arch/powerpc/kernel/prom.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Procedures for creating, accessing and interpreting the device tree.
   4  *
   5  * Paul Mackerras       August 1996.
   6  * Copyright (C) 1996-2005 Paul Mackerras.
   7  *
   8  *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
   9  *    {engebret|bergner}@us.ibm.com
  10  */
  11
  12 #undef DEBUG
  13
  14 #include <stdarg.h>
  15 #include <linux/kernel.h>
  16 #include <linux/string.h>
  17 #include <linux/init.h>
  18 #include <linux/threads.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/types.h>
  21 #include <linux/pci.h>
  22 #include <linux/delay.h>
  23 #include <linux/initrd.h>
  24 #include <linux/bitops.h>
  25 #include <linux/export.h>
  26 #include <linux/kexec.h>
  27 #include <linux/irq.h>
  28 #include <linux/memblock.h>
  29 #include <linux/of.h>
  30 #include <linux/of_fdt.h>
  31 #include <linux/libfdt.h>
  32 #include <linux/cpu.h>
  33 #include <linux/pgtable.h>
  34
  35 #include <asm/prom.h>
  36 #include <asm/rtas.h>
  37 #include <asm/page.h>
  38 #include <asm/processor.h>
  39 #include <asm/irq.h>
  40 #include <asm/io.h>
  41 #include <asm/kdump.h>
  42 #include <asm/smp.h>
  43 #include <asm/mmu.h>
  44 #include <asm/paca.h>
  45 #include <asm/powernv.h>
  46 #include <asm/iommu.h>
  47 #include <asm/btext.h>
  48 #include <asm/sections.h>
  49 #include <asm/machdep.h>
  50 #include <asm/pci-bridge.h>
  51 #include <asm/kexec.h>
  52 #include <asm/opal.h>
  53 #include <asm/fadump.h>
  54 #include <asm/epapr_hcalls.h>
  55 #include <asm/firmware.h>
  56 #include <asm/dt_cpu_ftrs.h>
  57 #include <asm/drmem.h>
  58 #include <asm/ultravisor.h>
  59
  60 #include <mm/mmu_decl.h>
  61
  62 #ifdef DEBUG
  63 #define DBG(fmt...) printk(KERN_ERR fmt)
  64 #else
  65 #define DBG(fmt...)
  66 #endif
  67
  68 #ifdef CONFIG_PPC64
  69 int __initdata iommu_is_off;
  70 int __initdata iommu_force_on;
  71 unsigned long tce_alloc_start, tce_alloc_end;
  72 u64 ppc64_rma_size;
  73 #endif
  74 static phys_addr_t first_memblock_size;
  75 static int __initdata boot_cpu_count;
  76
  77 static int __init early_parse_mem(char *p)
  78 {
  79         if (!p)
  80                 return 1;
  81
  82         memory_limit = PAGE_ALIGN(memparse(p, &p));
  83         DBG("memory limit = 0x%llx\n", memory_limit);
  84
  85         return 0;
  86 }
  87 early_param("mem", early_parse_mem);
  88
  89 /*
  90  * overlaps_initrd - check for overlap with page aligned extension of
  91  * initrd.
  92  */
  93 static inline int overlaps_initrd(unsigned long start, unsigned long size)
  94 {
  95 #ifdef CONFIG_BLK_DEV_INITRD
  96         if (!initrd_start)
  97                 return 0;
  98
  99         return  (start + size) > ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
 100                         start <= ALIGN(initrd_end, PAGE_SIZE);
 101 #else
 102         return 0;
 103 #endif
 104 }
 105
 106 /**
 107  * move_device_tree - move tree to an unused area, if needed.
 108  *
 109  * The device tree may be allocated beyond our memory limit, or inside the
 110  * crash kernel region for kdump, or within the page aligned range of initrd.
 111  * If so, move it out of the way.
