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/config.h>
  16 #include <linux/module.h>
  17 #include <linux/string.h>
  18 #include <linux/sched.h>
  19 #include <linux/init.h>
  20 #include <linux/kernel.h>
  21 #include <linux/reboot.h>
  22 #include <linux/delay.h>
  23 #include <linux/initrd.h>
  24 #include <linux/ide.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/ioport.h>
  27 #include <linux/console.h>
  28 #include <linux/utsname.h>
  29 #include <linux/tty.h>
  30 #include <linux/root_dev.h>
  31 #include <linux/notifier.h>
  32 #include <linux/cpu.h>
  33 #include <linux/unistd.h>
  34 #include <linux/serial.h>
  35 #include <linux/serial_8250.h>
  36 #include <linux/bootmem.h>
  37 #include <asm/io.h>
  38 #include <asm/kdump.h>
  39 #include <asm/prom.h>
  40 #include <asm/processor.h>
  41 #include <asm/pgtable.h>
  42 #include <asm/smp.h>
  43 #include <asm/elf.h>
  44 #include <asm/machdep.h>
  45 #include <asm/paca.h>
  46 #include <asm/time.h>
  47 #include <asm/cputable.h>
  48 #include <asm/sections.h>
  49 #include <asm/btext.h>
  50 #include <asm/nvram.h>
  51 #include <asm/setup.h>
  52 #include <asm/system.h>
  53 #include <asm/rtas.h>
  54 #include <asm/iommu.h>
  55 #include <asm/serial.h>
  56 #include <asm/cache.h>
  57 #include <asm/page.h>
  58 #include <asm/mmu.h>
  59 #include <asm/lmb.h>
  60 #include <asm/iseries/it_lp_naca.h>
  61 #include <asm/firmware.h>
  62 #include <asm/xmon.h>
  63 #include <asm/udbg.h>
  64 #include <asm/kexec.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 have_of = 1;
  75 int boot_cpuid = 0;
  76 dev_t boot_dev;
  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 = 0x80,
  84         .log_dline_size = 7,
  85         .iline_size = 0x80,
  86         .log_iline_size = 7
  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_MAGIC_SYSRQ
  99 unsigned long SYSRQ_KEY;
 100 #endif /* CONFIG_MAGIC_SYSRQ */
 101
 102
 103 #ifdef CONFIG_SMP
 104
 105 static int smt_enabled_cmdline;
 106
 107 /* Look for ibm,smt-enabled OF option */
 108 static void check_smt_enabled(void)
 109 {
 110         struct device_node *dn;
 111         char *smt_option;
 112
 113         /* Allow the command line to overrule the OF option */
 114         if (smt_enabled_cmdline)
 115                 return;
 116
 117         dn = of_find_node_by_path("/options");
 118
 119         if (dn) {
 120                 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
 121
 122                 if (smt_option) {
 123                         if (!strcmp(smt_option, "on"))
 124                                 smt_enabled_at_boot = 1;
 125                         else if (!strcmp(smt_option, "off"))
 126                                 smt_enabled_at_boot = 0;
 127                 }
 128         }
 129 }
 130
 131 /* Look for smt-enabled= cmdline option */
 132 static int __init early_smt_enabled(char *p)
 133 {
 134         smt_enabled_cmdline = 1;
 135
 136         if (!p)
 137                 return 0;
 138
 139         if (!strcmp(p, "on") || !strcmp(p, "1"))
 140                 smt_enabled_at_boot = 1;
 141         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 142                 smt_enabled_at_boot = 0;
 143
 144         return 0;
 145 }
 146 early_param("smt-enabled", early_smt_enabled);
 147
 148 #else
 149 #define check_smt_enabled()
 150 #endif /* CONFIG_SMP */
 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         /* Enable early debugging if any specified (see udbg.h) */
 174         udbg_early_init();
 175
 176         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 177
 178         /*
 179          * Do early initializations using the flattened device
 180          * tree, like retreiving the physical memory map or
 181          * calculating/retreiving the hash table size
 182          */
 183         early_init_devtree(__va(dt_ptr));
 184
 185         /* Now we know the logical id of our boot cpu, setup the paca. */
 186         setup_boot_paca();
 187
 188         /* Fix up paca fields required for the boot cpu */
 189         get_paca()->cpu_start = 1;
 190         get_paca()->stab_real = __pa((u64)&initial_stab);
 191         get_paca()->stab_addr = (u64)&initial_stab;
 192
 193         /* Probe the machine type */
 194         probe_machine();
 195
 196 #ifdef CONFIG_CRASH_DUMP
 197         kdump_setup();
 198 #endif
 199
 200         DBG("Found, Initializing memory management...\n");
 201
 202         /*
 203          * Initialize the MMU Hash table and create the linear mapping
 204          * of memory. Has to be done before stab/slb initialization as
 205          * this is currently where the page size encoding is obtained
 206          */
 207         htab_initialize();
 208
 209         /*
 210          * Initialize stab / SLB management except on iSeries
 211          */
 212         if (cpu_has_feature(CPU_FTR_SLB))
 213                 slb_initialize();
 214         else if (!firmware_has_feature(FW_FEATURE_ISERIES))
 215                 stab_initialize(get_paca()->stab_real);
 216
 217         DBG(" <- early_setup()\n");
 218 }
 219
 220 #ifdef CONFIG_SMP
 221 void early_setup_secondary(void)
