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 <linux/kernel.h>
  15 #include <linux/string.h>
  16 #include <linux/init.h>
  17 #include <linux/threads.h>
  18 #include <linux/spinlock.h>
  19 #include <linux/types.h>
  20 #include <linux/pci.h>
  21 #include <linux/delay.h>
  22 #include <linux/initrd.h>
  23 #include <linux/bitops.h>
  24 #include <linux/export.h>
  25 #include <linux/kexec.h>
  26 #include <linux/irq.h>
  27 #include <linux/memblock.h>
  28 #include <linux/of.h>
  29 #include <linux/of_fdt.h>
  30 #include <linux/libfdt.h>
  31 #include <linux/cpu.h>
  32 #include <linux/pgtable.h>
  33 #include <linux/seq_buf.h>
  34
  35 #include <asm/rtas.h>
  36 #include <asm/page.h>
  37 #include <asm/processor.h>
  38 #include <asm/irq.h>
  39 #include <asm/io.h>
  40 #include <asm/kdump.h>
  41 #include <asm/smp.h>
  42 #include <asm/mmu.h>
  43 #include <asm/paca.h>
  44 #include <asm/powernv.h>
  45 #include <asm/iommu.h>
  46 #include <asm/btext.h>
  47 #include <asm/sections.h>
  48 #include <asm/setup.h>
  49 #include <asm/pci-bridge.h>
  50 #include <asm/kexec.h>
  51 #include <asm/opal.h>
  52 #include <asm/fadump.h>
  53 #include <asm/epapr_hcalls.h>
  54 #include <asm/firmware.h>
  55 #include <asm/dt_cpu_ftrs.h>
  56 #include <asm/drmem.h>
  57 #include <asm/ultravisor.h>
  58 #include <asm/prom.h>
  59 #include <asm/plpks.h>
  60
  61 #include <mm/mmu_decl.h>
  62
  63 #ifdef DEBUG
  64 #define DBG(fmt...) printk(KERN_ERR fmt)
  65 #else
  66 #define DBG(fmt...)
  67 #endif
  68
  69 int *chip_id_lookup_table;
  70
  71 #ifdef CONFIG_PPC64
  72 int __initdata iommu_is_off;
  73 int __initdata iommu_force_on;
  74 unsigned long tce_alloc_start, tce_alloc_end;
  75 u64 ppc64_rma_size;
  76 unsigned int boot_cpu_node_count __ro_after_init;
  77 #endif
  78 static phys_addr_t first_memblock_size;
  79 static int __initdata boot_cpu_count;
  80
  81 static int __init early_parse_mem(char *p)
  82 {
  83         if (!p)
  84                 return 1;
  85
  86         memory_limit = PAGE_ALIGN(memparse(p, &p));
  87         DBG("memory limit = 0x%llx\n", memory_limit);
  88
  89         return 0;
  90 }
  91 early_param("mem", early_parse_mem);
  92
  93 /*
  94  * overlaps_initrd - check for overlap with page aligned extension of
  95  * initrd.
  96  */
  97 static inline int overlaps_initrd(unsigned long start, unsigned long size)
  98 {
  99 #ifdef CONFIG_BLK_DEV_INITRD
 100         if (!initrd_start)
 101                 return 0;
 102
 103         return  (start + size) > ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
 104                         start <= ALIGN(initrd_end, PAGE_SIZE);
 105 #else
 106         return 0;
 107 #endif
 108 }
 109
 110 /**
 111  * move_device_tree - move tree to an unused area, if needed.
 112  *
 113  * The device tree may be allocated beyond our memory limit, or inside the
 114  * crash kernel region for kdump, or within the page aligned range of initrd.
 115  * If so, move it out of the way.
 116  */
 117 static void __init move_device_tree(void)
 118 {
 119         unsigned long start, size;
 120         void *p;
 121
 122         DBG("-> move_device_tree\n");
 123
 124         start = __pa(initial_boot_params);
 125         size = fdt_totalsize(initial_boot_params);
 126
 127         if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
 128             !memblock_is_memory(start + size - 1) ||
 129             overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
 130                 p = memblock_alloc_raw(size, PAGE_SIZE);
 131                 if (!p)
 132                         panic("Failed to allocate %lu bytes to move device tree\n",
 133                               size);
 134                 memcpy(p, initial_boot_params, size);
 135                 initial_boot_params = p;
 136                 DBG("Moved device tree to 0x%px\n", p);
 137         }
 138
 139         DBG("<- move_device_tree\n");
 140 }
 141
 142 /*
 143  * ibm,pa/pi-features is a per-cpu property that contains a string of
 144  * attribute descriptors, each of which has a 2 byte header plus up
 145  * to 254 bytes worth of processor attribute bits.  First header
 146  * byte specifies the number of bytes following the header.
