arch/powerpc/kernel/setup_64.c

   1 /*
   2  *
   3  * Common boot and setup code.
   4  *
   5  * Copyright (C) 2001 PPC64 Team, IBM Corp
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #define DEBUG
  14
  15 #include <linux/export.h>
  16 #include <linux/string.h>
  17 #include <linux/sched.h>
  18 #include <linux/init.h>
  19 #include <linux/kernel.h>
  20 #include <linux/reboot.h>
  21 #include <linux/delay.h>
  22 #include <linux/initrd.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/ioport.h>
  25 #include <linux/console.h>
  26 #include <linux/utsname.h>
  27 #include <linux/tty.h>
  28 #include <linux/root_dev.h>
  29 #include <linux/notifier.h>
  30 #include <linux/cpu.h>
  31 #include <linux/unistd.h>
  32 #include <linux/serial.h>
  33 #include <linux/serial_8250.h>
  34 #include <linux/bootmem.h>
  35 #include <linux/pci.h>
  36 #include <linux/lockdep.h>
  37 #include <linux/memblock.h>
  38 #include <linux/hugetlb.h>
  39
  40 #include <asm/io.h>
  41 #include <asm/kdump.h>
  42 #include <asm/prom.h>
  43 #include <asm/processor.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/smp.h>
  46 #include <asm/elf.h>
  47 #include <asm/machdep.h>
  48 #include <asm/paca.h>
  49 #include <asm/time.h>
  50 #include <asm/cputable.h>
  51 #include <asm/sections.h>
  52 #include <asm/btext.h>
  53 #include <asm/nvram.h>
  54 #include <asm/setup.h>
  55 #include <asm/rtas.h>
  56 #include <asm/iommu.h>
  57 #include <asm/serial.h>
  58 #include <asm/cache.h>
  59 #include <asm/page.h>
  60 #include <asm/mmu.h>
  61 #include <asm/firmware.h>
  62 #include <asm/xmon.h>
  63 #include <asm/udbg.h>
  64 #include <asm/kexec.h>
  65 #include <asm/mmu_context.h>
  66 #include <asm/code-patching.h>
  67 #include <asm/kvm_ppc.h>
  68 #include <asm/hugetlb.h>
  69 #include <asm/epapr_hcalls.h>
  70
  71 #include "setup.h"
  72
  73 #ifdef DEBUG
  74 #define DBG(fmt...) udbg_printf(fmt)
  75 #else
  76 #define DBG(fmt...)
  77 #endif
  78
  79 int boot_cpuid = 0;
  80 int spinning_secondaries;
  81 u64 ppc64_pft_size;
  82
  83 /* Pick defaults since we might want to patch instructions
  84  * before we've read this from the device tree.
  85  */
  86 struct ppc64_caches ppc64_caches = {
  87         .dline_size = 0x40,
  88         .log_dline_size = 6,
  89         .iline_size = 0x40,
  90         .log_iline_size = 6
  91 };
  92 EXPORT_SYMBOL_GPL(ppc64_caches);
  93
  94 /*
  95  * These are used in binfmt_elf.c to put aux entries on the stack
  96  * for each elf executable being started.
