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/swiotlb.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 SPRG3 */
 146 void __init setup_paca(int cpu)
 147 {
 148         local_paca = &paca[cpu];
 149         mtspr(SPRN_SPRG3, local_paca);
 150 }
 151
 152 /*
 153  * Early initialization entry point. This is called by head.S
 154  * with MMU translation disabled. We rely on the "feature" of
 155  * the CPU that ignores the top 2 bits of the address in real
 156  * mode so we can access kernel globals normally provided we
 157  * only toy with things in the RMO region. From here, we do
 158  * some early parsing of the device-tree to setup out LMB
 159  * data structures, and allocate & initialize the hash table
 160  * and segment tables so we can start running with translation
 161  * enabled.
 162  *
 163  * It is this function which will call the probe() callback of
 164  * the various platform types and copy the matching one to the
 165  * global ppc_md structure. Your platform can eventually do
 166  * some very early initializations from the probe() routine, but
 167  * this is not recommended, be very careful as, for example, the
 168  * device-tree is not accessible via normal means at this point.
 169  */
 170
 171 void __init early_setup(unsigned long dt_ptr)
 172 {
 173         /* -------- printk is _NOT_ safe to use here ! ------- */
 174
 175         /* Fill in any unititialised pacas */
 176         initialise_pacas();
 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         setup_paca(0);
 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(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 extern unsigned long __secondary_hold_spinloop;
 234 extern void generic_secondary_smp_init(void);
 235
 236 void smp_release_cpus(void)
 237 {
 238         unsigned long *ptr;
 239
 240         DBG(" -> smp_release_cpus()\n");
 241
 242         /* All secondary cpus are spinning on a common spinloop, release them
 243          * all now so they can start to spin on their individual paca
 244          * spinloops. For non SMP kernels, the secondary cpus never get out
 245          * of the common spinloop.
 246          */
 247
 248         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 249                         - PHYSICAL_START);
 250         *ptr = __pa(generic_secondary_smp_init);
 251         mb();
 252
 253         DBG(" <- smp_release_cpus()\n");
 254 }
 255 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 256
 257 /*
 258  * Initialize some remaining members of the ppc64_caches and systemcfg
 259  * structures
 260  * (at least until we get rid of them completely). This is mostly some
 261  * cache informations about the CPU that will be used by cache flush
 262  * routines and/or provided to userland
 263  */
 264 static void __init initialize_cache_info(void)
 265 {
 266         struct device_node *np;
 267         unsigned long num_cpus = 0;
 268
 269         DBG(" -> initialize_cache_info()\n");
 270
 271         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 272                 num_cpus += 1;
 273
 274                 /* We're assuming *all* of the CPUs have the same
 275                  * d-cache and i-cache sizes... -Peter
 276                  */
 277
 278                 if ( num_cpus == 1 ) {
 279                         const u32 *sizep, *lsizep;
 280                         u32 size, lsize;
 281
 282                         size = 0;
 283                         lsize = cur_cpu_spec->dcache_bsize;
 284                         sizep = of_get_property(np, "d-cache-size", NULL);
 285                         if (sizep != NULL)
 286                                 size = *sizep;
 287                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
 288                         /* fallback if block size missing */
 289                         if (lsizep == NULL)
 290                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
 291                         if (lsizep != NULL)
 292                                 lsize = *lsizep;
 293                         if (sizep == 0 || lsizep == 0)
 294                                 DBG("Argh, can't find dcache properties ! "
 295                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 296
 297                         ppc64_caches.dsize = size;
 298                         ppc64_caches.dline_size = lsize;
 299                         ppc64_caches.log_dline_size = __ilog2(lsize);
 300                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 301
 302                         size = 0;
 303                         lsize = cur_cpu_spec->icache_bsize;
 304                         sizep = of_get_property(np, "i-cache-size", NULL);
 305                         if (sizep != NULL)
 306                                 size = *sizep;
 307                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
 308                         if (lsizep == NULL)
 309                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
 310                         if (lsizep != NULL)
 311                                 lsize = *lsizep;
 312                         if (sizep == 0 || lsizep == 0)
 313                                 DBG("Argh, can't find icache properties ! "
 314                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 315
 316                         ppc64_caches.isize = size;
 317                         ppc64_caches.iline_size = lsize;
 318                         ppc64_caches.log_iline_size = __ilog2(lsize);
 319                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 320                 }
 321         }
 322
 323         DBG(" <- initialize_cache_info()\n");
 324 }
 325
 326
 327 /*
 328  * Do some initial setup of the system.  The parameters are those which
 329  * were passed in from the bootloader.