 147  * Second header byte is an "attribute-specifier" type, of which
 148  * zero is the only currently-defined value.
 149  * Implementation:  Pass in the byte and bit offset for the feature
 150  * that we are interested in.  The function will return -1 if the
 151  * pa-features property is missing, or a 1/0 to indicate if the feature
 152  * is supported/not supported.  Note that the bit numbers are
 153  * big-endian to match the definition in PAPR.
 154  * Note: the 'clear' flag clears the feature if the bit is set in the
 155  * ibm,pa/pi-features property, it does not set the feature if the
 156  * bit is clear.
 157  */
 158 struct ibm_feature {
 159         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
 160         unsigned long   mmu_features;   /* MMU_FTR_xxx bit */
 161         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
 162         unsigned int    cpu_user_ftrs2; /* PPC_FEATURE2_xxx bit */
 163         unsigned char   pabyte;         /* byte number in ibm,pa/pi-features */
 164         unsigned char   pabit;          /* bit number (big-endian) */
 165         unsigned char   clear;          /* if 1, pa bit set => clear feature */
 166 };
 167
 168 static struct ibm_feature ibm_pa_features[] __initdata = {
 169         { .pabyte = 0,  .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
 170         { .pabyte = 0,  .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
 171         { .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
 172         { .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
 173         { .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE },
 174 #ifdef CONFIG_PPC_RADIX_MMU
 175         { .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX | MMU_FTR_GTSE },
 176 #endif
 177         { .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
 178                                     .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
 179         /*
 180          * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
 181          * we don't want to turn on TM here, so we use the *_COMP versions
 182          * which are 0 if the kernel doesn't support TM.
 183          */
 184         { .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
 185           .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
 186
 187         { .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
 188         { .pabyte = 68, .pabit = 5, .cpu_features = CPU_FTR_DEXCR_NPHIE },
 189 };
 190
 191 /*
 192  * ibm,pi-features property provides the support of processor specific
 193  * options not described in ibm,pa-features. Right now use byte 0, bit 3
 194  * which indicates the occurrence of DSI interrupt when the paste operation
 195  * on the suspended NX window.
 196  */
 197 static struct ibm_feature ibm_pi_features[] __initdata = {
 198         { .pabyte = 0, .pabit = 3, .mmu_features  = MMU_FTR_NX_DSI },
 199         { .pabyte = 0, .pabit = 4, .cpu_features  = CPU_FTR_DBELL, .clear = 1 },
 200 };
 201
 202 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
 203                                  unsigned long tablelen,
 204                                  struct ibm_feature *fp,
 205                                  unsigned long ft_size)
 206 {
 207         unsigned long i, len, bit;
 208
 209         /* find descriptor with type == 0 */
 210         for (;;) {
 211                 if (tablelen < 3)
 212                         return;
 213                 len = 2 + ftrs[0];
 214                 if (tablelen < len)
 215                         return;         /* descriptor 0 not found */
 216                 if (ftrs[1] == 0)
 217                         break;
 218                 tablelen -= len;
 219                 ftrs += len;
 220         }
 221
 222         /* loop over bits we know about */
 223         for (i = 0; i < ft_size; ++i, ++fp) {
 224                 if (fp->pabyte >= ftrs[0])
 225                         continue;
 226                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
 227                 if (bit && !fp->clear) {
 228                         cur_cpu_spec->cpu_features |= fp->cpu_features;
 229                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
 230                         cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
 231                         cur_cpu_spec->mmu_features |= fp->mmu_features;
 232                 } else if (bit == fp->clear) {
 233                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
 234                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
 235                         cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