  97  */
  98 int dcache_bsize;
  99 int icache_bsize;
 100 int ucache_bsize;
 101
 102 #ifdef CONFIG_SMP
 103
 104 static char *smt_enabled_cmdline;
 105
 106 /* Look for ibm,smt-enabled OF option */
 107 static void check_smt_enabled(void)
 108 {
 109         struct device_node *dn;
 110         const char *smt_option;
 111
 112         /* Default to enabling all threads */
 113         smt_enabled_at_boot = threads_per_core;
 114
 115         /* Allow the command line to overrule the OF option */
 116         if (smt_enabled_cmdline) {
 117                 if (!strcmp(smt_enabled_cmdline, "on"))
 118                         smt_enabled_at_boot = threads_per_core;
 119                 else if (!strcmp(smt_enabled_cmdline, "off"))
 120                         smt_enabled_at_boot = 0;
 121                 else {
 122                         long smt;
 123                         int rc;
 124
 125                         rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
 126                         if (!rc)
 127                                 smt_enabled_at_boot =
 128                                         min(threads_per_core, (int)smt);
 129                 }
 130         } else {
 131                 dn = of_find_node_by_path("/options");
 132                 if (dn) {
 133                         smt_option = of_get_property(dn, "ibm,smt-enabled",
 134                                                      NULL);
 135
 136                         if (smt_option) {
 137                                 if (!strcmp(smt_option, "on"))
 138                                         smt_enabled_at_boot = threads_per_core;
 139                                 else if (!strcmp(smt_option, "off"))
 140                                         smt_enabled_at_boot = 0;
 141                         }
 142
 143                         of_node_put(dn);
 144                 }
 145         }
 146 }
 147
 148 /* Look for smt-enabled= cmdline option */
 149 static int __init early_smt_enabled(char *p)
 150 {
 151         smt_enabled_cmdline = p;
 152         return 0;
 153 }
 154 early_param("smt-enabled", early_smt_enabled);
 155
 156 #else
 157 #define check_smt_enabled()
 158 #endif /* CONFIG_SMP */
 159
 160 /** Fix up paca fields required for the boot cpu */
 161 static void fixup_boot_paca(void)
 162 {
 163         /* The boot cpu is started */
 164         get_paca()->cpu_start = 1;
 165         /* Allow percpu accesses to work until we setup percpu data */
 166         get_paca()->data_offset = 0;
 167 }
 168
 169 /*
 170  * Early initialization entry point. This is called by head.S
 171  * with MMU translation disabled. We rely on the "feature" of
 172  * the CPU that ignores the top 2 bits of the address in real
 173  * mode so we can access kernel globals normally provided we
 174  * only toy with things in the RMO region. From here, we do
 175  * some early parsing of the device-tree to setup out MEMBLOCK
 176  * data structures, and allocate & initialize the hash table
 177  * and segment tables so we can start running with translation
 178  * enabled.
 179  *
 180  * It is this function which will call the probe() callback of
 181  * the various platform types and copy the matching one to the
 182  * global ppc_md structure. Your platform can eventually do
 183  * some very early initializations from the probe() routine, but
 184  * this is not recommended, be very careful as, for example, the
 185  * device-tree is not accessible via normal means at this point.
 186  */
 187
 188 void __init early_setup(unsigned long dt_ptr)
 189 {
 190         static __initdata struct paca_struct boot_paca;
 191
 192         /* -------- printk is _NOT_ safe to use here ! ------- */
 193
 194         /* Identify CPU type */
 195         identify_cpu(0, mfspr(SPRN_PVR));
 196
 197         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 198         initialise_paca(&boot_paca, 0);
 199         setup_paca(&boot_paca);
 200         fixup_boot_paca();
 201
 202         /* Initialize lockdep early or else spinlocks will blow */
 203         lockdep_init();
 204
 205         /* -------- printk is now safe to use ------- */
 206
 207         /* Enable early debugging if any specified (see udbg.h) */
 208         udbg_early_init();
 209
 210         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 211
 212         /*
 213          * Do early initialization using the flattened device
 214          * tree, such as retrieving the physical memory map or
 215          * calculating/retrieving the hash table size.
 216          */
 217         early_init_devtree(__va(dt_ptr));
 218
 219         epapr_paravirt_early_init();
 220
 221         /* Now we know the logical id of our boot cpu, setup the paca. */
 222         setup_paca(&paca[boot_cpuid]);
 223         fixup_boot_paca();
 224
 225         /* Probe the machine type */
 226         probe_machine();
 227
 228         setup_kdump_trampoline();
 229
 230         DBG("Found, Initializing memory management...\n");
 231
 232         /* Initialize the hash table or TLB handling */
 233         early_init_mmu();
 234
 235         kvm_cma_reserve();
 236
 237         /*
 238          * Reserve any gigantic pages requested on the command line.
 239          * memblock needs to have been initialized by the time this is
 240          * called since this will reserve memory.