 330  */
 331 void __init setup_system(void)
 332 {
 333         DBG(" -> setup_system()\n");
 334
 335         /* Apply the CPUs-specific and firmware specific fixups to kernel
 336          * text (nop out sections not relevant to this CPU or this firmware)
 337          */
 338         do_feature_fixups(cur_cpu_spec->cpu_features,
 339                           &__start___ftr_fixup, &__stop___ftr_fixup);
 340         do_feature_fixups(cur_cpu_spec->mmu_features,
 341                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
 342         do_feature_fixups(powerpc_firmware_features,
 343                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 344         do_lwsync_fixups(cur_cpu_spec->cpu_features,
 345                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
 346
 347         /*
 348          * Unflatten the device-tree passed by prom_init or kexec
 349          */
 350         unflatten_device_tree();
 351
 352         /*
 353          * Fill the ppc64_caches & systemcfg structures with informations
 354          * retrieved from the device-tree.
 355          */
 356         initialize_cache_info();
 357
 358         /*
 359          * Initialize irq remapping subsystem
 360          */
 361         irq_early_init();
 362
 363 #ifdef CONFIG_PPC_RTAS
 364         /*
 365          * Initialize RTAS if available
 366          */
 367         rtas_initialize();
 368 #endif /* CONFIG_PPC_RTAS */
 369
 370         /*
 371          * Check if we have an initrd provided via the device-tree
 372          */
 373         check_for_initrd();
 374
 375         /*
 376          * Do some platform specific early initializations, that includes
 377          * setting up the hash table pointers. It also sets up some interrupt-mapping
 378          * related options that will be used by finish_device_tree()
 379          */
 380         if (ppc_md.init_early)
 381                 ppc_md.init_early();
 382
 383         /*
 384          * We can discover serial ports now since the above did setup the
 385          * hash table management for us, thus ioremap works. We do that early
 386          * so that further code can be debugged
 387          */
 388         find_legacy_serial_ports();
 389
 390         /*
 391          * Register early console
 392          */
 393         register_early_udbg_console();
 394
 395         /*
 396          * Initialize xmon
 397          */
 398         xmon_setup();
 399
 400         check_smt_enabled();
 401         smp_setup_cpu_maps();
 402
 403 #ifdef CONFIG_SMP
 404         /* Release secondary cpus out of their spinloops at 0x60 now that
 405          * we can map physical -> logical CPU ids
 406          */
 407         smp_release_cpus();
 408 #endif
 409
 410         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 411
 412         printk("-----------------------------------------------------\n");
 413         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
 414         printk("physicalMemorySize            = 0x%llx\n", lmb_phys_mem_size());
 415         if (ppc64_caches.dline_size != 0x80)
 416                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 417                        ppc64_caches.dline_size);
 418         if (ppc64_caches.iline_size != 0x80)
 419                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 420                        ppc64_caches.iline_size);
 421 #ifdef CONFIG_PPC_STD_MMU_64
 422         if (htab_address)
 423                 printk("htab_address                  = 0x%p\n", htab_address);
 424         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 425 #endif /* CONFIG_PPC_STD_MMU_64 */
 426         if (PHYSICAL_START > 0)
 427                 printk("physical_start                = 0x%llx\n",
 428                        (unsigned long long)PHYSICAL_START);
 429         printk("-----------------------------------------------------\n");
 430
 431         DBG(" <- setup_system()\n");
 432 }
 433
 434 #ifdef CONFIG_IRQSTACKS
 435 static void __init irqstack_early_init(void)
 436 {
 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(lmb_alloc_base(THREAD_SIZE,
 446                                             THREAD_SIZE, 0x10000000));
 447                 hardirq_ctx[i] = (struct thread_info *)
 448                         __va(lmb_alloc_base(THREAD_SIZE,
 449                                             THREAD_SIZE, 0x10000000));
 450         }
 451 }
 452 #else
 453 #define irqstack_early_init()
 454 #endif
 455
 456 /*
 457  * Stack space used when we detect a bad kernel stack pointer, and
 458  * early in SMP boots before relocation is enabled.
 459  */
 460 static void __init emergency_stack_init(void)
 461 {
 462         unsigned long limit;
 463         unsigned int i;
 464
 465         /*
 466          * Emergency stacks must be under 256MB, we cannot afford to take
 467          * SLB misses on them. The ABI also requires them to be 128-byte
 468          * aligned.