 236                         cur_cpu_spec->mmu_features &= ~fp->mmu_features;
 237                 }
 238         }
 239 }
 240
 241 static void __init check_cpu_features(unsigned long node, char *name,
 242                                       struct ibm_feature *fp,
 243                                       unsigned long size)
 244 {
 245         const unsigned char *pa_ftrs;
 246         int tablelen;
 247
 248         pa_ftrs = of_get_flat_dt_prop(node, name, &tablelen);
 249         if (pa_ftrs == NULL)
 250                 return;
 251
 252         scan_features(node, pa_ftrs, tablelen, fp, size);
 253 }
 254
 255 #ifdef CONFIG_PPC_64S_HASH_MMU
 256 static void __init init_mmu_slb_size(unsigned long node)
 257 {
 258         const __be32 *slb_size_ptr;
 259
 260         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
 261                         of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
 262
 263         if (slb_size_ptr)
 264                 mmu_slb_size = be32_to_cpup(slb_size_ptr);
 265 }
 266 #else
 267 #define init_mmu_slb_size(node) do { } while(0)
 268 #endif
 269
 270 static struct feature_property {
 271         const char *name;
 272         u32 min_value;
 273         unsigned long cpu_feature;
 274         unsigned long cpu_user_ftr;
 275 } feature_properties[] __initdata = {
 276 #ifdef CONFIG_ALTIVEC
 277         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 278         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 279 #endif /* CONFIG_ALTIVEC */
 280 #ifdef CONFIG_VSX
 281         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
 282         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
 283 #endif /* CONFIG_VSX */
 284 #ifdef CONFIG_PPC64
 285         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
 286         {"ibm,purr", 1, CPU_FTR_PURR, 0},
 287         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
 288 #endif /* CONFIG_PPC64 */
 289 };
 290
 291 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
 292 static __init void identical_pvr_fixup(unsigned long node)
 293 {
 294         unsigned int pvr;
 295         const char *model = of_get_flat_dt_prop(node, "model", NULL);
 296
 297         /*
 298          * Since 440GR(x)/440EP(x) processors have the same pvr,
 299          * we check the node path and set bit 28 in the cur_cpu_spec
 300          * pvr for EP(x) processor version. This bit is always 0 in
 301          * the "real" pvr. Then we call identify_cpu again with
 302          * the new logical pvr to enable FPU support.
 303          */
 304         if (model && strstr(model, "440EP")) {
 305                 pvr = cur_cpu_spec->pvr_value | 0x8;
 306                 identify_cpu(0, pvr);
 307                 DBG("Using logical pvr %x for %s\n", pvr, model);
 308         }
 309 }
 310 #else
 311 #define identical_pvr_fixup(node) do { } while(0)
 312 #endif
 313
 314 static void __init check_cpu_feature_properties(unsigned long node)
 315 {
 316         int i;
 317         struct feature_property *fp = feature_properties;
 318         const __be32 *prop;
 319
 320         for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
 321                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
 322                 if (prop && be32_to_cpup(prop) >= fp->min_value) {
 323                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
 324                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
 325                 }
 326         }
 327 }
 328
 329 static int __init early_init_dt_scan_cpus(unsigned long node,
 330                                           const char *uname, int depth,
 331                                           void *data)
 332 {
 333         const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
 334         const __be32 *cpu_version = NULL;
 335         const __be32 *prop;
 336         const __be32 *intserv;
 337         int i, nthreads;
 338         int len;
 339         int found = -1;
 340         int found_thread = 0;
 341
 342         /* We are scanning "cpu" nodes only */
 343         if (type == NULL || strcmp(type, "cpu") != 0)
 344                 return 0;
 345
 346         if (IS_ENABLED(CONFIG_PPC64))
 347                 boot_cpu_node_count++;
 348
 349         /* Get physical cpuid */
 350         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
 351         if (!intserv)
 352                 intserv = of_get_flat_dt_prop(node, "reg", &len);
 353
 354         nthreads = len / sizeof(int);
 355
 356         /*
 357          * Now see if any of these threads match our boot cpu.