 241          */
 242         reserve_hugetlb_gpages();
 243
 244         DBG(" <- early_setup()\n");
 245
 246 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
 247         /*
 248          * This needs to be done *last* (after the above DBG() even)
 249          *
 250          * Right after we return from this function, we turn on the MMU
 251          * which means the real-mode access trick that btext does will
 252          * no longer work, it needs to switch to using a real MMU
 253          * mapping. This call will ensure that it does
 254          */
 255         btext_map();
 256 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
 257 }
 258
 259 #ifdef CONFIG_SMP
 260 void early_setup_secondary(void)
 261 {
 262         /* Mark interrupts enabled in PACA */
 263         get_paca()->soft_enabled = 0;
 264
 265         /* Initialize the hash table or TLB handling */
 266         early_init_mmu_secondary();
 267 }
 268
 269 #endif /* CONFIG_SMP */
 270
 271 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 272 void smp_release_cpus(void)
 273 {
 274         unsigned long *ptr;
 275         int i;
 276
 277         DBG(" -> smp_release_cpus()\n");
 278
 279         /* All secondary cpus are spinning on a common spinloop, release them
 280          * all now so they can start to spin on their individual paca
 281          * spinloops. For non SMP kernels, the secondary cpus never get out
 282          * of the common spinloop.
 283          */
 284
 285         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 286                         - PHYSICAL_START);
 287         *ptr = __pa(generic_secondary_smp_init);
 288
 289         /* And wait a bit for them to catch up */
 290         for (i = 0; i < 100000; i++) {
 291                 mb();
 292                 HMT_low();
 293                 if (spinning_secondaries == 0)
 294                         break;
 295                 udelay(1);
 296         }
 297         DBG("spinning_secondaries = %d\n", spinning_secondaries);
 298
 299         DBG(" <- smp_release_cpus()\n");
 300 }
 301 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 302
 303 /*
 304  * Initialize some remaining members of the ppc64_caches and systemcfg
 305  * structures
 306  * (at least until we get rid of them completely). This is mostly some
 307  * cache informations about the CPU that will be used by cache flush
 308  * routines and/or provided to userland
 309  */
 310 static void __init initialize_cache_info(void)
 311 {
 312         struct device_node *np;
 313         unsigned long num_cpus = 0;
 314
 315         DBG(" -> initialize_cache_info()\n");
 316
 317         for_each_node_by_type(np, "cpu") {
 318                 num_cpus += 1;
 319
 320                 /*
 321                  * We're assuming *all* of the CPUs have the same
 322                  * d-cache and i-cache sizes... -Peter
 323                  */
 324                 if (num_cpus == 1) {
 325                         const __be32 *sizep, *lsizep;
 326                         u32 size, lsize;
 327
 328                         size = 0;
 329                         lsize = cur_cpu_spec->dcache_bsize;
 330                         sizep = of_get_property(np, "d-cache-size", NULL);
 331                         if (sizep != NULL)
 332                                 size = be32_to_cpu(*sizep);
 333                         lsizep = of_get_property(np, "d-cache-block-size",
 334                                                  NULL);
 335                         /* fallback if block size missing */
 336                         if (lsizep == NULL)
 337                                 lsizep = of_get_property(np,
 338                                                          "d-cache-line-size",
 339                                                          NULL);
 340                         if (lsizep != NULL)
 341                                 lsize = be32_to_cpu(*lsizep);
 342                         if (sizep == NULL || lsizep == NULL)
 343                                 DBG("Argh, can't find dcache properties ! "
 344                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 345
 346                         ppc64_caches.dsize = size;
 347                         ppc64_caches.dline_size = lsize;
 348                         ppc64_caches.log_dline_size = __ilog2(lsize);
 349                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 350
 351                         size = 0;
 352                         lsize = cur_cpu_spec->icache_bsize;
 353                         sizep = of_get_property(np, "i-cache-size", NULL);
 354                         if (sizep != NULL)
 355                                 size = be32_to_cpu(*sizep);
 356                         lsizep = of_get_property(np, "i-cache-block-size",
 357                                                  NULL);
 358                         if (lsizep == NULL)
 359                                 lsizep = of_get_property(np,
 360                                                          "i-cache-line-size",
 361                                                          NULL);
 362                         if (lsizep != NULL)
 363                                 lsize = be32_to_cpu(*lsizep);
 364                         if (sizep == NULL || lsizep == NULL)
 365                                 DBG("Argh, can't find icache properties ! "
 366                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 367
 368                         ppc64_caches.isize = size;
 369                         ppc64_caches.iline_size = lsize;
 370                         ppc64_caches.log_iline_size = __ilog2(lsize);
 371                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 372                 }
 373         }
 374
 375         DBG(" <- initialize_cache_info()\n");
 376 }
 377
 378
 379 /*
 380  * Do some initial setup of the system.  The parameters are those which
 381  * were passed in from the bootloader.
