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 #undef DEBUG
  14
  15 #include <linux/module.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/lmb.h>
  38 #include <asm/io.h>
  39 #include <asm/kdump.h>
  40 #include <asm/prom.h>
  41 #include <asm/processor.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/smp.h>
  44 #include <asm/elf.h>
  45 #include <asm/machdep.h>
  46 #include <asm/paca.h>
  47 #include <asm/time.h>
  48 #include <asm/cputable.h>
  49 #include <asm/sections.h>
  50 #include <asm/btext.h>
  51 #include <asm/nvram.h>
  52 #include <asm/setup.h>
  53 #include <asm/system.h>
  54 #include <asm/rtas.h>
  55 #include <asm/iommu.h>
  56 #include <asm/serial.h>
  57 #include <asm/cache.h>
  58 #include <asm/page.h>
  59 #include <asm/mmu.h>
  60 #include <asm/firmware.h>
  61 #include <asm/xmon.h>
  62 #include <asm/udbg.h>
  63 #include <asm/kexec.h>
  64 #include <asm/mmu_context.h>
  65
  66 #include "setup.h"
  67
  68 #ifdef DEBUG
  69 #define DBG(fmt...) udbg_printf(fmt)
  70 #else
  71 #define DBG(fmt...)
  72 #endif
  73
  74 int boot_cpuid = 0;
  75 u64 ppc64_pft_size;
  76
  77 /* Pick defaults since we might want to patch instructions
  78  * before we've read this from the device tree.
  79  */
  80 struct ppc64_caches ppc64_caches = {
  81         .dline_size = 0x40,
  82         .log_dline_size = 6,
  83         .iline_size = 0x40,
  84         .log_iline_size = 6
  85 };
  86 EXPORT_SYMBOL_GPL(ppc64_caches);
  87
  88 /*
  89  * These are used in binfmt_elf.c to put aux entries on the stack
  90  * for each elf executable being started.
  91  */
  92 int dcache_bsize;
  93 int icache_bsize;
  94 int ucache_bsize;
  95
  96 #ifdef CONFIG_SMP
  97
  98 static int smt_enabled_cmdline;
  99
 100 /* Look for ibm,smt-enabled OF option */
 101 static void check_smt_enabled(void)
 102 {
 103         struct device_node *dn;
 104         const char *smt_option;
 105
 106         /* Allow the command line to overrule the OF option */
 107         if (smt_enabled_cmdline)
 108                 return;
 109
 110         dn = of_find_node_by_path("/options");
 111
 112         if (dn) {
 113                 smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
 114
 115                 if (smt_option) {
 116                         if (!strcmp(smt_option, "on"))
 117                                 smt_enabled_at_boot = 1;
 118                         else if (!strcmp(smt_option, "off"))
 119                                 smt_enabled_at_boot = 0;
 120                 }
 121         }
 122 }
 123
 124 /* Look for smt-enabled= cmdline option */
 125 static int __init early_smt_enabled(char *p)
 126 {
 127         smt_enabled_cmdline = 1;
 128
 129         if (!p)
 130                 return 0;
 131
 132         if (!strcmp(p, "on") || !strcmp(p, "1"))
 133                 smt_enabled_at_boot = 1;
 134         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 135                 smt_enabled_at_boot = 0;
 136
 137         return 0;
 138 }
 139 early_param("smt-enabled", early_smt_enabled);
 140
 141 #else
 142 #define check_smt_enabled()
 143 #endif /* CONFIG_SMP */
 144
 145 /* Put the paca pointer into r13 and SPRG_PACA */
 146 static void __init setup_paca(struct paca_struct *new_paca)
 147 {
 148         local_paca = new_paca;
 149         mtspr(SPRN_SPRG_PACA, local_paca);
 150 #ifdef CONFIG_PPC_BOOK3E
 151         mtspr(SPRN_SPRG_TLB_EXFRAME, local_paca->extlb);
 152 #endif
 153 }
 154
 155 /*
 156  * Early initialization entry point. This is called by head.S
 157  * with MMU translation disabled. We rely on the "feature" of
 158  * the CPU that ignores the top 2 bits of the address in real
 159  * mode so we can access kernel globals normally provided we
 160  * only toy with things in the RMO region. From here, we do
 161  * some early parsing of the device-tree to setup out LMB
 162  * data structures, and allocate & initialize the hash table
 163  * and segment tables so we can start running with translation
 164  * enabled.
