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 /* The main machine-dep calls structure
  99  */
 100 struct machdep_calls ppc_md;
 101 EXPORT_SYMBOL(ppc_md);
 102
 103 #ifdef CONFIG_MAGIC_SYSRQ
 104 unsigned long SYSRQ_KEY;
 105 #endif /* CONFIG_MAGIC_SYSRQ */
 106
 107
 108 static int ppc64_panic_event(struct notifier_block *, unsigned long, void *);
 109 static struct notifier_block ppc64_panic_block = {
 110         .notifier_call = ppc64_panic_event,
 111         .priority = INT_MIN /* may not return; must be done last */
 112 };
 113
 114 #ifdef CONFIG_SMP
 115
 116 static int smt_enabled_cmdline;
 117
 118 /* Look for ibm,smt-enabled OF option */
 119 static void check_smt_enabled(void)
 120 {
 121         struct device_node *dn;
 122         char *smt_option;
 123
 124         /* Allow the command line to overrule the OF option */
 125         if (smt_enabled_cmdline)
 126                 return;
 127
 128         dn = of_find_node_by_path("/options");
 129
 130         if (dn) {
 131                 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
 132
 133                 if (smt_option) {
 134                         if (!strcmp(smt_option, "on"))
 135                                 smt_enabled_at_boot = 1;
 136                         else if (!strcmp(smt_option, "off"))
 137                                 smt_enabled_at_boot = 0;
 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 = 1;
 146
 147         if (!p)
 148                 return 0;
 149
 150         if (!strcmp(p, "on") || !strcmp(p, "1"))
 151                 smt_enabled_at_boot = 1;
 152         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 153                 smt_enabled_at_boot = 0;
 154
 155         return 0;
 156 }
 157 early_param("smt-enabled", early_smt_enabled);
 158
 159 #else
 160 #define check_smt_enabled()
 161 #endif /* CONFIG_SMP */
 162
 163 extern struct machdep_calls pSeries_md;
 164 extern struct machdep_calls pmac_md;
 165 extern struct machdep_calls maple_md;
 166 extern struct machdep_calls cell_md;
 167 extern struct machdep_calls iseries_md;
 168
 169 /* Ultimately, stuff them in an elf section like initcalls... */
 170 static struct machdep_calls __initdata *machines[] = {
 171 #ifdef CONFIG_PPC_PSERIES
 172         &pSeries_md,
 173 #endif /* CONFIG_PPC_PSERIES */
 174 #ifdef CONFIG_PPC_PMAC
 175         &pmac_md,
 176 #endif /* CONFIG_PPC_PMAC */
 177 #ifdef CONFIG_PPC_MAPLE
 178         &maple_md,
 179 #endif /* CONFIG_PPC_MAPLE */
 180 #ifdef CONFIG_PPC_CELL
 181         &cell_md,
 182 #endif
 183 #ifdef CONFIG_PPC_ISERIES
 184         &iseries_md,
 185 #endif
 186         NULL
 187 };
 188
 189 /*
 190  * Early initialization entry point. This is called by head.S
 191  * with MMU translation disabled. We rely on the "feature" of
 192  * the CPU that ignores the top 2 bits of the address in real
 193  * mode so we can access kernel globals normally provided we
 194  * only toy with things in the RMO region. From here, we do
 195  * some early parsing of the device-tree to setup out LMB
 196  * data structures, and allocate & initialize the hash table
 197  * and segment tables so we can start running with translation
 198  * enabled.
 199  *
 200  * It is this function which will call the probe() callback of
 201  * the various platform types and copy the matching one to the
 202  * global ppc_md structure. Your platform can eventually do
 203  * some very early initializations from the probe() routine, but
 204  * this is not recommended, be very careful as, for example, the
 205  * device-tree is not accessible via normal means at this point.
