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/memblock.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 char *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         /* Default to enabling all threads */
 107         smt_enabled_at_boot = threads_per_core;
 108
 109         /* Allow the command line to overrule the OF option */
 110         if (smt_enabled_cmdline) {
 111                 if (!strcmp(smt_enabled_cmdline, "on"))
 112                         smt_enabled_at_boot = threads_per_core;
 113                 else if (!strcmp(smt_enabled_cmdline, "off"))
 114                         smt_enabled_at_boot = 0;
 115                 else {
 116                         long smt;
 117                         int rc;
 118
 119                         rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
 120                         if (!rc)
 121                                 smt_enabled_at_boot =
 122                                         min(threads_per_core, (int)smt);
 123                 }
 124         } else {
 125                 dn = of_find_node_by_path("/options");
 126                 if (dn) {
 127                         smt_option = of_get_property(dn, "ibm,smt-enabled",
 128                                                      NULL);
 129
 130                         if (smt_option) {
 131                                 if (!strcmp(smt_option, "on"))
 132                                         smt_enabled_at_boot = threads_per_core;
 133                                 else if (!strcmp(smt_option, "off"))
 134                                         smt_enabled_at_boot = 0;
 135                         }
 136
 137                         of_node_put(dn);
 138                 }
 139         }
 140 }
 141
 142 /* Look for smt-enabled= cmdline option */
 143 static int __init early_smt_enabled(char *p)
 144 {
 145         smt_enabled_cmdline = p;
 146         return 0;
 147 }
 148 early_param("smt-enabled", early_smt_enabled);
 149
 150 #else
 151 #define check_smt_enabled()
 152 #endif /* CONFIG_SMP */
 153
 154 /*
 155  * Early initialization entry point. This is called by head.S
 156  * with MMU translation disabled. We rely on the "feature" of
 157  * the CPU that ignores the top 2 bits of the address in real
 158  * mode so we can access kernel globals normally provided we
 159  * only toy with things in the RMO region. From here, we do
 160  * some early parsing of the device-tree to setup out MEMBLOCK
 161  * data structures, and allocate & initialize the hash table
 162  * and segment tables so we can start running with translation
 163  * enabled.
 164  *
 165  * It is this function which will call the probe() callback of
 166  * the various platform types and copy the matching one to the
 167  * global ppc_md structure. Your platform can eventually do
 168  * some very early initializations from the probe() routine, but
 169  * this is not recommended, be very careful as, for example, the
 170  * device-tree is not accessible via normal means at this point.
 171  */
 172
 173 void __init early_setup(unsigned long dt_ptr)
 174 {
 175         /* -------- printk is _NOT_ safe to use here ! ------- */
 176
 177         /* Identify CPU type */
 178         identify_cpu(0, mfspr(SPRN_PVR));
 179
 180         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 181         initialise_paca(&boot_paca, 0);
 182         setup_paca(&boot_paca);
 183
 184         /* Initialize lockdep early or else spinlocks will blow */
 185         lockdep_init();
 186
 187         /* -------- printk is now safe to use ------- */
 188
 189         /* Enable early debugging if any specified (see udbg.h) */
 190         udbg_early_init();
 191
 192         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 193
 194         /*
 195          * Do early initialization using the flattened device
 196          * tree, such as retrieving the physical memory map or
 197          * calculating/retrieving the hash table size.
 198          */
 199         early_init_devtree(__va(dt_ptr));
 200
 201         /* Now we know the logical id of our boot cpu, setup the paca. */
 202         setup_paca(&paca[boot_cpuid]);
 203
 204         /* Fix up paca fields required for the boot cpu */
 205         get_paca()->cpu_start = 1;
 206
 207         /* Probe the machine type */
 208         probe_machine();
 209
 210         setup_kdump_trampoline();
 211
 212         DBG("Found, Initializing memory management...\n");
 213
 214         /* Initialize the hash table or TLB handling */
 215         early_init_mmu();
 216
 217         DBG(" <- early_setup()\n");
 218 }
 219
 220 #ifdef CONFIG_SMP
 221 void early_setup_secondary(void)
 222 {
 223         /* Mark interrupts enabled in PACA */
 224         get_paca()->soft_enabled = 0;
 225
 226         /* Initialize the hash table or TLB handling */
 227         early_init_mmu_secondary();
 228 }
 229
 230 #endif /* CONFIG_SMP */
 231
 232 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 233 void smp_release_cpus(void)
 234 {
 235         unsigned long *ptr;
 236
 237         DBG(" -> smp_release_cpus()\n");
 238
 239         /* All secondary cpus are spinning on a common spinloop, release them
 240          * all now so they can start to spin on their individual paca
 241          * spinloops. For non SMP kernels, the secondary cpus never get out
 242          * of the common spinloop.