 112  */
 113 static void __init move_device_tree(void)
 114 {
 115         unsigned long start, size;
 116         void *p;
 117
 118         DBG("-> move_device_tree\n");
 119
 120         start = __pa(initial_boot_params);
 121         size = fdt_totalsize(initial_boot_params);
 122
 123         if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
 124             !memblock_is_memory(start + size - 1) ||
 125             overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
 126                 p = memblock_alloc_raw(size, PAGE_SIZE);
 127                 if (!p)
 128                         panic("Failed to allocate %lu bytes to move device tree\n",
 129                               size);
 130                 memcpy(p, initial_boot_params, size);
 131                 initial_boot_params = p;
 132                 DBG("Moved device tree to 0x%px\n", p);
 133         }
 134
 135         DBG("<- move_device_tree\n");
 136 }
 137
 138 /*
 139  * ibm,pa-features is a per-cpu property that contains a string of
 140  * attribute descriptors, each of which has a 2 byte header plus up
 141  * to 254 bytes worth of processor attribute bits.  First header
 142  * byte specifies the number of bytes following the header.
 143  * Second header byte is an "attribute-specifier" type, of which
 144  * zero is the only currently-defined value.
 145  * Implementation:  Pass in the byte and bit offset for the feature
 146  * that we are interested in.  The function will return -1 if the
 147  * pa-features property is missing, or a 1/0 to indicate if the feature
 148  * is supported/not supported.  Note that the bit numbers are
 149  * big-endian to match the definition in PAPR.
 150  */
 151 static struct ibm_pa_feature {
 152         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 153         unsigned long   mmu_features;   /* MMU_FTR_xxx bit */
 154         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 155         unsigned int    cpu_user_ftrs2; /* PPC_FEATURE2_xxx bit */
 156         unsigned char   pabyte;         /* byte number in ibm,pa-features */
 157         unsigned char   pabit;          /* bit number (big-endian) */
 158         unsigned char   invert;         /* if 1, pa bit set => clear feature */
 159 } ibm_pa_features[] __initdata = {
 160         { .pabyte = 0,  .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
 161         { .pabyte = 0,  .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
 162         { .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
 163         { .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
 164         { .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE },
 165 #ifdef CONFIG_PPC_RADIX_MMU
 166         { .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX | MMU_FTR_GTSE },
 167 #endif
 168         { .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
 169                                     .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
 170         /*
 171          * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
 172          * we don't want to turn on TM here, so we use the *_COMP versions
 173          * which are 0 if the kernel doesn't support TM.
 174          */
 175         { .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
 176           .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
 177
 178         { .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
 179 };
 180
 181 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
 182                                  unsigned long tablelen,
 183                                  struct ibm_pa_feature *fp,
 184                                  unsigned long ft_size)
 185 {
 186         unsigned long i, len, bit;
 187
 188         /* find descriptor with type == 0 */
 189         for (;;) {
 190                 if (tablelen < 3)
 191                         return;
 192                 len = 2 + ftrs[0];
 193                 if (tablelen < len)
 194                         return;         /* descriptor 0 not found */
 195                 if (ftrs[1] == 0)
 196                         break;
 197                 tablelen -= len;
 198                 ftrs += len;
 199         }
 200
 201         /* loop over bits we know about */
 202         for (i = 0; i < ft_size; ++i, ++fp) {
 203                 if (fp->pabyte >= ftrs[0])
 204                         continue;
 205                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 206                 if (bit ^ fp->invert) {
 207                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 208                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 209                         cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
 210                         cur_cpu_spec->mmu_features |= fp->mmu_features;
 211                 } else {
 212                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
 213                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
 214                         cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
 215                         cur_cpu_spec->mmu_features &= ~fp->mmu_features;
 216                 }
 217         }
 218 }
 219
 220 static void __init check_cpu_pa_features(unsigned long node)
 221 {
 222         const unsigned char *pa_ftrs;
 223         int tablelen;
 224
 225         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
 226         if (pa_ftrs == NULL)
 227                 return;
 228
 229         scan_features(node, pa_ftrs, tablelen,
 230                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 231 }
 232
 233 #ifdef CONFIG_PPC_BOOK3S_64
 234 static void __init init_mmu_slb_size(unsigned long node)
 235 {
 236         const __be32 *slb_size_ptr;
 237
 238         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
 239                         of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 240
 241         if (slb_size_ptr)
 242                 mmu_slb_size = be32_to_cpup(slb_size_ptr);
 243 }
 244 #else
 245 #define init_mmu_slb_size(node) do { } while(0)
 246 #endif
 247
 248 static struct feature_property {
 249         const char *name;
 250         u32 min_value;
 251         unsigned long cpu_feature;
 252         unsigned long cpu_user_ftr;
 253 } feature_properties[] __initdata = {
 254 #ifdef CONFIG_ALTIVEC
 255         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 256         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 257 #endif /* CONFIG_ALTIVEC */
 258 #ifdef CONFIG_VSX
 259         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 260         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 261 #endif /* CONFIG_VSX */
 262 #ifdef CONFIG_PPC64
 263         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 264         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 265         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 266 #endif /* CONFIG_PPC64 */
 267 };
 268
 269 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 270 static inline void identical_pvr_fixup(unsigned long node)
 271 {
 272         unsigned int pvr;
 273         const char *model = of_get_flat_dt_prop(node, "model", NULL);
 274
 275         /*
 276          * Since 440GR(x)/440EP(x) processors have the same pvr,
 277          * we check the node path and set bit 28 in the cur_cpu_spec
 278          * pvr for EP(x) processor version. This bit is always 0 in
 279          * the "real" pvr. Then we call identify_cpu again with
 280          * the new logical pvr to enable FPU support.