 382  */
 383 void __init setup_system(void)
 384 {
 385         DBG(" -> setup_system()\n");
 386
 387         /* Apply the CPUs-specific and firmware specific fixups to kernel
 388          * text (nop out sections not relevant to this CPU or this firmware)
 389          */
 390         do_feature_fixups(cur_cpu_spec->cpu_features,
 391                           &__start___ftr_fixup, &__stop___ftr_fixup);
 392         do_feature_fixups(cur_cpu_spec->mmu_features,
 393                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
 394         do_feature_fixups(powerpc_firmware_features,
 395                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 396         do_lwsync_fixups(cur_cpu_spec->cpu_features,
 397                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
 398         do_final_fixups();
 399
 400         /*
 401          * Unflatten the device-tree passed by prom_init or kexec
 402          */
 403         unflatten_device_tree();
 404
 405         /*
 406          * Fill the ppc64_caches & systemcfg structures with informations
 407          * retrieved from the device-tree.
 408          */
 409         initialize_cache_info();
 410
 411 #ifdef CONFIG_PPC_RTAS
 412         /*
 413          * Initialize RTAS if available
 414          */
 415         rtas_initialize();
 416 #endif /* CONFIG_PPC_RTAS */
 417
 418         /*
 419          * Check if we have an initrd provided via the device-tree
 420          */
 421         check_for_initrd();
 422
 423         /*
 424          * Do some platform specific early initializations, that includes
 425          * setting up the hash table pointers. It also sets up some interrupt-mapping
 426          * related options that will be used by finish_device_tree()
 427          */
 428         if (ppc_md.init_early)
 429                 ppc_md.init_early();
 430
 431         /*
 432          * We can discover serial ports now since the above did setup the
 433          * hash table management for us, thus ioremap works. We do that early
 434          * so that further code can be debugged
 435          */
 436         find_legacy_serial_ports();
 437
 438         /*
 439          * Register early console
 440          */
 441         register_early_udbg_console();
 442
 443         /*
 444          * Initialize xmon
 445          */
 446         xmon_setup();
 447
 448         smp_setup_cpu_maps();
 449         check_smt_enabled();
 450
 451 #ifdef CONFIG_SMP
 452         /* Release secondary cpus out of their spinloops at 0x60 now that
 453          * we can map physical -> logical CPU ids
 454          */
 455         smp_release_cpus();
 456 #endif
 457
 458         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 459
 460         printk("-----------------------------------------------------\n");
 461         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
 462         printk("physicalMemorySize            = 0x%llx\n", memblock_phys_mem_size());
 463         if (ppc64_caches.dline_size != 0x80)
 464                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 465                        ppc64_caches.dline_size);
 466         if (ppc64_caches.iline_size != 0x80)
 467                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 468                        ppc64_caches.iline_size);
 469 #ifdef CONFIG_PPC_STD_MMU_64
 470         if (htab_address)
 471                 printk("htab_address                  = 0x%p\n", htab_address);
 472         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 473 #endif /* CONFIG_PPC_STD_MMU_64 */
 474         if (PHYSICAL_START > 0)
 475                 printk("physical_start                = 0x%llx\n",
 476                        (unsigned long long)PHYSICAL_START);
 477         printk("-----------------------------------------------------\n");
 478
 479         DBG(" <- setup_system()\n");
 480 }
 481
 482 /* This returns the limit below which memory accesses to the linear
 483  * mapping are guarnateed not to cause a TLB or SLB miss. This is
 484  * used to allocate interrupt or emergency stacks for which our
 485  * exception entry path doesn't deal with being interrupted.