 222 {
 223         struct paca_struct *lpaca = get_paca();
 224
 225         /* Mark enabled in PACA */
 226         lpaca->proc_enabled = 0;
 227
 228         /* Initialize hash table for that CPU */
 229         htab_initialize_secondary();
 230
 231         /* Initialize STAB/SLB. We use a virtual address as it works
 232          * in real mode on pSeries and we want a virutal address on
 233          * iSeries anyway
 234          */
 235         if (cpu_has_feature(CPU_FTR_SLB))
 236                 slb_initialize();
 237         else
 238                 stab_initialize(lpaca->stab_addr);
 239 }
 240
 241 #endif /* CONFIG_SMP */
 242
 243 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 244 void smp_release_cpus(void)
 245 {
 246         extern unsigned long __secondary_hold_spinloop;
 247         unsigned long *ptr;
 248
 249         DBG(" -> smp_release_cpus()\n");
 250
 251         /* All secondary cpus are spinning on a common spinloop, release them
 252          * all now so they can start to spin on their individual paca
 253          * spinloops. For non SMP kernels, the secondary cpus never get out
 254          * of the common spinloop.
 255          * This is useless but harmless on iSeries, secondaries are already
 256          * waiting on their paca spinloops. */
 257
 258         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 259                         - PHYSICAL_START);
 260         *ptr = 1;
 261         mb();
 262
 263         DBG(" <- smp_release_cpus()\n");
 264 }
 265 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 266
 267 /*
 268  * Initialize some remaining members of the ppc64_caches and systemcfg
 269  * structures
 270  * (at least until we get rid of them completely). This is mostly some
 271  * cache informations about the CPU that will be used by cache flush
 272  * routines and/or provided to userland
 273  */
 274 static void __init initialize_cache_info(void)
 275 {
 276         struct device_node *np;
 277         unsigned long num_cpus = 0;
 278
 279         DBG(" -> initialize_cache_info()\n");
 280
 281         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 282                 num_cpus += 1;
 283
 284                 /* We're assuming *all* of the CPUs have the same
 285                  * d-cache and i-cache sizes... -Peter
 286                  */
 287
 288                 if ( num_cpus == 1 ) {
 289                         u32 *sizep, *lsizep;
 290                         u32 size, lsize;
 291                         const char *dc, *ic;
 292
 293                         /* Then read cache informations */
 294                         if (machine_is(powermac)) {
 295                                 dc = "d-cache-block-size";
 296                                 ic = "i-cache-block-size";
 297                         } else {
 298                                 dc = "d-cache-line-size";
 299                                 ic = "i-cache-line-size";
 300                         }
 301
 302                         size = 0;
 303                         lsize = cur_cpu_spec->dcache_bsize;
 304                         sizep = (u32 *)get_property(np, "d-cache-size", NULL);
 305                         if (sizep != NULL)
 306                                 size = *sizep;
 307                         lsizep = (u32 *) get_property(np, dc, NULL);
 308                         if (lsizep != NULL)
 309                                 lsize = *lsizep;
 310                         if (sizep == 0 || lsizep == 0)
 311                                 DBG("Argh, can't find dcache properties ! "
 312                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 313
 314                         ppc64_caches.dsize = size;
 315                         ppc64_caches.dline_size = lsize;
 316                         ppc64_caches.log_dline_size = __ilog2(lsize);
 317                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 318
 319                         size = 0;
 320                         lsize = cur_cpu_spec->icache_bsize;
 321                         sizep = (u32 *)get_property(np, "i-cache-size", NULL);
 322                         if (sizep != NULL)
 323                                 size = *sizep;
 324                         lsizep = (u32 *)get_property(np, ic, NULL);
 325                         if (lsizep != NULL)
 326                                 lsize = *lsizep;
 327                         if (sizep == 0 || lsizep == 0)
 328                                 DBG("Argh, can't find icache properties ! "
 329                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 330
 331                         ppc64_caches.isize = size;
 332                         ppc64_caches.iline_size = lsize;
 333                         ppc64_caches.log_iline_size = __ilog2(lsize);
 334                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 335                 }
 336         }
 337
 338         DBG(" <- initialize_cache_info()\n");
 339 }
 340
 341
 342 /*
 343  * Do some initial setup of the system.  The parameters are those which
 344  * were passed in from the bootloader.