 206  */
 207
 208 void __init early_setup(unsigned long dt_ptr)
 209 {
 210         static struct machdep_calls **mach;
 211
 212         /* Enable early debugging if any specified (see udbg.h) */
 213         udbg_early_init();
 214
 215         DBG(" -> early_setup()\n");
 216
 217         /*
 218          * Do early initializations using the flattened device
 219          * tree, like retreiving the physical memory map or
 220          * calculating/retreiving the hash table size
 221          */
 222         early_init_devtree(__va(dt_ptr));
 223
 224         /* Now we know the logical id of our boot cpu, setup the paca. */
 225         setup_boot_paca();
 226
 227         /* Fix up paca fields required for the boot cpu */
 228         get_paca()->cpu_start = 1;
 229         get_paca()->stab_real = __pa((u64)&initial_stab);
 230         get_paca()->stab_addr = (u64)&initial_stab;
 231
 232         /*
 233          * Iterate all ppc_md structures until we find the proper
 234          * one for the current machine type
 235          */
 236         DBG("Probing machine type for platform %x...\n", _machine);
 237
 238         for (mach = machines; *mach; mach++) {
 239                 if ((*mach)->probe(_machine))
 240                         break;
 241         }
 242         /* What can we do if we didn't find ? */
 243         if (*mach == NULL) {
 244                 DBG("No suitable machine found !\n");
 245                 for (;;);
 246         }
 247         ppc_md = **mach;
 248
 249 #ifdef CONFIG_CRASH_DUMP
 250         kdump_setup();
 251 #endif
 252
 253         DBG("Found, Initializing memory management...\n");
 254
 255         /*
 256          * Initialize the MMU Hash table and create the linear mapping
 257          * of memory. Has to be done before stab/slb initialization as
 258          * this is currently where the page size encoding is obtained
 259          */
 260         htab_initialize();
 261
 262         /*
 263          * Initialize stab / SLB management except on iSeries
 264          */
 265         if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
 266                 if (cpu_has_feature(CPU_FTR_SLB))
 267                         slb_initialize();
 268                 else
 269                         stab_initialize(get_paca()->stab_real);
 270         }
 271
 272         DBG(" <- early_setup()\n");
 273 }
 274
 275 #ifdef CONFIG_SMP
 276 void early_setup_secondary(void)
 277 {
 278         struct paca_struct *lpaca = get_paca();
 279
 280         /* Mark enabled in PACA */
 281         lpaca->proc_enabled = 0;
 282
 283         /* Initialize hash table for that CPU */
 284         htab_initialize_secondary();
 285
 286         /* Initialize STAB/SLB. We use a virtual address as it works
 287          * in real mode on pSeries and we want a virutal address on
 288          * iSeries anyway
 289          */
 290         if (cpu_has_feature(CPU_FTR_SLB))
 291                 slb_initialize();
 292         else
 293                 stab_initialize(lpaca->stab_addr);
 294 }
 295
 296 #endif /* CONFIG_SMP */
 297
 298 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 299 void smp_release_cpus(void)
 300 {
 301         extern unsigned long __secondary_hold_spinloop;
 302         unsigned long *ptr;
 303
 304         DBG(" -> smp_release_cpus()\n");
 305
 306         /* All secondary cpus are spinning on a common spinloop, release them
 307          * all now so they can start to spin on their individual paca
 308          * spinloops. For non SMP kernels, the secondary cpus never get out
 309          * of the common spinloop.
