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/mmu-hash64.h>
  61 #include <asm/firmware.h>
  62 #include <asm/xmon.h>
  63 #include <asm/udbg.h>
  64 #include <asm/kexec.h>
  65 #include <asm/swiotlb.h>
  66 #include <asm/mmu_context.h>
  67
  68 #include "setup.h"
  69
  70 #ifdef DEBUG
  71 #define DBG(fmt...) udbg_printf(fmt)
  72 #else
  73 #define DBG(fmt...)
  74 #endif
  75
  76 int boot_cpuid = 0;
  77 u64 ppc64_pft_size;
  78
  79 /* Pick defaults since we might want to patch instructions
  80  * before we've read this from the device tree.
  81  */
  82 struct ppc64_caches ppc64_caches = {
  83         .dline_size = 0x40,
  84         .log_dline_size = 6,
  85         .iline_size = 0x40,
  86         .log_iline_size = 6
  87 };
  88 EXPORT_SYMBOL_GPL(ppc64_caches);
  89
  90 /*
  91  * These are used in binfmt_elf.c to put aux entries on the stack
  92  * for each elf executable being started.
  93  */
  94 int dcache_bsize;
  95 int icache_bsize;
  96 int ucache_bsize;
  97
  98 #ifdef CONFIG_SMP
  99
 100 static int smt_enabled_cmdline;
 101
 102 /* Look for ibm,smt-enabled OF option */
 103 static void check_smt_enabled(void)
 104 {
 105         struct device_node *dn;
 106         const char *smt_option;
 107
 108         /* Allow the command line to overrule the OF option */
 109         if (smt_enabled_cmdline)
 110                 return;
 111
 112         dn = of_find_node_by_path("/options");
 113
 114         if (dn) {
 115                 smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
 116
 117                 if (smt_option) {
 118                         if (!strcmp(smt_option, "on"))
 119                                 smt_enabled_at_boot = 1;
 120                         else if (!strcmp(smt_option, "off"))
 121                                 smt_enabled_at_boot = 0;
 122                 }
 123         }
 124 }
 125
 126 /* Look for smt-enabled= cmdline option */
 127 static int __init early_smt_enabled(char *p)
 128 {
 129         smt_enabled_cmdline = 1;
 130
 131         if (!p)
 132                 return 0;
 133
 134         if (!strcmp(p, "on") || !strcmp(p, "1"))
 135                 smt_enabled_at_boot = 1;
 136         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 137                 smt_enabled_at_boot = 0;
 138
 139         return 0;
 140 }
 141 early_param("smt-enabled", early_smt_enabled);
 142
 143 #else
 144 #define check_smt_enabled()
 145 #endif /* CONFIG_SMP */
 146
 147 /* Put the paca pointer into r13 and SPRG_PACA */
 148 void __init setup_paca(int cpu)
 149 {
 150         local_paca = &paca[cpu];
 151         mtspr(SPRN_SPRG_PACA, local_paca);
 152 #ifdef CONFIG_PPC_BOOK3E
 153         mtspr(SPRN_SPRG_TLB_EXFRAME, local_paca->extlb);
 154 #endif
 155 }
 156
 157 /*
 158  * Early initialization entry point. This is called by head.S
 159  * with MMU translation disabled. We rely on the "feature" of
 160  * the CPU that ignores the top 2 bits of the address in real
 161  * mode so we can access kernel globals normally provided we
 162  * only toy with things in the RMO region. From here, we do
 163  * some early parsing of the device-tree to setup out LMB
 164  * data structures, and allocate & initialize the hash table
 165  * and segment tables so we can start running with translation
 166  * enabled.
 167  *
 168  * It is this function which will call the probe() callback of
 169  * the various platform types and copy the matching one to the
 170  * global ppc_md structure. Your platform can eventually do
 171  * some very early initializations from the probe() routine, but
 172  * this is not recommended, be very careful as, for example, the
 173  * device-tree is not accessible via normal means at this point.