 345  */
 346 void __init setup_system(void)
 347 {
 348         DBG(" -> setup_system()\n");
 349
 350 #ifdef CONFIG_KEXEC
 351         kdump_move_device_tree();
 352 #endif
 353         /*
 354          * Unflatten the device-tree passed by prom_init or kexec
 355          */
 356         unflatten_device_tree();
 357
 358 #ifdef CONFIG_KEXEC
 359         kexec_setup();  /* requires unflattened device tree. */
 360 #endif
 361
 362         /*
 363          * Fill the ppc64_caches & systemcfg structures with informations
 364          * retrieved from the device-tree. Need to be called before
 365          * finish_device_tree() since the later requires some of the
 366          * informations filled up here to properly parse the interrupt
 367          * tree.
 368          * It also sets up the cache line sizes which allows to call
 369          * routines like flush_icache_range (used by the hash init
 370          * later on).
 371          */
 372         initialize_cache_info();
 373
 374 #ifdef CONFIG_PPC_RTAS
 375         /*
 376          * Initialize RTAS if available
 377          */
 378         rtas_initialize();
 379 #endif /* CONFIG_PPC_RTAS */
 380
 381         /*
 382          * Check if we have an initrd provided via the device-tree
 383          */
 384         check_for_initrd();
 385
 386         /*
 387          * Do some platform specific early initializations, that includes
 388          * setting up the hash table pointers. It also sets up some interrupt-mapping
 389          * related options that will be used by finish_device_tree()
 390          */
 391         ppc_md.init_early();
 392
 393         /*
 394          * We can discover serial ports now since the above did setup the
 395          * hash table management for us, thus ioremap works. We do that early
 396          * so that further code can be debugged
 397          */
 398         find_legacy_serial_ports();
 399
 400         /*
 401          * "Finish" the device-tree, that is do the actual parsing of
 402          * some of the properties like the interrupt map
 403          */
 404         finish_device_tree();
 405
 406         /*
 407          * Initialize xmon
 408          */
 409 #ifdef CONFIG_XMON_DEFAULT
 410         xmon_init(1);
 411 #endif
 412         /*
 413          * Register early console
 414          */
 415         register_early_udbg_console();
 416
 417         if (do_early_xmon)
 418                 debugger(NULL);
 419
 420         check_smt_enabled();
 421         smp_setup_cpu_maps();
 422
 423 #ifdef CONFIG_SMP
 424         /* Release secondary cpus out of their spinloops at 0x60 now that
 425          * we can map physical -> logical CPU ids
 426          */
 427         smp_release_cpus();
 428 #endif
 429
 430         printk("Starting Linux PPC64 %s\n", system_utsname.version);
 431
 432         printk("-----------------------------------------------------\n");
 433         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 434         printk("ppc64_interrupt_controller    = 0x%ld\n",
 435                ppc64_interrupt_controller);
 436         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 437         printk("ppc64_caches.dcache_line_size = 0x%x\n",
 438                ppc64_caches.dline_size);
 439         printk("ppc64_caches.icache_line_size = 0x%x\n",
 440                ppc64_caches.iline_size);
 441         printk("htab_address                  = 0x%p\n", htab_address);
 442         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 443 #if PHYSICAL_START > 0
 444         printk("physical_start                = 0x%x\n", PHYSICAL_START);
 445 #endif
 446         printk("-----------------------------------------------------\n");
 447
 448         DBG(" <- setup_system()\n");
 449 }
 450
 451 #ifdef CONFIG_IRQSTACKS
 452 static void __init irqstack_early_init(void)
 453 {
 454         unsigned int i;
 455
 456         /*
 457          * interrupt stacks must be under 256MB, we cannot afford to take
 458          * SLB misses on them.