 310          * This is useless but harmless on iSeries, secondaries are already
 311          * waiting on their paca spinloops. */
 312
 313         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 314                         - PHYSICAL_START);
 315         *ptr = 1;
 316         mb();
 317
 318         DBG(" <- smp_release_cpus()\n");
 319 }
 320 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 321
 322 /*
 323  * Initialize some remaining members of the ppc64_caches and systemcfg
 324  * structures
 325  * (at least until we get rid of them completely). This is mostly some
 326  * cache informations about the CPU that will be used by cache flush
 327  * routines and/or provided to userland
 328  */
 329 static void __init initialize_cache_info(void)
 330 {
 331         struct device_node *np;
 332         unsigned long num_cpus = 0;
 333
 334         DBG(" -> initialize_cache_info()\n");
 335
 336         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 337                 num_cpus += 1;
 338
 339                 /* We're assuming *all* of the CPUs have the same
 340                  * d-cache and i-cache sizes... -Peter
 341                  */
 342
 343                 if ( num_cpus == 1 ) {
 344                         u32 *sizep, *lsizep;
 345                         u32 size, lsize;
 346                         const char *dc, *ic;
 347
 348                         /* Then read cache informations */
 349                         if (_machine == PLATFORM_POWERMAC) {
 350                                 dc = "d-cache-block-size";
 351                                 ic = "i-cache-block-size";
 352                         } else {
 353                                 dc = "d-cache-line-size";
 354                                 ic = "i-cache-line-size";
 355                         }
 356
 357                         size = 0;
 358                         lsize = cur_cpu_spec->dcache_bsize;
 359                         sizep = (u32 *)get_property(np, "d-cache-size", NULL);
 360                         if (sizep != NULL)
 361                                 size = *sizep;
 362                         lsizep = (u32 *) get_property(np, dc, NULL);
 363                         if (lsizep != NULL)
 364                                 lsize = *lsizep;
 365                         if (sizep == 0 || lsizep == 0)
 366                                 DBG("Argh, can't find dcache properties ! "
 367                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 368
 369                         ppc64_caches.dsize = size;
 370                         ppc64_caches.dline_size = lsize;
 371                         ppc64_caches.log_dline_size = __ilog2(lsize);
 372                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 373
 374                         size = 0;
 375                         lsize = cur_cpu_spec->icache_bsize;
 376                         sizep = (u32 *)get_property(np, "i-cache-size", NULL);
 377                         if (sizep != NULL)
 378                                 size = *sizep;
 379                         lsizep = (u32 *)get_property(np, ic, NULL);
 380                         if (lsizep != NULL)
 381                                 lsize = *lsizep;
 382                         if (sizep == 0 || lsizep == 0)
 383                                 DBG("Argh, can't find icache properties ! "
 384                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 385
 386                         ppc64_caches.isize = size;
 387                         ppc64_caches.iline_size = lsize;
 388                         ppc64_caches.log_iline_size = __ilog2(lsize);
 389                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 390                 }
 391         }
 392
 393         DBG(" <- initialize_cache_info()\n");
 394 }
 395
 396
 397 /*
 398  * Do some initial setup of the system.  The parameters are those which
 399  * were passed in from the bootloader.
 400  */
 401 void __init setup_system(void)
 402 {
 403         DBG(" -> setup_system()\n");
 404
 405 #ifdef CONFIG_KEXEC
 406         kdump_move_device_tree();
 407 #endif
 408         /*
 409          * Unflatten the device-tree passed by prom_init or kexec
 410          */
 411         unflatten_device_tree();
 412
 413 #ifdef CONFIG_KEXEC
 414         kexec_setup();  /* requires unflattened device tree. */
 415 #endif
 416
 417         /*
 418          * Fill the ppc64_caches & systemcfg structures with informations
 419          * retrieved from the device-tree. Need to be called before
 420          * finish_device_tree() since the later requires some of the
 421          * informations filled up here to properly parse the interrupt
 422          * tree.
 423          * It also sets up the cache line sizes which allows to call
 424          * routines like flush_icache_range (used by the hash init
 425          * later on).