 243          */
 244
 245         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 246                         - PHYSICAL_START);
 247         *ptr = __pa(generic_secondary_smp_init);
 248         mb();
 249
 250         DBG(" <- smp_release_cpus()\n");
 251 }
 252 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 253
 254 /*
 255  * Initialize some remaining members of the ppc64_caches and systemcfg
 256  * structures
 257  * (at least until we get rid of them completely). This is mostly some
 258  * cache informations about the CPU that will be used by cache flush
 259  * routines and/or provided to userland
 260  */
 261 static void __init initialize_cache_info(void)
 262 {
 263         struct device_node *np;
 264         unsigned long num_cpus = 0;
 265
 266         DBG(" -> initialize_cache_info()\n");
 267
 268         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 269                 num_cpus += 1;
 270
 271                 /* We're assuming *all* of the CPUs have the same
 272                  * d-cache and i-cache sizes... -Peter
 273                  */
 274
 275                 if ( num_cpus == 1 ) {
 276                         const u32 *sizep, *lsizep;
 277                         u32 size, lsize;
 278
 279                         size = 0;
 280                         lsize = cur_cpu_spec->dcache_bsize;
 281                         sizep = of_get_property(np, "d-cache-size", NULL);
 282                         if (sizep != NULL)
 283                                 size = *sizep;
 284                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
 285                         /* fallback if block size missing */
 286                         if (lsizep == NULL)
 287                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
 288                         if (lsizep != NULL)
 289                                 lsize = *lsizep;
 290                         if (sizep == 0 || lsizep == 0)
 291                                 DBG("Argh, can't find dcache properties ! "
 292                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 293
 294                         ppc64_caches.dsize = size;
 295                         ppc64_caches.dline_size = lsize;
 296                         ppc64_caches.log_dline_size = __ilog2(lsize);
 297                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 298
 299                         size = 0;
 300                         lsize = cur_cpu_spec->icache_bsize;
 301                         sizep = of_get_property(np, "i-cache-size", NULL);
 302                         if (sizep != NULL)
 303                                 size = *sizep;
 304                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
 305                         if (lsizep == NULL)
 306                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
 307                         if (lsizep != NULL)
 308                                 lsize = *lsizep;
 309                         if (sizep == 0 || lsizep == 0)
 310                                 DBG("Argh, can't find icache properties ! "
 311                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 312
 313                         ppc64_caches.isize = size;
 314                         ppc64_caches.iline_size = lsize;
 315                         ppc64_caches.log_iline_size = __ilog2(lsize);
 316                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 317                 }
 318         }
 319
 320         DBG(" <- initialize_cache_info()\n");
 321 }
 322
 323
 324 /*
 325  * Do some initial setup of the system.  The parameters are those which
 326  * were passed in from the bootloader.
 327  */
 328 void __init setup_system(void)
 329 {
 330         DBG(" -> setup_system()\n");
 331
 332         /* Apply the CPUs-specific and firmware specific fixups to kernel
 333          * text (nop out sections not relevant to this CPU or this firmware)
 334          */
 335         do_feature_fixups(cur_cpu_spec->cpu_features,
 336                           &__start___ftr_fixup, &__stop___ftr_fixup);
 337         do_feature_fixups(cur_cpu_spec->mmu_features,
 338                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
 339         do_feature_fixups(powerpc_firmware_features,
 340                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 341         do_lwsync_fixups(cur_cpu_spec->cpu_features,
 342                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
 343
 344         /*
 345          * Unflatten the device-tree passed by prom_init or kexec
 346          */
 347         unflatten_device_tree();
 348
 349         /*
 350          * Fill the ppc64_caches & systemcfg structures with informations
 351          * retrieved from the device-tree.