 459          */
 460         for_each_possible_cpu(i) {
 461                 softirq_ctx[i] = (struct thread_info *)
 462                         __va(lmb_alloc_base(THREAD_SIZE,
 463                                             THREAD_SIZE, 0x10000000));
 464                 hardirq_ctx[i] = (struct thread_info *)
 465                         __va(lmb_alloc_base(THREAD_SIZE,
 466                                             THREAD_SIZE, 0x10000000));
 467         }
 468 }
 469 #else
 470 #define irqstack_early_init()
 471 #endif
 472
 473 /*
 474  * Stack space used when we detect a bad kernel stack pointer, and
 475  * early in SMP boots before relocation is enabled.
 476  */
 477 static void __init emergency_stack_init(void)
 478 {
 479         unsigned long limit;
 480         unsigned int i;
 481
 482         /*
 483          * Emergency stacks must be under 256MB, we cannot afford to take
 484          * SLB misses on them. The ABI also requires them to be 128-byte
 485          * aligned.
 486          *
 487          * Since we use these as temporary stacks during secondary CPU
 488          * bringup, we need to get at them in real mode. This means they
 489          * must also be within the RMO region.
 490          */
 491         limit = min(0x10000000UL, lmb.rmo_size);
 492
 493         for_each_possible_cpu(i)
 494                 paca[i].emergency_sp =
 495                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 496 }
 497
 498 /*
 499  * Called into from start_kernel, after lock_kernel has been called.
 500  * Initializes bootmem, which is unsed to manage page allocation until
 501  * 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 = PAGE_OFFSET;
 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         emergency_stack_init();
 530
 531         stabs_alloc();
 532
 533         /* set up the bootmem stuff with available memory */
 534         do_init_bootmem();
 535         sparse_init();
 536
 537 #ifdef CONFIG_DUMMY_CONSOLE
 538         conswitchp = &dummy_con;
 539 #endif
 540
 541         ppc_md.setup_arch();
 542
 543         paging_init();
 544         ppc64_boot_msg(0x15, "Setup Done");
 545 }
 546
 547
 548 /* ToDo: do something useful if ppc_md is not yet setup. */
 549 #define PPC64_LINUX_FUNCTION 0x0f000000
 550 #define PPC64_IPL_MESSAGE 0xc0000000
 551 #define PPC64_TERM_MESSAGE 0xb0000000
 552
 553 static void ppc64_do_msg(unsigned int src, const char *msg)
 554 {
 555         if (ppc_md.progress) {
 556                 char buf[128];
 557
 558                 sprintf(buf, "%08X\n", src);
 559                 ppc_md.progress(buf, 0);
 560                 snprintf(buf, 128, "%s", msg);
 561                 ppc_md.progress(buf, 0);
 562         }
 563 }
 564
 565 /* Print a boot progress message. */
 566 void ppc64_boot_msg(unsigned int src, const char *msg)
 567 {
 568         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 569         printk("[boot]%04x %s\n", src, msg);
 570 }
 571
 572 /* Print a termination message (print only -- does not stop the kernel) */
 573 void ppc64_terminate_msg(unsigned int src, const char *msg)
 574 {
 575         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 576         printk("[terminate]%04x %s\n", src, msg);
 577 }
 578
 579 void cpu_die(void)
 580 {
 581         if (ppc_md.cpu_die)
 582                 ppc_md.cpu_die();
 583 }
 584
 585 #ifdef CONFIG_SMP
 586 void __init setup_per_cpu_areas(void)
 587 {
 588         int i;
 589         unsigned long size;
 590         char *ptr;
 591
 592         /* Copy section for each CPU (we discard the original) */
 593         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 594 #ifdef CONFIG_MODULES
 595         if (size < PERCPU_ENOUGH_ROOM)
 596                 size = PERCPU_ENOUGH_ROOM;
 597 #endif
 598
 599         for_each_possible_cpu(i) {
 600                 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
 601                 if (!ptr)
 602                         panic("Cannot allocate cpu data for CPU %d\n", i);
 603
 604                 paca[i].data_offset = ptr - __per_cpu_start;
 605                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 606         }
 607 }
 608 #endif