 358          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
 359          */
 360         for (i = 0; i < nthreads; i++) {
 361                 if (be32_to_cpu(intserv[i]) ==
 362                         fdt_boot_cpuid_phys(initial_boot_params)) {
 363                         found = boot_cpu_count;
 364                         found_thread = i;
 365                 }
 366 #ifdef CONFIG_SMP
 367                 /* logical cpu id is always 0 on UP kernels */
 368                 boot_cpu_count++;
 369 #endif
 370         }
 371
 372         /* Not the boot CPU */
 373         if (found < 0)
 374                 return 0;
 375
 376         boot_cpuid = found;
 377
 378         if (IS_ENABLED(CONFIG_PPC64))
 379                 boot_cpu_hwid = be32_to_cpu(intserv[found_thread]);
 380
 381         if (nr_cpu_ids % nthreads != 0) {
 382                 set_nr_cpu_ids(ALIGN(nr_cpu_ids, nthreads));
 383                 pr_warn("nr_cpu_ids was not a multiple of threads_per_core, adjusted to %d\n",
 384                         nr_cpu_ids);
 385         }
 386
 387         if (boot_cpuid >= nr_cpu_ids) {
 388                 // Remember boot core for smp_setup_cpu_maps()
 389                 boot_core_hwid = be32_to_cpu(intserv[0]);
 390
 391                 pr_warn("Boot CPU %d (core hwid %d) >= nr_cpu_ids, adjusted boot CPU to %d\n",
 392                         boot_cpuid, boot_core_hwid, found_thread);
 393
 394                 // Adjust boot CPU to appear on logical core 0
 395                 boot_cpuid = found_thread;
 396         }
 397
 398         DBG("boot cpu: logical %d physical %d\n", boot_cpuid,
 399             be32_to_cpu(intserv[found_thread]));
 400
 401         /*
 402          * PAPR defines "logical" PVR values for cpus that
 403          * meet various levels of the architecture:
 404          * 0x0f000001   Architecture version 2.04
 405          * 0x0f000002   Architecture version 2.05
 406          * If the cpu-version property in the cpu node contains
 407          * such a value, we call identify_cpu again with the
 408          * logical PVR value in order to use the cpu feature
 409          * bits appropriate for the architecture level.
 410          *
 411          * A POWER6 partition in "POWER6 architected" mode
 412          * uses the 0x0f000002 PVR value; in POWER5+ mode
 413          * it uses 0x0f000001.
 414          *
 415          * If we're using device tree CPU feature discovery then we don't
 416          * support the cpu-version property, and it's the responsibility of the
 417          * firmware/hypervisor to provide the correct feature set for the
 418          * architecture level via the ibm,powerpc-cpu-features binding.
 419          */
 420         if (!dt_cpu_ftrs_in_use()) {
 421                 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
 422                 if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000) {
 423                         identify_cpu(0, be32_to_cpup(prop));
 424                         cpu_version = prop;
 425                 }
 426
 427                 check_cpu_feature_properties(node);
 428                 check_cpu_features(node, "ibm,pa-features", ibm_pa_features,
 429                                    ARRAY_SIZE(ibm_pa_features));
 430                 check_cpu_features(node, "ibm,pi-features", ibm_pi_features,
 431                                    ARRAY_SIZE(ibm_pi_features));
 432         }
 433
 434         identical_pvr_fixup(node);
 435
 436         // We can now add the CPU name & PVR to the hardware description
 437         seq_buf_printf(&ppc_hw_desc, "%s 0x%04lx ", cur_cpu_spec->cpu_name, mfspr(SPRN_PVR));
 438         if (cpu_version)
 439                 seq_buf_printf(&ppc_hw_desc, "0x%04x ", be32_to_cpup(cpu_version));
 440
 441         init_mmu_slb_size(node);
 442
 443 #ifdef CONFIG_PPC64
 444         if (nthreads == 1)
 445                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
 446         else if (!dt_cpu_ftrs_in_use())
 447                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
 448 #endif
 449
 450         return 0;
 451 }
 452
 453 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
 454                                                 const char *uname,
 455                                                 int depth, void *data)
 456 {
 457         const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
 458
 459         /* Use common scan routine to determine if this is the chosen node */
 460         if (early_init_dt_scan_chosen(data) < 0)
 461                 return 0;
 462
 463 #ifdef CONFIG_PPC64
 464         /* check if iommu is forced on or off */
 465         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
 466                 iommu_is_off = 1;
 467         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
 468                 iommu_force_on = 1;
 469 #endif
 470
 471         /* mem=x on the command line is the preferred mechanism */
 472         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
 473         if (lprop)
 474                 memory_limit = *lprop;
 475
 476 #ifdef CONFIG_PPC64
 477         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
 478         if (lprop)
 479                 tce_alloc_start = *lprop;
 480         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
 481         if (lprop)
 482                 tce_alloc_end = *lprop;
 483 #endif
 484
 485 #ifdef CONFIG_CRASH_RESERVE
 486         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
 487         if (lprop)
 488                 crashk_res.start = *lprop;
 489
 490         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
 491         if (lprop)
 492                 crashk_res.end = crashk_res.start + *lprop - 1;
 493 #endif
 494
 495         /* break now */
 496         return 1;
 497 }
 498
 499 /*
 500  * Compare the range against max mem limit and update
 501  * size if it cross the limit.