 174  */
 175
 176 void __init early_setup(unsigned long dt_ptr)
 177 {
 178         /* -------- printk is _NOT_ safe to use here ! ------- */
 179
 180         /* Fill in any unititialised pacas */
 181         initialise_pacas();
 182
 183         /* Identify CPU type */
 184         identify_cpu(0, mfspr(SPRN_PVR));
 185
 186         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 187         setup_paca(0);
 188
 189         /* Initialize lockdep early or else spinlocks will blow */
 190         lockdep_init();
 191
 192         /* -------- printk is now safe to use ------- */
 193
 194         /* Enable early debugging if any specified (see udbg.h) */
 195         udbg_early_init();
 196
 197         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 198
 199         /*
 200          * Do early initialization using the flattened device
 201          * tree, such as retrieving the physical memory map or
 202          * calculating/retrieving the hash table size.
 203          */
 204         early_init_devtree(__va(dt_ptr));
 205
 206         /* Now we know the logical id of our boot cpu, setup the paca. */
 207         setup_paca(boot_cpuid);
 208
 209         /* Fix up paca fields required for the boot cpu */
 210         get_paca()->cpu_start = 1;
 211
 212         /* Probe the machine type */
 213         probe_machine();
 214
 215         setup_kdump_trampoline();
 216
 217         DBG("Found, Initializing memory management...\n");
 218
 219         /* Initialize the hash table or TLB handling */
 220         early_init_mmu();
 221
 222         DBG(" <- early_setup()\n");
 223 }
 224
 225 #ifdef CONFIG_SMP
 226 void early_setup_secondary(void)
 227 {
 228         /* Mark interrupts enabled in PACA */
 229         get_paca()->soft_enabled = 0;
 230
 231         /* Initialize the hash table or TLB handling */
 232         early_init_mmu_secondary();
 233 }
 234
 235 #endif /* CONFIG_SMP */
 236
 237 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 238 void smp_release_cpus(void)
 239 {
 240         unsigned long *ptr;
 241
 242         DBG(" -> smp_release_cpus()\n");
 243
 244         /* All secondary cpus are spinning on a common spinloop, release them
 245          * all now so they can start to spin on their individual paca
 246          * spinloops. For non SMP kernels, the secondary cpus never get out
 247          * of the common spinloop.
 248          */
 249
 250         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 251                         - PHYSICAL_START);
 252         *ptr = __pa(generic_secondary_smp_init);
 253         mb();
 254
 255         DBG(" <- smp_release_cpus()\n");
 256 }
 257 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 258
 259 /*
 260  * Initialize some remaining members of the ppc64_caches and systemcfg
 261  * structures
 262  * (at least until we get rid of them completely). This is mostly some
 263  * cache informations about the CPU that will be used by cache flush
 264  * routines and/or provided to userland
 265  */
 266 static void __init initialize_cache_info(void)
 267 {
 268         struct device_node *np;
 269         unsigned long num_cpus = 0;
 270
 271         DBG(" -> initialize_cache_info()\n");
 272
 273         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 274                 num_cpus += 1;
 275
 276                 /* We're assuming *all* of the CPUs have the same
 277                  * d-cache and i-cache sizes... -Peter
 278                  */
 279
 280                 if ( num_cpus == 1 ) {
 281                         const u32 *sizep, *lsizep;
 282                         u32 size, lsize;
 283
 284                         size = 0;
 285                         lsize = cur_cpu_spec->dcache_bsize;
 286                         sizep = of_get_property(np, "d-cache-size", NULL);
 287                         if (sizep != NULL)
 288                                 size = *sizep;
 289                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
 290                         /* fallback if block size missing */
 291                         if (lsizep == NULL)
 292                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
 293                         if (lsizep != NULL)
 294                                 lsize = *lsizep;
 295                         if (sizep == 0 || lsizep == 0)
 296                                 DBG("Argh, can't find dcache properties ! "
 297                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 298
 299                         ppc64_caches.dsize = size;
 300                         ppc64_caches.dline_size = lsize;
 301                         ppc64_caches.log_dline_size = __ilog2(lsize);
 302                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 303
 304                         size = 0;
 305                         lsize = cur_cpu_spec->icache_bsize;
 306                         sizep = of_get_property(np, "i-cache-size", NULL);
 307                         if (sizep != NULL)
 308                                 size = *sizep;
 309                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
 310                         if (lsizep == NULL)
 311                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
 312                         if (lsizep != NULL)
 313                                 lsize = *lsizep;
 314                         if (sizep == 0 || lsizep == 0)
 315                                 DBG("Argh, can't find icache properties ! "
 316                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 317
 318                         ppc64_caches.isize = size;
 319                         ppc64_caches.iline_size = lsize;
 320                         ppc64_caches.log_iline_size = __ilog2(lsize);
 321                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 322                 }
 323         }
 324
 325         DBG(" <- initialize_cache_info()\n");
 326 }
 327
 328
 329 /*
 330  * Do some initial setup of the system.  The parameters are those which
 331  * were passed in from the bootloader.