 486  */
 487 static u64 safe_stack_limit(void)
 488 {
 489 #ifdef CONFIG_PPC_BOOK3E
 490         /* Freescale BookE bolts the entire linear mapping */
 491         if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
 492                 return linear_map_top;
 493         /* Other BookE, we assume the first GB is bolted */
 494         return 1ul << 30;
 495 #else
 496         /* BookS, the first segment is bolted */
 497         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
 498                 return 1UL << SID_SHIFT_1T;
 499         return 1UL << SID_SHIFT;
 500 #endif
 501 }
 502
 503 static void __init irqstack_early_init(void)
 504 {
 505         u64 limit = safe_stack_limit();
 506         unsigned int i;
 507
 508         /*
 509          * Interrupt stacks must be in the first segment since we
 510          * cannot afford to take SLB misses on them.
 511          */
 512         for_each_possible_cpu(i) {
 513                 softirq_ctx[i] = (struct thread_info *)
 514                         __va(memblock_alloc_base(THREAD_SIZE,
 515                                             THREAD_SIZE, limit));
 516                 hardirq_ctx[i] = (struct thread_info *)
 517                         __va(memblock_alloc_base(THREAD_SIZE,
 518                                             THREAD_SIZE, limit));
 519         }
 520 }
 521
 522 #ifdef CONFIG_PPC_BOOK3E
 523 static void __init exc_lvl_early_init(void)
 524 {
 525         extern unsigned int interrupt_base_book3e;
 526         extern unsigned int exc_debug_debug_book3e;
 527
 528         unsigned int i;
 529
 530         for_each_possible_cpu(i) {
 531                 critirq_ctx[i] = (struct thread_info *)
 532                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
 533                 dbgirq_ctx[i] = (struct thread_info *)
 534                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
 535                 mcheckirq_ctx[i] = (struct thread_info *)
 536                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
 537         }
 538
 539         if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
 540                 patch_branch(&interrupt_base_book3e + (0x040 / 4) + 1,
 541                              (unsigned long)&exc_debug_debug_book3e, 0);
 542 }
 543 #else
 544 #define exc_lvl_early_init()
 545 #endif
 546
 547 /*
 548  * Stack space used when we detect a bad kernel stack pointer, and
 549  * early in SMP boots before relocation is enabled.
 550  */
 551 static void __init emergency_stack_init(void)
 552 {
 553         u64 limit;
 554         unsigned int i;
 555
 556         /*
 557          * Emergency stacks must be under 256MB, we cannot afford to take
 558          * SLB misses on them. The ABI also requires them to be 128-byte
 559          * aligned.
 560          *
 561          * Since we use these as temporary stacks during secondary CPU
 562          * bringup, we need to get at them in real mode. This means they
 563          * must also be within the RMO region.
 564          */
 565         limit = min(safe_stack_limit(), ppc64_rma_size);
 566
 567         for_each_possible_cpu(i) {
 568                 unsigned long sp;
 569                 sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
 570                 sp += THREAD_SIZE;
 571                 paca[i].emergency_sp = __va(sp);
 572         }
 573 }
 574
 575 /*
 576  * Called into from start_kernel this initializes bootmem, which is used
 577  * to manage page allocation until mem_init is called.
 578  */
 579 void __init setup_arch(char **cmdline_p)
 580 {
 581         ppc64_boot_msg(0x12, "Setup Arch");
 582
 583         *cmdline_p = cmd_line;
 584
 585         /*
 586          * Set cache line size based on type of cpu as a default.
 587          * Systems with OF can look in the properties on the cpu node(s)
 588          * for a possibly more accurate value.
