         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
 387         if (lprop)
 388                 tce_alloc_end = *lprop;
 389 #endif
 390
 391 #ifdef CONFIG_KEXEC
 392         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
 393         if (lprop)
 394                 crashk_res.start = *lprop;
 395
 396         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
 397         if (lprop)
 398                 crashk_res.end = crashk_res.start + *lprop - 1;
 399 #endif
 400
 401         /* break now */
 402         return 1;
 403 }
 404
 405 #ifdef CONFIG_PPC_PSERIES
 406 /*
 407  * Interpret the ibm,dynamic-memory property in the
 408  * /ibm,dynamic-reconfiguration-memory node.
 409  * This contains a list of memory blocks along with NUMA affinity
 410  * information.
 411  */
 412 static int __init early_init_dt_scan_drconf_memory(unsigned long node)
 413 {
 414         __be32 *dm, *ls, *usm;
 415         unsigned long l, n, flags;
 416         u64 base, size, memblock_size;
 417         unsigned int is_kexec_kdump = 0, rngs;
 418
 419         ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
 420         if (ls == NULL || l < dt_root_size_cells * sizeof(__be32))
 421                 return 0;
 422         memblock_size = dt_mem_next_cell(dt_root_size_cells, &ls);
 423
 424         dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
 425         if (dm == NULL || l < sizeof(__be32))
 426                 return 0;
 427
 428         n = *dm++;      /* number of entries */
 429         if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(__be32))
 430                 return 0;
 431
 432         /* check if this is a kexec/kdump kernel. */
 433         usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
 434                                                  &l);
 435         if (usm != NULL)
 436                 is_kexec_kdump = 1;
 437
 438         for (; n != 0; --n) {
 439                 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
 440                 flags = dm[3];
 441                 /* skip DRC index, pad, assoc. list index, flags */
 442                 dm += 4;
 443                 /* skip this block if the reserved bit is set in flags (0x80)
 444                    or if the block is not assigned to this partition (0x8) */
 445                 if ((flags & 0x80) || !(flags & 0x8))
 446                         continue;
 447                 size = memblock_size;
 448                 rngs = 1;
 449                 if (is_kexec_kdump) {
 450                         /*
 451                          * For each memblock in ibm,dynamic-memory, a corresponding
 452                          * entry in linux,drconf-usable-memory property contains
 453                          * a counter 'p' followed by 'p' (base, size) duple.
 454                          * Now read the counter from
 455                          * linux,drconf-usable-memory property
 456                          */
 457                         rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
 458                         if (!rngs) /* there are no (base, size) duple */
 459                                 continue;
 460                 }
 461                 do {
 462                         if (is_kexec_kdump) {
 463                                 base = dt_mem_next_cell(dt_root_addr_cells,
 464                                                          &usm);
 465                                 size = dt_mem_next_cell(dt_root_size_cells,
 466                                                          &usm);
 467                         }
 468                         if (iommu_is_off) {
 469                                 if (base >= 0x80000000ul)
 470                                         continue;
 471                                 if ((base + size) > 0x80000000ul)
 472                                         size = 0x80000000ul - base;
 473                         }
 474                         memblock_add(base, size);
 475                 } while (--rngs);
 476         }
 477         memblock_dump_all();
 478         return 0;
 479 }
 480 #else
 481 #define early_init_dt_scan_drconf_memory(node)  0
 482 #endif /* CONFIG_PPC_PSERIES */
 483
 484 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
 485                                                 const char *uname,
 486                                                 int depth, void *data)
 487 {
 488         if (depth == 1 &&
 489             strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
 490                 return early_init_dt_scan_drconf_memory(node);
 491
 492         return early_init_dt_scan_memory(node, uname, depth, data);
 493 }
 494
 495 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