 332  */
 333 void __init setup_system(void)
 334 {
 335         DBG(" -> setup_system()\n");
 336
 337         /* Apply the CPUs-specific and firmware specific fixups to kernel
 338          * text (nop out sections not relevant to this CPU or this firmware)
 339          */
 340         do_feature_fixups(cur_cpu_spec->cpu_features,
 341                           &__start___ftr_fixup, &__stop___ftr_fixup);
 342         do_feature_fixups(cur_cpu_spec->mmu_features,
 343                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
 344         do_feature_fixups(powerpc_firmware_features,
 345                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 346         do_lwsync_fixups(cur_cpu_spec->cpu_features,
 347                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
 348
 349         /*
 350          * Unflatten the device-tree passed by prom_init or kexec
 351          */
 352         unflatten_device_tree();
 353
 354         /*
 355          * Fill the ppc64_caches & systemcfg structures with informations
 356          * retrieved from the device-tree.
 357          */
 358         initialize_cache_info();
 359
 360         /*
 361          * Initialize irq remapping subsystem
 362          */
 363         irq_early_init();
 364
 365 #ifdef CONFIG_PPC_RTAS
 366         /*
 367          * Initialize RTAS if available
 368          */
 369         rtas_initialize();
 370 #endif /* CONFIG_PPC_RTAS */
 371
 372         /*
 373          * Check if we have an initrd provided via the device-tree
 374          */
 375         check_for_initrd();
 376
 377         /*
 378          * Do some platform specific early initializations, that includes
 379          * setting up the hash table pointers. It also sets up some interrupt-mapping
 380          * related options that will be used by finish_device_tree()
 381          */
 382         if (ppc_md.init_early)
 383                 ppc_md.init_early();
 384
 385         /*
 386          * We can discover serial ports now since the above did setup the
 387          * hash table management for us, thus ioremap works. We do that early
 388          * so that further code can be debugged
 389          */
 390         find_legacy_serial_ports();
 391
 392         /*
 393          * Register early console
 394          */
 395         register_early_udbg_console();
 396
 397         /*
 398          * Initialize xmon
 399          */
 400         xmon_setup();
 401
 402         check_smt_enabled();
 403         smp_setup_cpu_maps();
 404
 405 #ifdef CONFIG_SMP
 406         /* Release secondary cpus out of their spinloops at 0x60 now that
 407          * we can map physical -> logical CPU ids
 408          */
 409         smp_release_cpus();
 410 #endif
 411
 412         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 413
 414         printk("-----------------------------------------------------\n");
 415         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
 416         printk("physicalMemorySize            = 0x%llx\n", lmb_phys_mem_size());
 417         if (ppc64_caches.dline_size != 0x80)
 418                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 419                        ppc64_caches.dline_size);
 420         if (ppc64_caches.iline_size != 0x80)
 421                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 422                        ppc64_caches.iline_size);
 423 #ifdef CONFIG_PPC_STD_MMU_64
 424         if (htab_address)
 425                 printk("htab_address                  = 0x%p\n", htab_address);
 426         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 427 #endif /* CONFIG_PPC_STD_MMU_64 */
 428         if (PHYSICAL_START > 0)
 429                 printk("physical_start                = 0x%llx\n",
 430                        (unsigned long long)PHYSICAL_START);
 431         printk("-----------------------------------------------------\n");
 432
 433         DBG(" <- setup_system()\n");
 434 }
 435
 436 #ifdef CONFIG_IRQSTACKS
 437 static void __init irqstack_early_init(void)
 438 {
 439         unsigned int i;
 440
 441         /*
 442          * interrupt stacks must be under 256MB, we cannot afford to take
 443          * SLB misses on them.