 589          */
 590         dcache_bsize = ppc64_caches.dline_size;
 591         icache_bsize = ppc64_caches.iline_size;
 592
 593         /* reboot on panic */
 594         panic_timeout = 180;
 595
 596         if (ppc_md.panic)
 597                 setup_panic();
 598
 599         init_mm.start_code = (unsigned long)_stext;
 600         init_mm.end_code = (unsigned long) _etext;
 601         init_mm.end_data = (unsigned long) _edata;
 602         init_mm.brk = klimit;
 603 #ifdef CONFIG_PPC_64K_PAGES
 604         init_mm.context.pte_frag = NULL;
 605 #endif
 606         irqstack_early_init();
 607         exc_lvl_early_init();
 608         emergency_stack_init();
 609
 610 #ifdef CONFIG_PPC_STD_MMU_64
 611         stabs_alloc();
 612 #endif
 613         /* set up the bootmem stuff with available memory */
 614         do_init_bootmem();
 615         sparse_init();
 616
 617 #ifdef CONFIG_DUMMY_CONSOLE
 618         conswitchp = &dummy_con;
 619 #endif
 620
 621         if (ppc_md.setup_arch)
 622                 ppc_md.setup_arch();
 623
 624         paging_init();
 625
 626         /* Initialize the MMU context management stuff */
 627         mmu_context_init();
 628
 629         /* Interrupt code needs to be 64K-aligned */
 630         if ((unsigned long)_stext & 0xffff)
 631                 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
 632                       (unsigned long)_stext);
 633
 634         ppc64_boot_msg(0x15, "Setup Done");
 635 }
 636
 637
 638 /* ToDo: do something useful if ppc_md is not yet setup. */
 639 #define PPC64_LINUX_FUNCTION 0x0f000000
 640 #define PPC64_IPL_MESSAGE 0xc0000000
 641 #define PPC64_TERM_MESSAGE 0xb0000000
 642
 643 static void ppc64_do_msg(unsigned int src, const char *msg)
 644 {
 645         if (ppc_md.progress) {
 646                 char buf[128];
 647
 648                 sprintf(buf, "%08X\n", src);
 649                 ppc_md.progress(buf, 0);
 650                 snprintf(buf, 128, "%s", msg);
 651                 ppc_md.progress(buf, 0);
 652         }
 653 }
 654
 655 /* Print a boot progress message. */
 656 void ppc64_boot_msg(unsigned int src, const char *msg)
 657 {
 658         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 659         printk("[boot]%04x %s\n", src, msg);
 660 }
 661
 662 #ifdef CONFIG_SMP
 663 #define PCPU_DYN_SIZE           ()
 664
 665 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
 666 {
 667         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
 668                                     __pa(MAX_DMA_ADDRESS));
 669 }
 670
 671 static void __init pcpu_fc_free(void *ptr, size_t size)
 672 {
 673         free_bootmem(__pa(ptr), size);
 674 }
 675
 676 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
 677 {
 678         if (cpu_to_node(from) == cpu_to_node(to))
 679                 return LOCAL_DISTANCE;
 680         else
 681                 return REMOTE_DISTANCE;
 682 }
 683
 684 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
 685 EXPORT_SYMBOL(__per_cpu_offset);
 686
 687 void __init setup_per_cpu_areas(void)
 688 {
 689         const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
 690         size_t atom_size;
 691         unsigned long delta;
 692         unsigned int cpu;
 693         int rc;
 694
 695         /*
 696          * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
 697          * to group units.  For larger mappings, use 1M atom which
 698          * should be large enough to contain a number of units.
 699          */
 700         if (mmu_linear_psize == MMU_PAGE_4K)
 701                 atom_size = PAGE_SIZE;
 702         else
 703                 atom_size = 1 << 20;
 704
 705         rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
 706                                     pcpu_fc_alloc, pcpu_fc_free);
 707         if (rc < 0)
 708                 panic("cannot initialize percpu area (err=%d)", rc);
 709
 710         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
 711         for_each_possible_cpu(cpu) {
 712                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
 713                 paca[cpu].data_offset = __per_cpu_offset[cpu];
 714         }
 715 }
 716 #endif
 717
 718
 719 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
 720 struct ppc_pci_io ppc_pci_io;
 721 EXPORT_SYMBOL(ppc_pci_io);
 722 #endif