 502  */
 503
 504 #ifdef CONFIG_SPARSEMEM
 505 static bool __init validate_mem_limit(u64 base, u64 *size)
 506 {
 507         u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
 508
 509         if (base >= max_mem)
 510                 return false;
 511         if ((base + *size) > max_mem)
 512                 *size = max_mem - base;
 513         return true;
 514 }
 515 #else
 516 static bool __init validate_mem_limit(u64 base, u64 *size)
 517 {
 518         return true;
 519 }
 520 #endif
 521
 522 #ifdef CONFIG_PPC_PSERIES
 523 /*
 524  * Interpret the ibm dynamic reconfiguration memory LMBs.
 525  * This contains a list of memory blocks along with NUMA affinity
 526  * information.
 527  */
 528 static int  __init early_init_drmem_lmb(struct drmem_lmb *lmb,
 529                                         const __be32 **usm,
 530                                         void *data)
 531 {
 532         u64 base, size;
 533         int is_kexec_kdump = 0, rngs;
 534
 535         base = lmb->base_addr;
 536         size = drmem_lmb_size();
 537         rngs = 1;
 538
 539         /*
 540          * Skip this block if the reserved bit is set in flags
 541          * or if the block is not assigned to this partition.
 542          */
 543         if ((lmb->flags & DRCONF_MEM_RESERVED) ||
 544             !(lmb->flags & DRCONF_MEM_ASSIGNED))
 545                 return 0;
 546
 547         if (*usm)
 548                 is_kexec_kdump = 1;
 549
 550         if (is_kexec_kdump) {
 551                 /*
 552                  * For each memblock in ibm,dynamic-memory, a
 553                  * corresponding entry in linux,drconf-usable-memory
 554                  * property contains a counter 'p' followed by 'p'
 555                  * (base, size) duple. Now read the counter from
 556                  * linux,drconf-usable-memory property
 557                  */
 558                 rngs = dt_mem_next_cell(dt_root_size_cells, usm);
 559                 if (!rngs) /* there are no (base, size) duple */
 560                         return 0;
 561         }
 562
 563         do {
 564                 if (is_kexec_kdump) {
 565                         base = dt_mem_next_cell(dt_root_addr_cells, usm);
 566                         size = dt_mem_next_cell(dt_root_size_cells, usm);
 567                 }
 568
 569                 if (iommu_is_off) {
 570                         if (base >= 0x80000000ul)
 571                                 continue;
 572                         if ((base + size) > 0x80000000ul)
 573                                 size = 0x80000000ul - base;
 574                 }
 575
 576                 if (!validate_mem_limit(base, &size))
 577                         continue;
 578
 579                 DBG("Adding: %llx -> %llx\n", base, size);
 580                 memblock_add(base, size);
 581
 582                 if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
 583                         memblock_mark_hotplug(base, size);
 584         } while (--rngs);
 585
 586         return 0;
 587 }
 588 #endif /* CONFIG_PPC_PSERIES */
 589
 590 static int __init early_init_dt_scan_memory_ppc(void)
 591 {
 592 #ifdef CONFIG_PPC_PSERIES
 593         const void *fdt = initial_boot_params;
 594         int node = fdt_path_offset(fdt, "/ibm,dynamic-reconfiguration-memory");
 595
 596         if (node > 0)
 597                 walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
 598
 599 #endif
 600
 601         return early_init_dt_scan_memory();
 602 }
 603
 604 /*
 605  * For a relocatable kernel, we need to get the memstart_addr first,
 606  * then use it to calculate the virtual kernel start address. This has
 607  * to happen at a very early stage (before machine_init). In this case,
 608  * we just want to get the memstart_address and would not like to mess the
 609  * memblock at this stage. So introduce a variable to skip the memblock_add()
 610  * for this reason.