 444          */
 445         for_each_possible_cpu(i) {
 446                 softirq_ctx[i] = (struct thread_info *)
 447                         __va(lmb_alloc_base(THREAD_SIZE,
 448                                             THREAD_SIZE, 0x10000000));
 449                 hardirq_ctx[i] = (struct thread_info *)
 450                         __va(lmb_alloc_base(THREAD_SIZE,
 451                                             THREAD_SIZE, 0x10000000));
 452         }
 453 }
 454 #else
 455 #define irqstack_early_init()
 456 #endif
 457
 458 #ifdef CONFIG_PPC_BOOK3E
 459 static void __init exc_lvl_early_init(void)
 460 {
 461         unsigned int i;
 462
 463         for_each_possible_cpu(i) {
 464                 critirq_ctx[i] = (struct thread_info *)
 465                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 466                 dbgirq_ctx[i] = (struct thread_info *)
 467                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 468                 mcheckirq_ctx[i] = (struct thread_info *)
 469                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 470         }
 471 }
 472 #else
 473 #define exc_lvl_early_init()
 474 #endif
 475
 476 /*
 477  * Stack space used when we detect a bad kernel stack pointer, and
 478  * early in SMP boots before relocation is enabled.
 479  */
 480 static void __init emergency_stack_init(void)
 481 {
 482         unsigned long limit;
 483         unsigned int i;
 484
 485         /*
 486          * Emergency stacks must be under 256MB, we cannot afford to take
 487          * SLB misses on them. The ABI also requires them to be 128-byte
 488          * aligned.
 489          *
 490          * Since we use these as temporary stacks during secondary CPU
 491          * bringup, we need to get at them in real mode. This means they
 492          * must also be within the RMO region.
 493          */
 494         limit = min(0x10000000ULL, lmb.rmo_size);
 495
 496         for_each_possible_cpu(i) {
 497                 unsigned long sp;
 498                 sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
 499                 sp += THREAD_SIZE;
 500                 paca[i].emergency_sp = __va(sp);
 501         }
 502 }
 503
 504 /*
 505  * Called into from start_kernel, after lock_kernel has been called.
 506  * Initializes bootmem, which is unsed to manage page allocation until
 507  * mem_init is called.
 508  */
 509 void __init setup_arch(char **cmdline_p)
 510 {
 511         ppc64_boot_msg(0x12, "Setup Arch");
 512
 513         *cmdline_p = cmd_line;
 514
 515         /*
 516          * Set cache line size based on type of cpu as a default.
 517          * Systems with OF can look in the properties on the cpu node(s)
 518          * for a possibly more accurate value.