 281          */
 282         if (model && strstr(model, "440EP")) {
 283                 pvr = cur_cpu_spec->pvr_value | 0x8;
 284                 identify_cpu(0, pvr);
 285                 DBG("Using logical pvr %x for %s\n", pvr, model);
 286         }
 287 }
 288 #else
 289 #define identical_pvr_fixup(node) do { } while(0)
 290 #endif
 291
 292 static void __init check_cpu_feature_properties(unsigned long node)
 293 {
 294         int i;
 295         struct feature_property *fp = feature_properties;
 296         const __be32 *prop;
 297
 298         for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
 299                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 300                 if (prop && be32_to_cpup(prop) >= fp->min_value) {
 301                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 302                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 303                 }
 304         }
 305 }
 306
 307 static int __init early_init_dt_scan_cpus(unsigned long node,
 308                                           const char *uname, int depth,
 309                                           void *data)
 310 {
 311         const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 312         const __be32 *prop;
 313         const __be32 *intserv;
 314         int i, nthreads;
 315         int len;
 316         int found = -1;
 317         int found_thread = 0;
 318
 319         /* We are scanning "cpu" nodes only */
 320         if (type == NULL || strcmp(type, "cpu") != 0)
 321                 return 0;
 322
 323         /* Get physical cpuid */
 324         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
 325         if (!intserv)
 326                 intserv = of_get_flat_dt_prop(node, "reg", &len);
 327
 328         nthreads = len / sizeof(int);
 329
 330         /*
 331          * Now see if any of these threads match our boot cpu.
 332          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
 333          */
 334         for (i = 0; i < nthreads; i++) {
 335                 if (be32_to_cpu(intserv[i]) ==
 336                         fdt_boot_cpuid_phys(initial_boot_params)) {
 337                         found = boot_cpu_count;
 338                         found_thread = i;
 339                 }
 340 #ifdef CONFIG_SMP
 341                 /* logical cpu id is always 0 on UP kernels */
 342                 boot_cpu_count++;
 343 #endif
 344         }
 345
 346         /* Not the boot CPU */
 347         if (found < 0)
 348                 return 0;
 349
 350         DBG("boot cpu: logical %d physical %d\n", found,
 351             be32_to_cpu(intserv[found_thread]));
 352         boot_cpuid = found;
 353
 354         /*
 355          * PAPR defines "logical" PVR values for cpus that
 356          * meet various levels of the architecture:
 357          * 0x0f000001   Architecture version 2.04
 358          * 0x0f000002   Architecture version 2.05
 359          * If the cpu-version property in the cpu node contains
 360          * such a value, we call identify_cpu again with the
 361          * logical PVR value in order to use the cpu feature
 362          * bits appropriate for the architecture level.
 363          *
 364          * A POWER6 partition in "POWER6 architected" mode
 365          * uses the 0x0f000002 PVR value; in POWER5+ mode
 366          * it uses 0x0f000001.
 367          *
 368          * If we're using device tree CPU feature discovery then we don't
 369          * support the cpu-version property, and it's the responsibility of the
 370          * firmware/hypervisor to provide the correct feature set for the
 371          * architecture level via the ibm,powerpc-cpu-features binding.