 611  */
 612 #ifdef CONFIG_RELOCATABLE
 613 static int add_mem_to_memblock = 1;
 614 #else
 615 #define add_mem_to_memblock 1
 616 #endif
 617
 618 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
 619 {
 620 #ifdef CONFIG_PPC64
 621         if (iommu_is_off) {
 622                 if (base >= 0x80000000ul)
 623                         return;
 624                 if ((base + size) > 0x80000000ul)
 625                         size = 0x80000000ul - base;
 626         }
 627 #endif
 628         /* Keep track of the beginning of memory -and- the size of
 629          * the very first block in the device-tree as it represents
 630          * the RMA on ppc64 server
 631          */
 632         if (base < memstart_addr) {
 633                 memstart_addr = base;
 634                 first_memblock_size = size;
 635         }
 636
 637         /* Add the chunk to the MEMBLOCK list */
 638         if (add_mem_to_memblock) {
 639                 if (validate_mem_limit(base, &size))
 640                         memblock_add(base, size);
 641         }
 642 }
 643
 644 static void __init early_reserve_mem_dt(void)
 645 {
 646         unsigned long i, dt_root;
 647         int len;
 648         const __be32 *prop;
 649
 650         early_init_fdt_reserve_self();
 651         early_init_fdt_scan_reserved_mem();
 652
 653         dt_root = of_get_flat_dt_root();
 654
 655         prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
 656
 657         if (!prop)
 658                 return;
 659
 660         DBG("Found new-style reserved-ranges\n");
 661
 662         /* Each reserved range is an (address,size) pair, 2 cells each,
 663          * totalling 4 cells per range. */
 664         for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
 665                 u64 base, size;
 666
 667                 base = of_read_number(prop + (i * 4) + 0, 2);
 668                 size = of_read_number(prop + (i * 4) + 2, 2);
 669
 670                 if (size) {
 671                         DBG("reserving: %llx -> %llx\n", base, size);
 672                         memblock_reserve(base, size);
 673                 }
 674         }
 675 }
 676
 677 static void __init early_reserve_mem(void)
 678 {
 679         __be64 *reserve_map;
 680
 681         reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
 682                         fdt_off_mem_rsvmap(initial_boot_params));
 683
 684         /* Look for the new "reserved-regions" property in the DT */
 685         early_reserve_mem_dt();
 686
 687 #ifdef CONFIG_BLK_DEV_INITRD
 688         /* Then reserve the initrd, if any */
 689         if (initrd_start && (initrd_end > initrd_start)) {
 690                 memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
 691                         ALIGN(initrd_end, PAGE_SIZE) -
 692                         ALIGN_DOWN(initrd_start, PAGE_SIZE));
 693         }
 694 #endif /* CONFIG_BLK_DEV_INITRD */
 695
 696         if (!IS_ENABLED(CONFIG_PPC32))
 697                 return;
 698
 699         /*
 700          * Handle the case where we might be booting from an old kexec
 701          * image that setup the mem_rsvmap as pairs of 32-bit values
 702          */
 703         if (be64_to_cpup(reserve_map) > 0xffffffffull) {
 704                 u32 base_32, size_32;
 705                 __be32 *reserve_map_32 = (__be32 *)reserve_map;
 706
 707                 DBG("Found old 32-bit reserve map\n");
 708
 709                 while (1) {
 710                         base_32 = be32_to_cpup(reserve_map_32++);
 711                         size_32 = be32_to_cpup(reserve_map_32++);
 712                         if (size_32 == 0)
 713                                 break;
 714                         DBG("reserving: %x -> %x\n", base_32, size_32);
 715                         memblock_reserve(base_32, size_32);
 716                 }
 717                 return;
 718         }
 719 }
 720
 721 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 722 static bool tm_disabled __initdata;
 723
 724 static int __init parse_ppc_tm(char *str)
 725 {
 726         bool res;
 727
 728         if (kstrtobool(str, &res))
 729                 return -EINVAL;
 730
 731         tm_disabled = !res;
 732
 733         return 0;
 734 }
 735 early_param("ppc_tm", parse_ppc_tm);
 736
 737 static void __init tm_init(void)
 738 {
 739         if (tm_disabled) {
 740                 pr_info("Disabling hardware transactional memory (HTM)\n");
 741                 cur_cpu_spec->cpu_user_features2 &=
 742                         ~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
 743                 cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
 744                 return;
 745         }
 746
 747         pnv_tm_init();
 748 }
 749 #else
 750 static void tm_init(void) { }
 751 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
 752
 753 static int __init
 754 early_init_dt_scan_model(unsigned long node, const char *uname,
 755                          int depth, void *data)
 756 {
 757         const char *prop;
 758
 759         if (depth != 0)
 760                 return 0;
 761
 762         prop = of_get_flat_dt_prop(node, "model", NULL);
 763         if (prop)
 764                 seq_buf_printf(&ppc_hw_desc, "%s ", prop);
 765
 766         /* break now */
 767         return 1;
 768 }
 769
 770 #ifdef CONFIG_PPC64
 771 static void __init save_fscr_to_task(void)
 772 {
 773         /*
 774          * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
 775          * have configured via the device tree features or via __init_FSCR().
 776          * That value will then be propagated to pid 1 (init) and all future
 777          * processes.