 469          *
 470          * Since we use these as temporary stacks during secondary CPU
 471          * bringup, we need to get at them in real mode. This means they
 472          * must also be within the RMO region.
 473          */
 474         limit = min(0x10000000ULL, lmb.rmo_size);
 475
 476         for_each_possible_cpu(i) {
 477                 unsigned long sp;
 478                 sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
 479                 sp += THREAD_SIZE;
 480                 paca[i].emergency_sp = __va(sp);
 481         }
 482 }
 483
 484 /*
 485  * Called into from start_kernel, after lock_kernel has been called.
 486  * Initializes bootmem, which is unsed to manage page allocation until
 487  * mem_init is called.
 488  */
 489 void __init setup_arch(char **cmdline_p)
 490 {
 491         ppc64_boot_msg(0x12, "Setup Arch");
 492
 493         *cmdline_p = cmd_line;
 494
 495         /*
 496          * Set cache line size based on type of cpu as a default.
 497          * Systems with OF can look in the properties on the cpu node(s)
 498          * for a possibly more accurate value.
 499          */
 500         dcache_bsize = ppc64_caches.dline_size;
 501         icache_bsize = ppc64_caches.iline_size;
 502
 503         /* reboot on panic */
 504         panic_timeout = 180;
 505
 506         if (ppc_md.panic)
 507                 setup_panic();
 508
 509         init_mm.start_code = (unsigned long)_stext;
 510         init_mm.end_code = (unsigned long) _etext;
 511         init_mm.end_data = (unsigned long) _edata;
 512         init_mm.brk = klimit;
 513
 514         irqstack_early_init();
 515         emergency_stack_init();
 516
 517 #ifdef CONFIG_PPC_STD_MMU_64
 518         stabs_alloc();
 519 #endif
 520         /* set up the bootmem stuff with available memory */
 521         do_init_bootmem();
 522         sparse_init();
 523
 524 #ifdef CONFIG_DUMMY_CONSOLE
 525         conswitchp = &dummy_con;
 526 #endif
 527
 528         if (ppc_md.setup_arch)
 529                 ppc_md.setup_arch();
 530
 531 #ifdef CONFIG_SWIOTLB
 532         if (ppc_swiotlb_enable)
 533                 swiotlb_init();
 534 #endif
 535
 536         paging_init();
 537         ppc64_boot_msg(0x15, "Setup Done");
 538 }
 539
 540
 541 /* ToDo: do something useful if ppc_md is not yet setup. */
 542 #define PPC64_LINUX_FUNCTION 0x0f000000
 543 #define PPC64_IPL_MESSAGE 0xc0000000
 544 #define PPC64_TERM_MESSAGE 0xb0000000
 545
 546 static void ppc64_do_msg(unsigned int src, const char *msg)
 547 {
 548         if (ppc_md.progress) {
 549                 char buf[128];
 550
 551                 sprintf(buf, "%08X\n", src);
 552                 ppc_md.progress(buf, 0);
 553                 snprintf(buf, 128, "%s", msg);
 554                 ppc_md.progress(buf, 0);
 555         }
 556 }
 557
 558 /* Print a boot progress message. */
 559 void ppc64_boot_msg(unsigned int src, const char *msg)
 560 {
 561         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 562         printk("[boot]%04x %s\n", src, msg);
 563 }
 564
 565 void cpu_die(void)
 566 {
 567         if (ppc_md.cpu_die)
 568                 ppc_md.cpu_die();
 569 }
 570
 571 #ifdef CONFIG_SMP
 572 void __init setup_per_cpu_areas(void)
 573 {
 574         int i;
 575         unsigned long size;
 576         char *ptr;
 577
 578         /* Copy section for each CPU (we discard the original) */
 579         size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
 580 #ifdef CONFIG_MODULES
 581         if (size < PERCPU_ENOUGH_ROOM)
 582                 size = PERCPU_ENOUGH_ROOM;
 583 #endif
 584
 585         for_each_possible_cpu(i) {
 586                 ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 587
 588                 paca[i].data_offset = ptr - __per_cpu_start;
 589                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 590         }
 591 }
 592 #endif
 593
 594
 595 #ifdef CONFIG_PPC_INDIRECT_IO
 596 struct ppc_pci_io ppc_pci_io;
 597 EXPORT_SYMBOL(ppc_pci_io);
 598 #endif /* CONFIG_PPC_INDIRECT_IO */
 599