 426          */
 427         initialize_cache_info();
 428
 429 #ifdef CONFIG_PPC_RTAS
 430         /*
 431          * Initialize RTAS if available
 432          */
 433         rtas_initialize();
 434 #endif /* CONFIG_PPC_RTAS */
 435
 436         /*
 437          * Check if we have an initrd provided via the device-tree
 438          */
 439         check_for_initrd();
 440
 441         /*
 442          * Do some platform specific early initializations, that includes
 443          * setting up the hash table pointers. It also sets up some interrupt-mapping
 444          * related options that will be used by finish_device_tree()
 445          */
 446         ppc_md.init_early();
 447
 448         /*
 449          * We can discover serial ports now since the above did setup the
 450          * hash table management for us, thus ioremap works. We do that early
 451          * so that further code can be debugged
 452          */
 453         find_legacy_serial_ports();
 454
 455         /*
 456          * "Finish" the device-tree, that is do the actual parsing of
 457          * some of the properties like the interrupt map
 458          */
 459         finish_device_tree();
 460
 461         /*
 462          * Initialize xmon
 463          */
 464 #ifdef CONFIG_XMON_DEFAULT
 465         xmon_init(1);
 466 #endif
 467         /*
 468          * Register early console
 469          */
 470         register_early_udbg_console();
 471
 472         /* Save unparsed command line copy for /proc/cmdline */
 473         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
 474
 475         parse_early_param();
 476
 477         check_smt_enabled();
 478         smp_setup_cpu_maps();
 479
 480 #ifdef CONFIG_SMP
 481         /* Release secondary cpus out of their spinloops at 0x60 now that
 482          * we can map physical -> logical CPU ids
 483          */
 484         smp_release_cpus();
 485 #endif
 486
 487         printk("Starting Linux PPC64 %s\n", system_utsname.version);
 488
 489         printk("-----------------------------------------------------\n");
 490         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 491         printk("ppc64_interrupt_controller    = 0x%ld\n",
 492                ppc64_interrupt_controller);
 493         printk("platform                      = 0x%x\n", _machine);
 494         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 495         printk("ppc64_caches.dcache_line_size = 0x%x\n",
 496                ppc64_caches.dline_size);
 497         printk("ppc64_caches.icache_line_size = 0x%x\n",
 498                ppc64_caches.iline_size);
 499         printk("htab_address                  = 0x%p\n", htab_address);
 500         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 501 #if PHYSICAL_START > 0
 502         printk("physical_start                = 0x%x\n", PHYSICAL_START);
 503 #endif
 504         printk("-----------------------------------------------------\n");
 505
 506         DBG(" <- setup_system()\n");
 507 }
 508
 509 static int ppc64_panic_event(struct notifier_block *this,
 510                              unsigned long event, void *ptr)
 511 {
 512         ppc_md.panic((char *)ptr);  /* May not return */
 513         return NOTIFY_DONE;
 514 }
 515
 516 #ifdef CONFIG_IRQSTACKS
 517 static void __init irqstack_early_init(void)
 518 {
 519         unsigned int i;
 520
 521         /*
 522          * interrupt stacks must be under 256MB, we cannot afford to take
 523          * SLB misses on them.
 524          */
 525         for_each_cpu(i) {
 526                 softirq_ctx[i] = (struct thread_info *)
 527                         __va(lmb_alloc_base(THREAD_SIZE,
 528                                             THREAD_SIZE, 0x10000000));
 529                 hardirq_ctx[i] = (struct thread_info *)
 530                         __va(lmb_alloc_base(THREAD_SIZE,
 531                                             THREAD_SIZE, 0x10000000));
 532         }
 533 }
 534 #else
 535 #define irqstack_early_init()
 536 #endif
 537
 538 /*
 539  * Stack space used when we detect a bad kernel stack pointer, and
 540  * early in SMP boots before relocation is enabled.
 541  */
 542 static void __init emergency_stack_init(void)
 543 {
 544         unsigned long limit;
 545         unsigned int i;
 546
 547         /*
 548          * Emergency stacks must be under 256MB, we cannot afford to take
 549          * SLB misses on them. The ABI also requires them to be 128-byte
 550          * aligned.
 551          *
 552          * Since we use these as temporary stacks during secondary CPU
 553          * bringup, we need to get at them in real mode. This means they
 554          * must also be within the RMO region.
 555          */
 556         limit = min(0x10000000UL, lmb.rmo_size);
 557
 558         for_each_cpu(i)
 559                 paca[i].emergency_sp =
 560                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 561 }
 562
 563 /*
 564  * Called into from start_kernel, after lock_kernel has been called.
 565  * Initializes bootmem, which is unsed to manage page allocation until
 566  * mem_init is called.