 165  *
 166  * It is this function which will call the probe() callback of
 167  * the various platform types and copy the matching one to the
 168  * global ppc_md structure. Your platform can eventually do
 169  * some very early initializations from the probe() routine, but
 170  * this is not recommended, be very careful as, for example, the
 171  * device-tree is not accessible via normal means at this point.
 172  */
 173
 174 void __init early_setup(unsigned long dt_ptr)
 175 {
 176         /* -------- printk is _NOT_ safe to use here ! ------- */
 177
 178         /* Identify CPU type */
 179         identify_cpu(0, mfspr(SPRN_PVR));
 180
 181         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 182         initialise_paca(&boot_paca, 0);
 183         setup_paca(&boot_paca);
 184
 185         /* Initialize lockdep early or else spinlocks will blow */
 186         lockdep_init();
 187
 188         /* -------- printk is now safe to use ------- */
 189
 190         /* Enable early debugging if any specified (see udbg.h) */
 191         udbg_early_init();
 192
 193         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 194
 195         /*
 196          * Do early initialization using the flattened device
 197          * tree, such as retrieving the physical memory map or
 198          * calculating/retrieving the hash table size.
 199          */
 200         early_init_devtree(__va(dt_ptr));
 201
 202         /* Now we know the logical id of our boot cpu, setup the paca. */
 203         setup_paca(&paca[boot_cpuid]);
 204
 205         /* Fix up paca fields required for the boot cpu */
 206         get_paca()->cpu_start = 1;
 207
 208         /* Probe the machine type */
 209         probe_machine();
 210
 211         setup_kdump_trampoline();
 212
 213         DBG("Found, Initializing memory management...\n");
 214
 215         /* Initialize the hash table or TLB handling */
 216         early_init_mmu();
 217
 218         DBG(" <- early_setup()\n");
 219 }
 220
 221 #ifdef CONFIG_SMP
 222 void early_setup_secondary(void)
 223 {
 224         /* Mark interrupts enabled in PACA */
 225         get_paca()->soft_enabled = 0;
 226
 227         /* Initialize the hash table or TLB handling */
 228         early_init_mmu_secondary();
 229 }
 230
 231 #endif /* CONFIG_SMP */
 232
 233 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 234 void smp_release_cpus(void)
 235 {
 236         unsigned long *ptr;
 237
 238         DBG(" -> smp_release_cpus()\n");
 239
 240         /* All secondary cpus are spinning on a common spinloop, release them
 241          * all now so they can start to spin on their individual paca
 242          * spinloops. For non SMP kernels, the secondary cpus never get out
 243          * of the common spinloop.