 372          */
 373         if (!dt_cpu_ftrs_in_use()) {
 374                 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
 375                 if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
 376                         identify_cpu(0, be32_to_cpup(prop));
 377
 378                 check_cpu_feature_properties(node);
 379                 check_cpu_pa_features(node);
 380         }
 381
 382         identical_pvr_fixup(node);
 383         init_mmu_slb_size(node);
 384
 385 #ifdef CONFIG_PPC64
 386         if (nthreads == 1)
 387                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
 388         else if (!dt_cpu_ftrs_in_use())
 389                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
 390         allocate_paca(boot_cpuid);
 391 #endif
 392         set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
 393
 394         return 0;
 395 }
 396
 397 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
 398                                                 const char *uname,
 399                                                 int depth, void *data)
 400 {
 401         const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
 402
 403         /* Use common scan routine to determine if this is the chosen node */
 404         if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
 405                 return 0;
 406
 407 #ifdef CONFIG_PPC64
 408         /* check if iommu is forced on or off */
 409         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
 410                 iommu_is_off = 1;
 411         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
 412                 iommu_force_on = 1;
 413 #endif
 414
 415         /* mem=x on the command line is the preferred mechanism */
 416         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
 417         if (lprop)
 418                 memory_limit = *lprop;
 419
 420 #ifdef CONFIG_PPC64
 421         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
 422         if (lprop)
 423                 tce_alloc_start = *lprop;
 424         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
 425         if (lprop)
 426                 tce_alloc_end = *lprop;
 427 #endif
 428
 429 #ifdef CONFIG_KEXEC_CORE
 430         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
 431         if (lprop)
 432                 crashk_res.start = *lprop;
 433
 434         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
 435         if (lprop)
 436                 crashk_res.end = crashk_res.start + *lprop - 1;
 437 #endif
 438
 439         /* break now */
 440         return 1;
 441 }
 442
 443 /*
 444  * Compare the range against max mem limit and update
 445  * size if it cross the limit.
 446  */
 447
 448 #ifdef CONFIG_SPARSEMEM
 449 static bool validate_mem_limit(u64 base, u64 *size)
 450 {
 451         u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
 452
 453         if (base >= max_mem)
 454                 return false;
 455         if ((base + *size) > max_mem)
 456                 *size = max_mem - base;
 457         return true;
 458 }
 459 #else
 460 static bool validate_mem_limit(u64 base, u64 *size)
 461 {
 462         return true;
 463 }
 464 #endif
 465
 466 #ifdef CONFIG_PPC_PSERIES
 467 /*
 468  * Interpret the ibm dynamic reconfiguration memory LMBs.
 469  * This contains a list of memory blocks along with NUMA affinity
 470  * information.
 471  */
 472 static int  __init early_init_drmem_lmb(struct drmem_lmb *lmb,
 473                                         const __be32 **usm,
 474                                         void *data)
 475 {
 476         u64 base, size;
 477         int is_kexec_kdump = 0, rngs;
 478
 479         base = lmb->base_addr;
 480         size = drmem_lmb_size();
 481         rngs = 1;
 482
 483         /*
 484          * Skip this block if the reserved bit is set in flags
 485          * or if the block is not assigned to this partition.
 486          */
 487         if ((lmb->flags & DRCONF_MEM_RESERVED) ||
 488             !(lmb->flags & DRCONF_MEM_ASSIGNED))
 489                 return 0;
 490
 491         if (*usm)
 492                 is_kexec_kdump = 1;
 493
 494         if (is_kexec_kdump) {
 495                 /*
 496                  * For each memblock in ibm,dynamic-memory, a
 497                  * corresponding entry in linux,drconf-usable-memory
 498                  * property contains a counter 'p' followed by 'p'
 499                  * (base, size) duple. Now read the counter from
 500                  * linux,drconf-usable-memory property
 501                  */
 502                 rngs = dt_mem_next_cell(dt_root_size_cells, usm);
 503                 if (!rngs) /* there are no (base, size) duple */
 504                         return 0;
 505         }
 506
 507         do {
 508                 if (is_kexec_kdump) {
 509                         base = dt_mem_next_cell(dt_root_addr_cells, usm);
 510                         size = dt_mem_next_cell(dt_root_size_cells, usm);
 511                 }
 512
 513                 if (iommu_is_off) {
 514                         if (base >= 0x80000000ul)
 515                                 continue;
 516                         if ((base + size) > 0x80000000ul)
 517                                 size = 0x80000000ul - base;
 518                 }
 519
 520                 if (!validate_mem_limit(base, &size))
 521                         continue;
 522
 523                 DBG("Adding: %llx -> %llx\n", base, size);
 524                 memblock_add(base, size);
 525
 526                 if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
 527                         memblock_mark_hotplug(base, size);
 528         } while (--rngs);
 529
 530         return 0;
 531 }
 532 #endif /* CONFIG_PPC_PSERIES */
 533
 534 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
 535                                                 const char *uname,
 536                                                 int depth, void *data)
 537 {
 538 #ifdef CONFIG_PPC_PSERIES
 539         if (depth == 1 &&
 540             strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0) {
 541                 walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
 542                 return 0;
 543         }
 544 #endif
 545
 546         return early_init_dt_scan_memory(node, uname, depth, data);
 547 }
 548
 549 /*
 550  * For a relocatable kernel, we need to get the memstart_addr first,
 551  * then use it to calculate the virtual kernel start address. This has
 552  * to happen at a very early stage (before machine_init). In this case,
 553  * we just want to get the memstart_address and would not like to mess the
 554  * memblock at this stage. So introduce a variable to skip the memblock_add()
 555  * for this reason.