 352          */
 353         initialize_cache_info();
 354
 355 #ifdef CONFIG_PPC_RTAS
 356         /*
 357          * Initialize RTAS if available
 358          */
 359         rtas_initialize();
 360 #endif /* CONFIG_PPC_RTAS */
 361
 362         /*
 363          * Check if we have an initrd provided via the device-tree
 364          */
 365         check_for_initrd();
 366
 367         /*
 368          * Do some platform specific early initializations, that includes
 369          * setting up the hash table pointers. It also sets up some interrupt-mapping
 370          * related options that will be used by finish_device_tree()
 371          */
 372         if (ppc_md.init_early)
 373                 ppc_md.init_early();
 374
 375         /*
 376          * We can discover serial ports now since the above did setup the
 377          * hash table management for us, thus ioremap works. We do that early
 378          * so that further code can be debugged
 379          */
 380         find_legacy_serial_ports();
 381
 382         /*
 383          * Register early console
 384          */
 385         register_early_udbg_console();
 386
 387         /*
 388          * Initialize xmon
 389          */
 390         xmon_setup();
 391
 392         smp_setup_cpu_maps();
 393         check_smt_enabled();
 394
 395 #ifdef CONFIG_SMP
 396         /* Release secondary cpus out of their spinloops at 0x60 now that
 397          * we can map physical -> logical CPU ids
 398          */
 399         smp_release_cpus();
 400 #endif
 401
 402         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 403
 404         printk("-----------------------------------------------------\n");
 405         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
 406         printk("physicalMemorySize            = 0x%llx\n", memblock_phys_mem_size());
 407         if (ppc64_caches.dline_size != 0x80)
 408                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 409                        ppc64_caches.dline_size);
 410         if (ppc64_caches.iline_size != 0x80)
 411                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 412                        ppc64_caches.iline_size);
 413 #ifdef CONFIG_PPC_STD_MMU_64
 414         if (htab_address)
 415                 printk("htab_address                  = 0x%p\n", htab_address);
 416         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 417 #endif /* CONFIG_PPC_STD_MMU_64 */
 418         if (PHYSICAL_START > 0)
 419                 printk("physical_start                = 0x%llx\n",
 420                        (unsigned long long)PHYSICAL_START);
 421         printk("-----------------------------------------------------\n");
 422
 423         DBG(" <- setup_system()\n");
 424 }
 425
 426 static u64 slb0_limit(void)
 427 {
 428         if (cpu_has_feature(CPU_FTR_1T_SEGMENT)) {
 429                 return 1UL << SID_SHIFT_1T;
 430         }
 431         return 1UL << SID_SHIFT;
 432 }
 433
 434 static void __init irqstack_early_init(void)
 435 {
 436         u64 limit = slb0_limit();
 437         unsigned int i;
 438
 439         /*
 440          * interrupt stacks must be under 256MB, we cannot afford to take
 441          * SLB misses on them.
 442          */
 443         for_each_possible_cpu(i) {
 444                 softirq_ctx[i] = (struct thread_info *)
 445                         __va(memblock_alloc_base(THREAD_SIZE,
 446                                             THREAD_SIZE, limit));
 447                 hardirq_ctx[i] = (struct thread_info *)
 448                         __va(memblock_alloc_base(THREAD_SIZE,
 449                                             THREAD_SIZE, limit));
 450         }
 451 }
 452
 453 #ifdef CONFIG_PPC_BOOK3E
 454 static void __init exc_lvl_early_init(void)
 455 {
 456         unsigned int i;
 457
 458         for_each_possible_cpu(i) {
 459                 critirq_ctx[i] = (struct thread_info *)
 460                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
 461                 dbgirq_ctx[i] = (struct thread_info *)
 462                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
 463                 mcheckirq_ctx[i] = (struct thread_info *)
 464                         __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
 465         }
 466 }
 467 #else
 468 #define exc_lvl_early_init()
 469 #endif
 470
 471 /*
 472  * Stack space used when we detect a bad kernel stack pointer, and
 473  * early in SMP boots before relocation is enabled.
 474  */
 475 static void __init emergency_stack_init(void)
 476 {
 477         u64 limit;
 478         unsigned int i;
 479
 480         /*
 481          * Emergency stacks must be under 256MB, we cannot afford to take
 482          * SLB misses on them. The ABI also requires them to be 128-byte
 483          * aligned.