 244          */
 245
 246         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 247                         - PHYSICAL_START);
 248         *ptr = __pa(generic_secondary_smp_init);
 249         mb();
 250
 251         DBG(" <- smp_release_cpus()\n");
 252 }
 253 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 254
 255 /*
 256  * Initialize some remaining members of the ppc64_caches and systemcfg
 257  * structures
 258  * (at least until we get rid of them completely). This is mostly some
 259  * cache informations about the CPU that will be used by cache flush
 260  * routines and/or provided to userland
 261  */
 262 static void __init initialize_cache_info(void)
 263 {
 264         struct device_node *np;
 265         unsigned long num_cpus = 0;
 266
 267         DBG(" -> initialize_cache_info()\n");
 268
 269         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 270                 num_cpus += 1;
 271
 272                 /* We're assuming *all* of the CPUs have the same
 273                  * d-cache and i-cache sizes... -Peter
 274                  */
 275
 276                 if ( num_cpus == 1 ) {
 277                         const u32 *sizep, *lsizep;
 278                         u32 size, lsize;
 279
 280                         size = 0;
 281                         lsize = cur_cpu_spec->dcache_bsize;
 282                         sizep = of_get_property(np, "d-cache-size", NULL);
 283                         if (sizep != NULL)
 284                                 size = *sizep;
 285                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
 286                         /* fallback if block size missing */
 287                         if (lsizep == NULL)
 288                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
 289                         if (lsizep != NULL)
 290                                 lsize = *lsizep;
 291                         if (sizep == 0 || lsizep == 0)
 292                                 DBG("Argh, can't find dcache properties ! "
 293                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 294
 295                         ppc64_caches.dsize = size;
 296                         ppc64_caches.dline_size = lsize;
 297                         ppc64_caches.log_dline_size = __ilog2(lsize);
 298                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 299
 300                         size = 0;
 301                         lsize = cur_cpu_spec->icache_bsize;
 302                         sizep = of_get_property(np, "i-cache-size", NULL);
 303                         if (sizep != NULL)
 304                                 size = *sizep;
 305                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
 306                         if (lsizep == NULL)
 307                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
 308                         if (lsizep != NULL)
 309                                 lsize = *lsizep;
 310                         if (sizep == 0 || lsizep == 0)
 311                                 DBG("Argh, can't find icache properties ! "
 312                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 313
 314                         ppc64_caches.isize = size;
 315                         ppc64_caches.iline_size = lsize;
 316                         ppc64_caches.log_iline_size = __ilog2(lsize);
 317                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 318                 }
 319         }
 320
 321         DBG(" <- initialize_cache_info()\n");
 322 }
 323
 324
 325 /*
 326  * Do some initial setup of the system.  The parameters are those which
 327  * were passed in from the bootloader.
 328  */
 329 void __init setup_system(void)
 330 {
 331         DBG(" -> setup_system()\n");
 332
 333         /* Apply the CPUs-specific and firmware specific fixups to kernel
 334          * text (nop out sections not relevant to this CPU or this firmware)
 335          */
 336         do_feature_fixups(cur_cpu_spec->cpu_features,
 337                           &__start___ftr_fixup, &__stop___ftr_fixup);
 338         do_feature_fixups(cur_cpu_spec->mmu_features,
 339                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
 340         do_feature_fixups(powerpc_firmware_features,
 341                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 342         do_lwsync_fixups(cur_cpu_spec->cpu_features,
 343                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
 344
 345         /*
 346          * Unflatten the device-tree passed by prom_init or kexec
 347          */
 348         unflatten_device_tree();
 349
 350         /*
 351          * Fill the ppc64_caches & systemcfg structures with informations
 352          * retrieved from the device-tree.
 353          */
 354         initialize_cache_info();
 355
 356 #ifdef CONFIG_PPC_RTAS
 357         /*
 358          * Initialize RTAS if available
 359          */
 360         rtas_initialize();
 361 #endif /* CONFIG_PPC_RTAS */
 362
 363         /*
 364          * Check if we have an initrd provided via the device-tree
 365          */
 366         check_for_initrd();
 367
 368         /*
 369          * Do some platform specific early initializations, that includes