 556  */
 557 #ifdef CONFIG_RELOCATABLE
 558 static int add_mem_to_memblock = 1;
 559 #else
 560 #define add_mem_to_memblock 1
 561 #endif
 562
 563 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
 564 {
 565 #ifdef CONFIG_PPC64
 566         if (iommu_is_off) {
 567                 if (base >= 0x80000000ul)
 568                         return;
 569                 if ((base + size) > 0x80000000ul)
 570                         size = 0x80000000ul - base;
 571         }
 572 #endif
 573         /* Keep track of the beginning of memory -and- the size of
 574          * the very first block in the device-tree as it represents
 575          * the RMA on ppc64 server
 576          */
 577         if (base < memstart_addr) {
 578                 memstart_addr = base;
 579                 first_memblock_size = size;
 580         }
 581
 582         /* Add the chunk to the MEMBLOCK list */
 583         if (add_mem_to_memblock) {
 584                 if (validate_mem_limit(base, &size))
 585                         memblock_add(base, size);
 586         }
 587 }
 588
 589 static void __init early_reserve_mem_dt(void)
 590 {
 591         unsigned long i, dt_root;
 592         int len;
 593         const __be32 *prop;
 594
 595         early_init_fdt_reserve_self();
 596         early_init_fdt_scan_reserved_mem();
 597
 598         dt_root = of_get_flat_dt_root();
 599
 600         prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
 601
 602         if (!prop)
 603                 return;
 604
 605         DBG("Found new-style reserved-ranges\n");
 606
 607         /* Each reserved range is an (address,size) pair, 2 cells each,
 608          * totalling 4 cells per range. */
 609         for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
 610                 u64 base, size;
 611
 612                 base = of_read_number(prop + (i * 4) + 0, 2);
 613                 size = of_read_number(prop + (i * 4) + 2, 2);
 614
 615                 if (size) {
 616                         DBG("reserving: %llx -> %llx\n", base, size);
 617                         memblock_reserve(base, size);
 618                 }
 619         }
 620 }
 621
 622 static void __init early_reserve_mem(void)
 623 {
 624         __be64 *reserve_map;
 625
 626         reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
 627                         fdt_off_mem_rsvmap(initial_boot_params));
 628
 629         /* Look for the new "reserved-regions" property in the DT */
 630         early_reserve_mem_dt();
 631
 632 #ifdef CONFIG_BLK_DEV_INITRD
 633         /* Then reserve the initrd, if any */
 634         if (initrd_start && (initrd_end > initrd_start)) {
 635                 memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
 636                         ALIGN(initrd_end, PAGE_SIZE) -
 637                         ALIGN_DOWN(initrd_start, PAGE_SIZE));
 638         }
 639 #endif /* CONFIG_BLK_DEV_INITRD */
 640
 641 #ifdef CONFIG_PPC32
 642         /*
 643          * Handle the case where we might be booting from an old kexec
 644          * image that setup the mem_rsvmap as pairs of 32-bit values
 645          */
 646         if (be64_to_cpup(reserve_map) > 0xffffffffull) {
 647                 u32 base_32, size_32;
 648                 __be32 *reserve_map_32 = (__be32 *)reserve_map;
 649
 650                 DBG("Found old 32-bit reserve map\n");
 651
 652                 while (1) {
 653                         base_32 = be32_to_cpup(reserve_map_32++);
 654                         size_32 = be32_to_cpup(reserve_map_32++);
 655                         if (size_32 == 0)
 656                                 break;
 657                         DBG("reserving: %x -> %x\n", base_32, size_32);
 658                         memblock_reserve(base_32, size_32);
 659                 }
 660                 return;
 661         }
 662 #endif
 663 }
 664
 665 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 666 static bool tm_disabled __initdata;
 667
 668 static int __init parse_ppc_tm(char *str)
 669 {
 670         bool res;
 671
 672         if (kstrtobool(str, &res))
 673                 return -EINVAL;
 674
 675         tm_disabled = !res;
 676
 677         return 0;
 678 }
 679 early_param("ppc_tm", parse_ppc_tm);
 680
 681 static void __init tm_init(void)
 682 {
 683         if (tm_disabled) {
 684                 pr_info("Disabling hardware transactional memory (HTM)\n");
 685                 cur_cpu_spec->cpu_user_features2 &=