 496 {
 497 #ifdef CONFIG_PPC64
 498         if (iommu_is_off) {
 499                 if (base >= 0x80000000ul)
 500                         return;
 501                 if ((base + size) > 0x80000000ul)
 502                         size = 0x80000000ul - base;
 503         }
 504 #endif
 505
 506         /* First MEMBLOCK added, do some special initializations */
 507         if (memstart_addr == ~(phys_addr_t)0)
 508                 setup_initial_memory_limit(base, size);
 509         memstart_addr = min((u64)memstart_addr, base);
 510
 511         /* Add the chunk to the MEMBLOCK list */
 512         memblock_add(base, size);
 513 }
 514
 515 void * __init early_init_dt_alloc_memory_arch(u64 size, u64 align)
 516 {
 517         return __va(memblock_alloc(size, align));
 518 }
 519
 520 #ifdef CONFIG_BLK_DEV_INITRD
 521 void __init early_init_dt_setup_initrd_arch(unsigned long start,
 522                 unsigned long end)
 523 {
 524         initrd_start = (unsigned long)__va(start);
 525         initrd_end = (unsigned long)__va(end);
 526         initrd_below_start_ok = 1;
 527 }
 528 #endif
 529
 530 static void __init early_reserve_mem(void)
 531 {
 532         u64 base, size;
 533         u64 *reserve_map;
 534         unsigned long self_base;
 535         unsigned long self_size;
 536
 537         reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
 538                                         initial_boot_params->off_mem_rsvmap);
 539
 540         /* before we do anything, lets reserve the dt blob */
 541         self_base = __pa((unsigned long)initial_boot_params);
 542         self_size = initial_boot_params->totalsize;
 543         memblock_reserve(self_base, self_size);
 544
 545 #ifdef CONFIG_BLK_DEV_INITRD
 546         /* then reserve the initrd, if any */
 547         if (initrd_start && (initrd_end > initrd_start))
 548                 memblock_reserve(__pa(initrd_start), initrd_end - initrd_start);
 549 #endif /* CONFIG_BLK_DEV_INITRD */
 550
 551 #ifdef CONFIG_PPC32
 552         /*
 553          * Handle the case where we might be booting from an old kexec
 554          * image that setup the mem_rsvmap as pairs of 32-bit values
 555          */
 556         if (*reserve_map > 0xffffffffull) {
 557                 u32 base_32, size_32;
 558                 u32 *reserve_map_32 = (u32 *)reserve_map;
 559
 560                 while (1) {
 561                         base_32 = *(reserve_map_32++);
 562                         size_32 = *(reserve_map_32++);
 563                         if (size_32 == 0)
 564                                 break;
 565                         /* skip if the reservation is for the blob */
 566                         if (base_32 == self_base && size_32 == self_size)
 567                                 continue;
 568                         DBG("reserving: %x -> %x\n", base_32, size_32);
 569                         memblock_reserve(base_32, size_32);
 570                 }
 571                 return;
 572         }
 573 #endif
 574         while (1) {
 575                 base = *(reserve_map++);
 576                 size = *(reserve_map++);
 577                 if (size == 0)
 578                         break;
 579                 DBG("reserving: %llx -> %llx\n", base, size);
 580                 memblock_reserve(base, size);
 581         }
 582 }
 583
 584 #ifdef CONFIG_PHYP_DUMP
 585 /**
 586  * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
 587  *
 588  * Function to find the largest size we need to reserve
 589  * during early boot process.
 590  *
 591  * It either looks for boot param and returns that OR
 592  * returns larger of 256 or 5% rounded down to multiples of 256MB.
 593  *
 594  */
 595 static inline unsigned long phyp_dump_calculate_reserve_size(void)
 596 {
 597         unsigned long tmp;
 598
 599         if (phyp_dump_info->reserve_bootvar)
 600                 return phyp_dump_info->reserve_bootvar;
 601
 602         /* divide by 20 to get 5% of value */
 603         tmp = memblock_end_of_DRAM();
 604         do_div(tmp, 20);
 605
 606         /* round it down in multiples of 256 */
 607         tmp = tmp & ~0x0FFFFFFFUL;
 608
 609         return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
 610 }
 611
 612 /**
 613  * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
 614  *
 615  * This routine may reserve memory regions in the kernel only
 616  * if the system is supported and a dump was taken in last
 617  * boot instance or if the hardware is supported and the
 618  * scratch area needs to be setup. In other instances it returns
 619  * without reserving anything. The memory in case of dump being
 620  * active is freed when the dump is collected (by userland tools).