 370          * setting up the hash table pointers. It also sets up some interrupt-mapping
 371          * related options that will be used by finish_device_tree()
 372          */
 373         if (ppc_md.init_early)
 374                 ppc_md.init_early();
 375
 376         /*
 377          * We can discover serial ports now since the above did setup the
 378          * hash table management for us, thus ioremap works. We do that early
 379          * so that further code can be debugged
 380          */
 381         find_legacy_serial_ports();
 382
 383         /*
 384          * Register early console
 385          */
 386         register_early_udbg_console();
 387
 388         /*
 389          * Initialize xmon
 390          */
 391         xmon_setup();
 392
 393         check_smt_enabled();
 394         smp_setup_cpu_maps();
 395
 396 #ifdef CONFIG_SMP
 397         /* Release secondary cpus out of their spinloops at 0x60 now that
 398          * we can map physical -> logical CPU ids
 399          */
 400         smp_release_cpus();
 401 #endif
 402
 403         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 404
 405         printk("-----------------------------------------------------\n");
 406         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
 407         printk("physicalMemorySize            = 0x%llx\n", lmb_phys_mem_size());
 408         if (ppc64_caches.dline_size != 0x80)
 409                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 410                        ppc64_caches.dline_size);
 411         if (ppc64_caches.iline_size != 0x80)
 412                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 413                        ppc64_caches.iline_size);
 414 #ifdef CONFIG_PPC_STD_MMU_64
 415         if (htab_address)
 416                 printk("htab_address                  = 0x%p\n", htab_address);
 417         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 418 #endif /* CONFIG_PPC_STD_MMU_64 */
 419         if (PHYSICAL_START > 0)
 420                 printk("physical_start                = 0x%llx\n",
 421                        (unsigned long long)PHYSICAL_START);
 422         printk("-----------------------------------------------------\n");
 423
 424         DBG(" <- setup_system()\n");
 425 }
 426
 427 #ifdef CONFIG_IRQSTACKS
 428 static void __init irqstack_early_init(void)
 429 {
 430         unsigned int i;
 431
 432         /*
 433          * interrupt stacks must be under 256MB, we cannot afford to take
 434          * SLB misses on them.
 435          */
 436         for_each_possible_cpu(i) {
 437                 softirq_ctx[i] = (struct thread_info *)
 438                         __va(lmb_alloc_base(THREAD_SIZE,
 439                                             THREAD_SIZE, 0x10000000));
 440                 hardirq_ctx[i] = (struct thread_info *)
 441                         __va(lmb_alloc_base(THREAD_SIZE,
 442                                             THREAD_SIZE, 0x10000000));
 443         }
 444 }
 445 #else
 446 #define irqstack_early_init()
 447 #endif
 448
 449 #ifdef CONFIG_PPC_BOOK3E
 450 static void __init exc_lvl_early_init(void)
 451 {
 452         unsigned int i;
 453
 454         for_each_possible_cpu(i) {
 455                 critirq_ctx[i] = (struct thread_info *)
 456                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 457                 dbgirq_ctx[i] = (struct thread_info *)
 458                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 459                 mcheckirq_ctx[i] = (struct thread_info *)
 460                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 461         }
 462 }
 463 #else
 464 #define exc_lvl_early_init()
 465 #endif
 466
 467 /*
 468  * Stack space used when we detect a bad kernel stack pointer, and
 469  * early in SMP boots before relocation is enabled.
 470  */
 471 static void __init emergency_stack_init(void)
 472 {
 473         unsigned long limit;
 474         unsigned int i;
 475
 476         /*
 477          * Emergency stacks must be under 256MB, we cannot afford to take
 478          * SLB misses on them. The ABI also requires them to be 128-byte
 479          * aligned.
 480          *
 481          * Since we use these as temporary stacks during secondary CPU
 482          * bringup, we need to get at them in real mode. This means they
 483          * must also be within the RMO region.
 484          */
 485         limit = min(0x10000000ULL, lmb.rmo_size);
 486
 487         for_each_possible_cpu(i) {
 488                 unsigned long sp;
 489                 sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
 490                 sp += THREAD_SIZE;
 491                 paca[i].emergency_sp = __va(sp);
 492         }
 493 }
 494
 495 /*
 496  * Called into from start_kernel, after lock_kernel has been called.
 497  * Initializes bootmem, which is unsed to manage page allocation until
 498  * mem_init is called.
 499  */
 500 void __init setup_arch(char **cmdline_p)
 501 {
 502         ppc64_boot_msg(0x12, "Setup Arch");
 503
 504         *cmdline_p = cmd_line;
 505
 506         /*
 507          * Set cache line size based on type of cpu as a default.
 508          * Systems with OF can look in the properties on the cpu node(s)
 509          * for a possibly more accurate value.