 686                         ~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
 687                 cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
 688                 return;
 689         }
 690
 691         pnv_tm_init();
 692 }
 693 #else
 694 static void tm_init(void) { }
 695 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
 696
 697 #ifdef CONFIG_PPC64
 698 static void __init save_fscr_to_task(void)
 699 {
 700         /*
 701          * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
 702          * have configured via the device tree features or via __init_FSCR().
 703          * That value will then be propagated to pid 1 (init) and all future
 704          * processes.
 705          */
 706         if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
 707                 init_task.thread.fscr = mfspr(SPRN_FSCR);
 708 }
 709 #else
 710 static inline void save_fscr_to_task(void) {};
 711 #endif
 712
 713
 714 void __init early_init_devtree(void *params)
 715 {
 716         phys_addr_t limit;
 717
 718         DBG(" -> early_init_devtree(%px)\n", params);
 719
 720         /* Too early to BUG_ON(), do it by hand */
 721         if (!early_init_dt_verify(params))
 722                 panic("BUG: Failed verifying flat device tree, bad version?");
 723
 724 #ifdef CONFIG_PPC_RTAS
 725         /* Some machines might need RTAS info for debugging, grab it now. */
 726         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 727 #endif
 728
 729 #ifdef CONFIG_PPC_POWERNV
 730         /* Some machines might need OPAL info for debugging, grab it now. */
 731         of_scan_flat_dt(early_init_dt_scan_opal, NULL);
 732
 733         /* Scan tree for ultravisor feature */
 734         of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
 735 #endif
 736
 737 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
 738         /* scan tree to see if dump is active during last boot */
 739         of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
 740 #endif
 741
 742         /* Retrieve various informations from the /chosen node of the
 743          * device-tree, including the platform type, initrd location and
 744          * size, TCE reserve, and more ...
 745          */
 746         of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
 747
 748         /* Scan memory nodes and rebuild MEMBLOCKs */
 749         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 750         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 751
 752         parse_early_param();
 753
 754         /* make sure we've parsed cmdline for mem= before this */
 755         if (memory_limit)
 756                 first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
 757         setup_initial_memory_limit(memstart_addr, first_memblock_size);
 758         /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
 759         memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
 760         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 761         if (PHYSICAL_START > MEMORY_START)
 762                 memblock_reserve(MEMORY_START, 0x8000);
 763         reserve_kdump_trampoline();
 764 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
 765         /*
 766          * If we fail to reserve memory for firmware-assisted dump then
 767          * fallback to kexec based kdump.
 768          */
 769         if (fadump_reserve_mem() == 0)
 770 #endif
 771                 reserve_crashkernel();
 772         early_reserve_mem();
 773
 774         /* Ensure that total memory size is page-aligned. */
 775         limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
 776         memblock_enforce_memory_limit(limit);
 777
 778 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
 779         if (!early_radix_enabled())
 780                 memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
 781 #endif
 782
 783         memblock_allow_resize();
 784         memblock_dump_all();
 785
 786         DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
 787
 788         /* We may need to relocate the flat tree, do it now.
 789          * FIXME .. and the initrd too? */
 790         move_device_tree();
 791
 792         allocate_paca_ptrs();
 793
 794         DBG("Scanning CPUs ...\n");
 795
 796         dt_cpu_ftrs_scan();
 797
 798         /* Retrieve CPU related informations from the flat tree
 799          * (altivec support, boot CPU ID, ...)