 621  */
 622 static void __init phyp_dump_reserve_mem(void)
 623 {
 624         unsigned long base, size;
 625         unsigned long variable_reserve_size;
 626
 627         if (!phyp_dump_info->phyp_dump_configured) {
 628                 printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
 629                 return;
 630         }
 631
 632         if (!phyp_dump_info->phyp_dump_at_boot) {
 633                 printk(KERN_INFO "Phyp-dump disabled at boot time\n");
 634                 return;
 635         }
 636
 637         variable_reserve_size = phyp_dump_calculate_reserve_size();
 638
 639         if (phyp_dump_info->phyp_dump_is_active) {
 640                 /* Reserve *everything* above RMR.Area freed by userland tools*/
 641                 base = variable_reserve_size;
 642                 size = memblock_end_of_DRAM() - base;
 643
 644                 /* XXX crashed_ram_end is wrong, since it may be beyond
 645                  * the memory_limit, it will need to be adjusted. */
 646                 memblock_reserve(base, size);
 647
 648                 phyp_dump_info->init_reserve_start = base;
 649                 phyp_dump_info->init_reserve_size = size;
 650         } else {
 651                 size = phyp_dump_info->cpu_state_size +
 652                         phyp_dump_info->hpte_region_size +
 653                         variable_reserve_size;
 654                 base = memblock_end_of_DRAM() - size;
 655                 memblock_reserve(base, size);
 656                 phyp_dump_info->init_reserve_start = base;
 657                 phyp_dump_info->init_reserve_size = size;
 658         }
 659 }
 660 #else
 661 static inline void __init phyp_dump_reserve_mem(void) {}
 662 #endif /* CONFIG_PHYP_DUMP  && CONFIG_PPC_RTAS */
 663
 664 void __init early_init_devtree(void *params)
 665 {
 666         phys_addr_t limit;
 667
 668         DBG(" -> early_init_devtree(%p)\n", params);
 669
 670         /* Setup flat device-tree pointer */
 671         initial_boot_params = params;
 672
 673 #ifdef CONFIG_PPC_RTAS
 674         /* Some machines might need RTAS info for debugging, grab it now. */
 675         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 676 #endif
 677
 678 #ifdef CONFIG_PHYP_DUMP
 679         /* scan tree to see if dump occurred during last boot */
 680         of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
 681 #endif
 682
 683         /* Retrieve various informations from the /chosen node of the
 684          * device-tree, including the platform type, initrd location and
 685          * size, TCE reserve, and more ...
 686          */
 687         of_scan_flat_dt(early_init_dt_scan_chosen_ppc, NULL);
 688
 689         /* Scan memory nodes and rebuild MEMBLOCKs */
 690         memblock_init();
 691
 692         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 693         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 694
 695         /* Save command line for /proc/cmdline and then parse parameters */
 696         strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
 697         parse_early_param();
 698
 699         /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
 700         memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
 701         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 702         if (PHYSICAL_START > MEMORY_START)
 703                 memblock_reserve(MEMORY_START, 0x8000);
 704         reserve_kdump_trampoline();
 705         reserve_crashkernel();
 706         early_reserve_mem();
 707         phyp_dump_reserve_mem();
 708
 709         limit = memory_limit;
 710         if (! limit) {
 711                 phys_addr_t memsize;
 712
 713                 /* Ensure that total memory size is page-aligned, because
 714                  * otherwise mark_bootmem() gets upset. */
 715                 memblock_analyze();
 716                 memsize = memblock_phys_mem_size();
 717                 if ((memsize & PAGE_MASK) != memsize)
 718                         limit = memsize & PAGE_MASK;
 719         }
 720         memblock_enforce_memory_limit(limit);
 721
 722         memblock_analyze();
 723         memblock_dump_all();
 724
 725         DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
 726
 727         /* We may need to relocate the flat tree, do it now.
 728          * FIXME .. and the initrd too? */
 729         move_device_tree();
 730
 731         allocate_pacas();
 732
 733         DBG("Scanning CPUs ...\n");
 734
 735         /* Retrieve CPU related informations from the flat tree
 736          * (altivec support, boot CPU ID, ...)
 737          */
 738         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 739
 740         DBG(" <- early_init_devtree()\n");
 741 }
 742
 743 /*******
 744  *
 745  * New implementation of the OF "find" APIs, return a refcounted
 746  * object, call of_node_put() when done.  The device tree and list
 747  * are protected by a rw_lock.
 748  *
 749  * Note that property management will need some locking as well,
 750  * this isn't dealt with yet.
 751  *
 752  *******/
 753
 754 /**
 755  *      of_find_next_cache_node - Find a node's subsidiary cache
 756  *      @np:    node of type "cpu" or "cache"
 757  *
 758  *      Returns a node pointer with refcount incremented, use
 759  *      of_node_put() on it when done.  Caller should hold a reference
 760  *      to np.
 761  */
 762 struct device_node *of_find_next_cache_node(struct device_node *np)
 763 {
 764         struct device_node *child;
 765         const phandle *handle;
 766
 767         handle = of_get_property(np, "l2-cache", NULL);
 768         if (!handle)
 769                 handle = of_get_property(np, "next-level-cache", NULL);
 770
 771         if (handle)
 772                 return of_find_node_by_phandle(*handle);
 773
 774         /* OF on pmac has nodes instead of properties named "l2-cache"
 775          * beneath CPU nodes.