 519          */
 520         dcache_bsize = ppc64_caches.dline_size;
 521         icache_bsize = ppc64_caches.iline_size;
 522
 523         /* reboot on panic */
 524         panic_timeout = 180;
 525
 526         if (ppc_md.panic)
 527                 setup_panic();
 528
 529         init_mm.start_code = (unsigned long)_stext;
 530         init_mm.end_code = (unsigned long) _etext;
 531         init_mm.end_data = (unsigned long) _edata;
 532         init_mm.brk = klimit;
 533
 534         irqstack_early_init();
 535         exc_lvl_early_init();
 536         emergency_stack_init();
 537
 538 #ifdef CONFIG_PPC_STD_MMU_64
 539         stabs_alloc();
 540 #endif
 541         /* set up the bootmem stuff with available memory */
 542         do_init_bootmem();
 543         sparse_init();
 544
 545 #ifdef CONFIG_DUMMY_CONSOLE
 546         conswitchp = &dummy_con;
 547 #endif
 548
 549         if (ppc_md.setup_arch)
 550                 ppc_md.setup_arch();
 551
 552 #ifdef CONFIG_SWIOTLB
 553         if (ppc_swiotlb_enable)
 554                 swiotlb_init();
 555 #endif
 556
 557         paging_init();
 558
 559         /* Initialize the MMU context management stuff */
 560         mmu_context_init();
 561
 562         ppc64_boot_msg(0x15, "Setup Done");
 563 }
 564
 565
 566 /* ToDo: do something useful if ppc_md is not yet setup. */
 567 #define PPC64_LINUX_FUNCTION 0x0f000000
 568 #define PPC64_IPL_MESSAGE 0xc0000000
 569 #define PPC64_TERM_MESSAGE 0xb0000000
 570
 571 static void ppc64_do_msg(unsigned int src, const char *msg)
 572 {
 573         if (ppc_md.progress) {
 574                 char buf[128];
 575
 576                 sprintf(buf, "%08X\n", src);
 577                 ppc_md.progress(buf, 0);
 578                 snprintf(buf, 128, "%s", msg);
 579                 ppc_md.progress(buf, 0);
 580         }
 581 }
 582
 583 /* Print a boot progress message. */
 584 void ppc64_boot_msg(unsigned int src, const char *msg)
 585 {
 586         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 587         printk("[boot]%04x %s\n", src, msg);
 588 }
 589
 590 void cpu_die(void)
 591 {
 592         if (ppc_md.cpu_die)
 593                 ppc_md.cpu_die();
 594 }
 595
 596 #ifdef CONFIG_SMP
 597 #define PCPU_DYN_SIZE           ()
 598
 599 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
 600 {
 601         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
 602                                     __pa(MAX_DMA_ADDRESS));
 603 }
 604
 605 static void __init pcpu_fc_free(void *ptr, size_t size)
 606 {
 607         free_bootmem(__pa(ptr), size);
 608 }
 609
 610 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
 611 {
 612         if (cpu_to_node(from) == cpu_to_node(to))
 613                 return LOCAL_DISTANCE;
 614         else
 615                 return REMOTE_DISTANCE;
 616 }
 617
 618 void __init setup_per_cpu_areas(void)
 619 {
 620         const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
 621         size_t atom_size;
 622         unsigned long delta;
 623         unsigned int cpu;
 624         int rc;
 625
 626         /*
 627          * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
 628          * to group units.  For larger mappings, use 1M atom which
 629          * should be large enough to contain a number of units.
 630          */
 631         if (mmu_linear_psize == MMU_PAGE_4K)
 632                 atom_size = PAGE_SIZE;
 633         else
 634                 atom_size = 1 << 20;
 635
 636         rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
 637                                     pcpu_fc_alloc, pcpu_fc_free);
 638         if (rc < 0)
 639                 panic("cannot initialize percpu area (err=%d)", rc);
 640
 641         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
 642         for_each_possible_cpu(cpu)
 643                 paca[cpu].data_offset = delta + pcpu_unit_offsets[cpu];
 644 }
 645 #endif
 646
 647
 648 #ifdef CONFIG_PPC_INDIRECT_IO
 649 struct ppc_pci_io ppc_pci_io;
 650 EXPORT_SYMBOL(ppc_pci_io);
 651 #endif /* CONFIG_PPC_INDIRECT_IO */
 652