 484          *
 485          * Since we use these as temporary stacks during secondary CPU
 486          * bringup, we need to get at them in real mode. This means they
 487          * must also be within the RMO region.
 488          */
 489         limit = min(slb0_limit(), ppc64_rma_size);
 490
 491         for_each_possible_cpu(i) {
 492                 unsigned long sp;
 493                 sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
 494                 sp += THREAD_SIZE;
 495                 paca[i].emergency_sp = __va(sp);
 496         }
 497 }
 498
 499 /*
 500  * Called into from start_kernel this initializes bootmem, which is used
 501  * to manage page allocation until mem_init is called.
 502  */
 503 void __init setup_arch(char **cmdline_p)
 504 {
 505         ppc64_boot_msg(0x12, "Setup Arch");
 506
 507         *cmdline_p = cmd_line;
 508
 509         /*
 510          * Set cache line size based on type of cpu as a default.
 511          * Systems with OF can look in the properties on the cpu node(s)
 512          * for a possibly more accurate value.
 513          */
 514         dcache_bsize = ppc64_caches.dline_size;
 515         icache_bsize = ppc64_caches.iline_size;
 516
 517         /* reboot on panic */
 518         panic_timeout = 180;
 519
 520         if (ppc_md.panic)
 521                 setup_panic();
 522
 523         init_mm.start_code = (unsigned long)_stext;
 524         init_mm.end_code = (unsigned long) _etext;
 525         init_mm.end_data = (unsigned long) _edata;
 526         init_mm.brk = klimit;
 527
 528         irqstack_early_init();
 529         exc_lvl_early_init();
 530         emergency_stack_init();
 531
 532 #ifdef CONFIG_PPC_STD_MMU_64
 533         stabs_alloc();
 534 #endif
 535         /* set up the bootmem stuff with available memory */
 536         do_init_bootmem();
 537         sparse_init();
 538
 539 #ifdef CONFIG_DUMMY_CONSOLE
 540         conswitchp = &dummy_con;
 541 #endif
 542
 543         if (ppc_md.setup_arch)
 544                 ppc_md.setup_arch();
 545
 546         paging_init();
 547
 548         /* Initialize the MMU context management stuff */
 549         mmu_context_init();
 550
 551         ppc64_boot_msg(0x15, "Setup Done");
 552 }
 553
 554
 555 /* ToDo: do something useful if ppc_md is not yet setup. */
 556 #define PPC64_LINUX_FUNCTION 0x0f000000
 557 #define PPC64_IPL_MESSAGE 0xc0000000
 558 #define PPC64_TERM_MESSAGE 0xb0000000
 559
 560 static void ppc64_do_msg(unsigned int src, const char *msg)
 561 {
 562         if (ppc_md.progress) {
 563                 char buf[128];
 564
 565                 sprintf(buf, "%08X\n", src);
 566                 ppc_md.progress(buf, 0);
 567                 snprintf(buf, 128, "%s", msg);
 568                 ppc_md.progress(buf, 0);
 569         }
 570 }
 571
 572 /* Print a boot progress message. */
 573 void ppc64_boot_msg(unsigned int src, const char *msg)
 574 {
 575         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 576         printk("[boot]%04x %s\n", src, msg);
 577 }
 578
 579 #ifdef CONFIG_SMP
 580 #define PCPU_DYN_SIZE           ()
 581
 582 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
 583 {
 584         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
 585                                     __pa(MAX_DMA_ADDRESS));
 586 }
 587
 588 static void __init pcpu_fc_free(void *ptr, size_t size)
 589 {
 590         free_bootmem(__pa(ptr), size);
 591 }
 592
 593 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
 594 {
 595         if (cpu_to_node(from) == cpu_to_node(to))
 596                 return LOCAL_DISTANCE;
 597         else
 598                 return REMOTE_DISTANCE;
 599 }
 600
 601 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
 602 EXPORT_SYMBOL(__per_cpu_offset);
 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                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
 630                 paca[cpu].data_offset = __per_cpu_offset[cpu];
 631         }
 632 }
 633 #endif
 634
 635
 636 #ifdef CONFIG_PPC_INDIRECT_IO
 637 struct ppc_pci_io ppc_pci_io;
 638 EXPORT_SYMBOL(ppc_pci_io);
 639 #endif /* CONFIG_PPC_INDIRECT_IO */
 640