 776          */
 777         if (!strcmp(np->type, "cpu"))
 778                 for_each_child_of_node(np, child)
 779                         if (!strcmp(child->type, "cache"))
 780                                 return child;
 781
 782         return NULL;
 783 }
 784
 785 #ifdef CONFIG_PPC_PSERIES
 786 /*
 787  * Fix up the uninitialized fields in a new device node:
 788  * name, type and pci-specific fields
 789  */
 790
 791 static int of_finish_dynamic_node(struct device_node *node)
 792 {
 793         struct device_node *parent = of_get_parent(node);
 794         int err = 0;
 795         const phandle *ibm_phandle;
 796
 797         node->name = of_get_property(node, "name", NULL);
 798         node->type = of_get_property(node, "device_type", NULL);
 799
 800         if (!node->name)
 801                 node->name = "<NULL>";
 802         if (!node->type)
 803                 node->type = "<NULL>";
 804
 805         if (!parent) {
 806                 err = -ENODEV;
 807                 goto out;
 808         }
 809
 810         /* We don't support that function on PowerMac, at least
 811          * not yet
 812          */
 813         if (machine_is(powermac))
 814                 return -ENODEV;
 815
 816         /* fix up new node's phandle field */
 817         if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
 818                 node->phandle = *ibm_phandle;
 819
 820 out:
 821         of_node_put(parent);
 822         return err;
 823 }
 824
 825 static int prom_reconfig_notifier(struct notifier_block *nb,
 826                                   unsigned long action, void *node)
 827 {
 828         int err;
 829
 830         switch (action) {
 831         case PSERIES_RECONFIG_ADD:
 832                 err = of_finish_dynamic_node(node);
 833                 if (err < 0) {
 834                         printk(KERN_ERR "finish_node returned %d\n", err);
 835                         err = NOTIFY_BAD;
 836                 }
 837                 break;
 838         default:
 839                 err = NOTIFY_DONE;
 840                 break;
 841         }
 842         return err;
 843 }
 844
 845 static struct notifier_block prom_reconfig_nb = {
 846         .notifier_call = prom_reconfig_notifier,
 847         .priority = 10, /* This one needs to run first */
 848 };
 849
 850 static int __init prom_reconfig_setup(void)
 851 {
 852         return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
 853 }
 854 __initcall(prom_reconfig_setup);
 855 #endif
 856
 857 /* Find the device node for a given logical cpu number, also returns the cpu
 858  * local thread number (index in ibm,interrupt-server#s) if relevant and
 859  * asked for (non NULL)
 860  */
 861 struct device_node *of_get_cpu_node(int cpu, unsigned int *thread)
 862 {
 863         int hardid;
 864         struct device_node *np;
 865
 866         hardid = get_hard_smp_processor_id(cpu);
 867
 868         for_each_node_by_type(np, "cpu") {
 869                 const u32 *intserv;
 870                 unsigned int plen, t;
 871
 872                 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
 873                  * fallback to "reg" property and assume no threads
 874                  */
 875                 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
 876                                 &plen);
 877                 if (intserv == NULL) {
 878                         const u32 *reg = of_get_property(np, "reg", NULL);
 879                         if (reg == NULL)
 880                                 continue;
 881                         if (*reg == hardid) {
 882                                 if (thread)
 883                                         *thread = 0;
 884                                 return np;
 885                         }
 886                 } else {
 887                         plen /= sizeof(u32);
 888                         for (t = 0; t < plen; t++) {
 889                                 if (hardid == intserv[t]) {
 890                                         if (thread)
 891                                                 *thread = t;
 892                                         return np;
 893                                 }
 894                         }
 895                 }
 896         }
 897         return NULL;
 898 }
 899 EXPORT_SYMBOL(of_get_cpu_node);
 900
 901 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
 902 static struct debugfs_blob_wrapper flat_dt_blob;
 903
 904 static int __init export_flat_device_tree(void)
 905 {
 906         struct dentry *d;
 907
 908         flat_dt_blob.data = initial_boot_params;
 909         flat_dt_blob.size = initial_boot_params->totalsize;
 910
 911         d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
 912                                 powerpc_debugfs_root, &flat_dt_blob);
 913         if (!d)
 914                 return 1;
 915
 916         return 0;
 917 }
 918 __initcall(export_flat_device_tree);
 919 #endif