 567  */
 568 void __init setup_arch(char **cmdline_p)
 569 {
 570         extern void do_init_bootmem(void);
 571
 572         ppc64_boot_msg(0x12, "Setup Arch");
 573
 574         *cmdline_p = cmd_line;
 575
 576         /*
 577          * Set cache line size based on type of cpu as a default.
 578          * Systems with OF can look in the properties on the cpu node(s)
 579          * for a possibly more accurate value.
 580          */
 581         dcache_bsize = ppc64_caches.dline_size;
 582         icache_bsize = ppc64_caches.iline_size;
 583
 584         /* reboot on panic */
 585         panic_timeout = 180;
 586
 587         if (ppc_md.panic)
 588                 notifier_chain_register(&panic_notifier_list, &ppc64_panic_block);
 589
 590         init_mm.start_code = PAGE_OFFSET;
 591         init_mm.end_code = (unsigned long) _etext;
 592         init_mm.end_data = (unsigned long) _edata;
 593         init_mm.brk = klimit;
 594
 595         irqstack_early_init();
 596         emergency_stack_init();
 597
 598         stabs_alloc();
 599
 600         /* set up the bootmem stuff with available memory */
 601         do_init_bootmem();
 602         sparse_init();
 603
 604 #ifdef CONFIG_DUMMY_CONSOLE
 605         conswitchp = &dummy_con;
 606 #endif
 607
 608         ppc_md.setup_arch();
 609
 610         /* Use the default idle loop if the platform hasn't provided one. */
 611         if (NULL == ppc_md.idle_loop) {
 612                 ppc_md.idle_loop = default_idle;
 613                 printk(KERN_INFO "Using default idle loop\n");
 614         }
 615
 616         paging_init();
 617         ppc64_boot_msg(0x15, "Setup Done");
 618 }
 619
 620
 621 /* ToDo: do something useful if ppc_md is not yet setup. */
 622 #define PPC64_LINUX_FUNCTION 0x0f000000
 623 #define PPC64_IPL_MESSAGE 0xc0000000
 624 #define PPC64_TERM_MESSAGE 0xb0000000
 625
 626 static void ppc64_do_msg(unsigned int src, const char *msg)
 627 {
 628         if (ppc_md.progress) {
 629                 char buf[128];
 630
 631                 sprintf(buf, "%08X\n", src);
 632                 ppc_md.progress(buf, 0);
 633                 snprintf(buf, 128, "%s", msg);
 634                 ppc_md.progress(buf, 0);
 635         }
 636 }
 637
 638 /* Print a boot progress message. */
 639 void ppc64_boot_msg(unsigned int src, const char *msg)
 640 {
 641         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 642         printk("[boot]%04x %s\n", src, msg);
 643 }
 644
 645 /* Print a termination message (print only -- does not stop the kernel) */
 646 void ppc64_terminate_msg(unsigned int src, const char *msg)
 647 {
 648         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 649         printk("[terminate]%04x %s\n", src, msg);
 650 }
 651
 652 int check_legacy_ioport(unsigned long base_port)
 653 {
 654         if (ppc_md.check_legacy_ioport == NULL)
 655                 return 0;
 656         return ppc_md.check_legacy_ioport(base_port);
 657 }
 658 EXPORT_SYMBOL(check_legacy_ioport);
 659
 660 void cpu_die(void)
 661 {
 662         if (ppc_md.cpu_die)
 663                 ppc_md.cpu_die();
 664 }
 665
 666 #ifdef CONFIG_SMP
 667 void __init setup_per_cpu_areas(void)
 668 {
 669         int i;
 670         unsigned long size;
 671         char *ptr;
 672
 673         /* Copy section for each CPU (we discard the original) */
 674         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 675 #ifdef CONFIG_MODULES
 676         if (size < PERCPU_ENOUGH_ROOM)
 677                 size = PERCPU_ENOUGH_ROOM;
 678 #endif
 679
 680         for_each_cpu(i) {
 681                 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
 682                 if (!ptr)
 683                         panic("Cannot allocate cpu data for CPU %d\n", i);
 684
 685                 paca[i].data_offset = ptr - __per_cpu_start;
 686                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 687         }
 688 }
 689 #endif