 800          */
 801         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 802         if (boot_cpuid < 0) {
 803                 printk("Failed to identify boot CPU !\n");
 804                 BUG();
 805         }
 806
 807         save_fscr_to_task();
 808
 809 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
 810         /* We'll later wait for secondaries to check in; there are
 811          * NCPUS-1 non-boot CPUs  :-)
 812          */
 813         spinning_secondaries = boot_cpu_count - 1;
 814 #endif
 815
 816         mmu_early_init_devtree();
 817
 818 #ifdef CONFIG_PPC_POWERNV
 819         /* Scan and build the list of machine check recoverable ranges */
 820         of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
 821 #endif
 822         epapr_paravirt_early_init();
 823
 824         /* Now try to figure out if we are running on LPAR and so on */
 825         pseries_probe_fw_features();
 826
 827         /*
 828          * Initialize pkey features and default AMR/IAMR values
 829          */
 830         pkey_early_init_devtree();
 831
 832 #ifdef CONFIG_PPC_PS3
 833         /* Identify PS3 firmware */
 834         if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
 835                 powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
 836 #endif
 837
 838         tm_init();
 839
 840         DBG(" <- early_init_devtree()\n");
 841 }
 842
 843 #ifdef CONFIG_RELOCATABLE
 844 /*
 845  * This function run before early_init_devtree, so we have to init
 846  * initial_boot_params.
 847  */
 848 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
 849 {
 850         /* Setup flat device-tree pointer */
 851         initial_boot_params = params;
 852
 853         /*
 854          * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
 855          * mess the memblock.
 856          */
 857         add_mem_to_memblock = 0;
 858         of_scan_flat_dt(early_init_dt_scan_root, NULL);
 859         of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
 860         add_mem_to_memblock = 1;
 861
 862         if (size)
 863                 *size = first_memblock_size;
 864 }
 865 #endif
 866
 867 /*******
 868  *
 869  * New implementation of the OF "find" APIs, return a refcounted
 870  * object, call of_node_put() when done.  The device tree and list
 871  * are protected by a rw_lock.
 872  *
 873  * Note that property management will need some locking as well,
 874  * this isn't dealt with yet.
 875  *
 876  *******/
 877
 878 /**
 879  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
 880  * @np: device node of the device
 881  *
 882  * This looks for a property "ibm,chip-id" in the node or any
 883  * of its parents and returns its content, or -1 if it cannot
 884  * be found.
 885  */
 886 int of_get_ibm_chip_id(struct device_node *np)
 887 {
 888         of_node_get(np);
 889         while (np) {
 890                 u32 chip_id;
 891
 892                 /*
 893                  * Skiboot may produce memory nodes that contain more than one
 894                  * cell in chip-id, we only read the first one here.
 895                  */
 896                 if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
 897                         of_node_put(np);
 898                         return chip_id;
 899                 }
 900
 901                 np = of_get_next_parent(np);
 902         }
 903         return -1;
 904 }
 905 EXPORT_SYMBOL(of_get_ibm_chip_id);
 906
 907 /**
 908  * cpu_to_chip_id - Return the cpus chip-id
 909  * @cpu: The logical cpu number.
 910  *
 911  * Return the value of the ibm,chip-id property corresponding to the given
 912  * logical cpu number. If the chip-id can not be found, returns -1.
 913  */
 914 int cpu_to_chip_id(int cpu)
 915 {
 916         struct device_node *np;
 917
 918         np = of_get_cpu_node(cpu, NULL);
 919         if (!np)
 920                 return -1;
 921
 922         of_node_put(np);
 923         return of_get_ibm_chip_id(np);
 924 }
 925 EXPORT_SYMBOL(cpu_to_chip_id);
 926
 927 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
 928 {
 929 #ifdef CONFIG_SMP
 930         /*
 931          * Early firmware scanning must use this rather than
 932          * get_hard_smp_processor_id because we don't have pacas allocated
 933          * until memory topology is discovered.
 934          */
 935         if (cpu_to_phys_id != NULL)
 936                 return (int)phys_id == cpu_to_phys_id[cpu];
 937 #endif
 938
 939         return (int)phys_id == get_hard_smp_processor_id(cpu);
 940 }