 778          */
 779         if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
 780                 init_task.thread.fscr = mfspr(SPRN_FSCR);
 781 }
 782 #else
 783 static inline void save_fscr_to_task(void) {}
 784 #endif
 785
 786
 787 void __init early_init_devtree(void *params)
 788 {
 789         phys_addr_t int_vector_size;
 790
 791         DBG(" -> early_init_devtree(%px)\n", params);
 792
 793         /* Too early to BUG_ON(), do it by hand */
 794         if (!early_init_dt_verify(params, __pa(params)))
 795                 panic("BUG: Failed verifying flat device tree, bad version?");
 796
 797         of_scan_flat_dt(early_init_dt_scan_model, NULL);
 798
 799 #ifdef CONFIG_PPC_RTAS
 800         /* Some machines might need RTAS info for debugging, grab it now. */
 801         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 802 #endif
 803
 804 #ifdef CONFIG_PPC_POWERNV
 805         /* Some machines might need OPAL info for debugging, grab it now. */
 806         of_scan_flat_dt(early_init_dt_scan_opal, NULL);
 807
 808         /* Scan tree for ultravisor feature */
 809         of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
 810 #endif
 811
 812 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
 813         /* scan tree to see if dump is active during last boot */
 814         of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
 815 #endif
 816
 817         /* Retrieve various informations from the /chosen node of the
 818          * device-tree, including the platform type, initrd location and
 819          * size, TCE reserve, and more ...
 820          */
 821         of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
 822
 823         /* Append additional parameters passed for fadump capture kernel */
 824         fadump_append_bootargs();
 825
 826         /* Scan memory nodes and rebuild MEMBLOCKs */
 827         early_init_dt_scan_root();
 828         early_init_dt_scan_memory_ppc();
 829
 830         /*
 831          * As generic code authors expect to be able to use static keys
 832          * in early_param() handlers, we initialize the static keys just
 833          * before parsing early params (it's fine to call jump_label_init()
 834          * more than once).
 835          */
 836         jump_label_init();
 837         parse_early_param();
 838
 839         /* make sure we've parsed cmdline for mem= before this */
 840         if (memory_limit)
 841                 first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
 842         setup_initial_memory_limit(memstart_addr, first_memblock_size);
 843         /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
 844         memblock_reserve(PHYSICAL_START, __pa(_end) - PHYSICAL_START);
 845 #ifdef CONFIG_PPC64
 846         /* If relocatable, reserve at least 32k for interrupt vectors etc. */
 847         int_vector_size = __end_interrupts - _stext;
 848         int_vector_size = max_t(phys_addr_t, SZ_32K, int_vector_size);
 849 #else
 850         /* If relocatable, reserve first 32k for interrupt vectors etc. */
 851         int_vector_size = SZ_32K;
 852 #endif
 853         if (PHYSICAL_START > MEMORY_START)
 854                 memblock_reserve(MEMORY_START, int_vector_size);
 855         reserve_kdump_trampoline();
 856 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
 857         /*
 858          * If we fail to reserve memory for firmware-assisted dump then
 859          * fallback to kexec based kdump.
 860          */
 861         if (fadump_reserve_mem() == 0)
 862 #endif
 863                 reserve_crashkernel();
 864         early_reserve_mem();
 865
 866         if (memory_limit > memblock_phys_mem_size())
 867                 memory_limit = 0;
 868
 869         /* Align down to 16 MB which is large page size with hash page translation */
 870         memory_limit = ALIGN_DOWN(memory_limit ?: memblock_phys_mem_size(), SZ_16M);
 871         memblock_enforce_memory_limit(memory_limit);
 872
 873 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
 874         if (!early_radix_enabled())
 875                 memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
 876 #endif
 877
 878         memblock_allow_resize();
 879         memblock_dump_all();
 880
 881         DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
 882
 883         /* We may need to relocate the flat tree, do it now.
 884          * FIXME .. and the initrd too? */
 885         move_device_tree();
 886
 887         DBG("Scanning CPUs ...\n");
 888
 889         dt_cpu_ftrs_scan();
 890
 891         /* Retrieve CPU related informations from the flat tree
 892          * (altivec support, boot CPU ID, ...)