 510          */
 511         dcache_bsize = ppc64_caches.dline_size;
 512         icache_bsize = ppc64_caches.iline_size;
 513
 514         /* reboot on panic */
 515         panic_timeout = 180;
 516
 517         if (ppc_md.panic)
 518                 setup_panic();
 519
 520         init_mm.start_code = (unsigned long)_stext;
 521         init_mm.end_code = (unsigned long) _etext;
 522         init_mm.end_data = (unsigned long) _edata;
 523         init_mm.brk = klimit;
 524
 525         irqstack_early_init();
 526         exc_lvl_early_init();
 527         emergency_stack_init();
 528
 529 #ifdef CONFIG_PPC_STD_MMU_64
 530         stabs_alloc();
 531 #endif
 532         /* set up the bootmem stuff with available memory */
 533         do_init_bootmem();
 534         sparse_init();
 535
 536 #ifdef CONFIG_DUMMY_CONSOLE
 537         conswitchp = &dummy_con;
 538 #endif
 539
 540         if (ppc_md.setup_arch)
 541                 ppc_md.setup_arch();
 542
 543         paging_init();
 544
 545         /* Initialize the MMU context management stuff */
 546         mmu_context_init();
 547
 548         ppc64_boot_msg(0x15, "Setup Done");
 549 }
 550
 551
 552 /* ToDo: do something useful if ppc_md is not yet setup. */
 553 #define PPC64_LINUX_FUNCTION 0x0f000000
 554 #define PPC64_IPL_MESSAGE 0xc0000000
 555 #define PPC64_TERM_MESSAGE 0xb0000000
 556
 557 static void ppc64_do_msg(unsigned int src, const char *msg)
 558 {
 559         if (ppc_md.progress) {
 560                 char buf[128];
 561
 562                 sprintf(buf, "%08X\n", src);
 563                 ppc_md.progress(buf, 0);
 564                 snprintf(buf, 128, "%s", msg);
 565                 ppc_md.progress(buf, 0);
 566         }
 567 }
 568
 569 /* Print a boot progress message. */
 570 void ppc64_boot_msg(unsigned int src, const char *msg)
 571 {
 572         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 573         printk("[boot]%04x %s\n", src, msg);
 574 }
 575
 576 void cpu_die(void)
 577 {
 578         if (ppc_md.cpu_die)
 579                 ppc_md.cpu_die();
 580 }
 581
 582 #ifdef CONFIG_SMP
 583 #define PCPU_DYN_SIZE           ()
 584
 585 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
 586 {
 587         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
 588                                     __pa(MAX_DMA_ADDRESS));
 589 }
 590
 591 static void __init pcpu_fc_free(void *ptr, size_t size)
 592 {
 593         free_bootmem(__pa(ptr), size);
 594 }
 595
 596 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
 597 {
 598         if (cpu_to_node(from) == cpu_to_node(to))
 599                 return LOCAL_DISTANCE;
 600         else
 601                 return REMOTE_DISTANCE;
 602 }
 603
 604 void __init setup_per_cpu_areas(void)
 605 {
 606         const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
 607         size_t atom_size;
 608         unsigned long delta;
 609         unsigned int cpu;
 610         int rc;
 611
 612         /*
 613          * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
 614          * to group units.  For larger mappings, use 1M atom which
 615          * should be large enough to contain a number of units.
 616          */
 617         if (mmu_linear_psize == MMU_PAGE_4K)
 618                 atom_size = PAGE_SIZE;
 619         else
 620                 atom_size = 1 << 20;
 621
 622         rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
 623                                     pcpu_fc_alloc, pcpu_fc_free);
 624         if (rc < 0)
 625                 panic("cannot initialize percpu area (err=%d)", rc);
 626
 627         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
 628         for_each_possible_cpu(cpu)
 629                 paca[cpu].data_offset = delta + pcpu_unit_offsets[cpu];
 630 }
 631 #endif
 632
 633
 634 #ifdef CONFIG_PPC_INDIRECT_IO
 635 struct ppc_pci_io ppc_pci_io;
 636 EXPORT_SYMBOL(ppc_pci_io);
 637 #endif /* CONFIG_PPC_INDIRECT_IO */
 638