 893          */
 894         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
 895         if (boot_cpuid < 0) {
 896                 printk("Failed to identify boot CPU !\n");
 897                 BUG();
 898         }
 899
 900         save_fscr_to_task();
 901
 902 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
 903         /* We'll later wait for secondaries to check in; there are
 904          * NCPUS-1 non-boot CPUs  :-)
 905          */
 906         spinning_secondaries = boot_cpu_count - 1;
 907 #endif
 908
 909         mmu_early_init_devtree();
 910
 911         /* Setup param area for passing additional parameters to fadump capture kernel. */
 912         fadump_setup_param_area();
 913
 914 #ifdef CONFIG_PPC_POWERNV
 915         /* Scan and build the list of machine check recoverable ranges */
 916         of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
 917 #endif
 918         epapr_paravirt_early_init();
 919
 920         /* Now try to figure out if we are running on LPAR and so on */
 921         pseries_probe_fw_features();
 922
 923         /*
 924          * Initialize pkey features and default AMR/IAMR values
 925          */
 926         pkey_early_init_devtree();
 927
 928 #ifdef CONFIG_PPC_PS3
 929         /* Identify PS3 firmware */
 930         if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
 931                 powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
 932 #endif
 933
 934         /* If kexec left a PLPKS password in the DT, get it and clear it */
 935         plpks_early_init_devtree();
 936
 937         tm_init();
 938
 939         DBG(" <- early_init_devtree()\n");
 940 }
 941
 942 #ifdef CONFIG_RELOCATABLE
 943 /*
 944  * This function run before early_init_devtree, so we have to init
 945  * initial_boot_params.
 946  */
 947 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
 948 {
 949         /* Setup flat device-tree pointer */
 950         initial_boot_params = params;
 951
 952         /*
 953          * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
 954          * mess the memblock.
 955          */
 956         add_mem_to_memblock = 0;
 957         early_init_dt_scan_root();
 958         early_init_dt_scan_memory_ppc();
 959         add_mem_to_memblock = 1;
 960
 961         if (size)
 962                 *size = first_memblock_size;
 963 }
 964 #endif
 965
 966 /*******
 967  *
 968  * New implementation of the OF "find" APIs, return a refcounted
 969  * object, call of_node_put() when done.  The device tree and list
 970  * are protected by a rw_lock.
 971  *
 972  * Note that property management will need some locking as well,
 973  * this isn't dealt with yet.
 974  *
 975  *******/
 976
 977 /**
 978  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
 979  * @np: device node of the device
 980  *
 981  * This looks for a property "ibm,chip-id" in the node or any
 982  * of its parents and returns its content, or -1 if it cannot
 983  * be found.
 984  */
 985 int of_get_ibm_chip_id(struct device_node *np)
 986 {
 987         of_node_get(np);
 988         while (np) {
 989                 u32 chip_id;
 990
 991                 /*
 992                  * Skiboot may produce memory nodes that contain more than one
 993                  * cell in chip-id, we only read the first one here.
 994                  */
 995                 if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
 996                         of_node_put(np);
 997                         return chip_id;
 998                 }
 999
1000                 np = of_get_next_parent(np);
1001         }
1002         return -1;
1003 }
1004 EXPORT_SYMBOL(of_get_ibm_chip_id);
1005
1006 /**
1007  * cpu_to_chip_id - Return the cpus chip-id
1008  * @cpu: The logical cpu number.
1009  *
1010  * Return the value of the ibm,chip-id property corresponding to the given
1011  * logical cpu number. If the chip-id can not be found, returns -1.
1012  */
1013 int cpu_to_chip_id(int cpu)
1014 {
1015         struct device_node *np;
1016         int ret = -1, idx;
1017
1018         idx = cpu / threads_per_core;
1019         if (chip_id_lookup_table && chip_id_lookup_table[idx] != -1)
1020                 return chip_id_lookup_table[idx];
1021
1022         np = of_get_cpu_node(cpu, NULL);
1023         if (np) {
1024                 ret = of_get_ibm_chip_id(np);
1025                 of_node_put(np);
1026
1027                 if (chip_id_lookup_table)
1028                         chip_id_lookup_table[idx] = ret;
1029         }
1030
1031         return ret;
1032 }
1033 EXPORT_SYMBOL(cpu_to_chip_id);
1034
1035 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
1036 {
1037 #ifdef CONFIG_SMP
1038         /*
1039          * Early firmware scanning must use this rather than
1040          * get_hard_smp_processor_id because we don't have pacas allocated
1041          * until memory topology is discovered.
1042          */
1043         if (cpu_to_phys_id != NULL)
1044                 return (int)phys_id == cpu_to_phys_id[cpu];
1045 #endif
1046
1047         return (int)phys_id == get_hard_smp_processor_id(cpu);
1048 }