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 <asm/io.h>
  37 #include <asm/prom.h>
  38 #include <asm/processor.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/smp.h>
  41 #include <asm/elf.h>
  42 #include <asm/machdep.h>
  43 #include <asm/paca.h>
  44 #include <asm/time.h>
  45 #include <asm/cputable.h>
  46 #include <asm/sections.h>
  47 #include <asm/btext.h>
  48 #include <asm/nvram.h>
  49 #include <asm/setup.h>
  50 #include <asm/system.h>
  51 #include <asm/rtas.h>
  52 #include <asm/iommu.h>
  53 #include <asm/serial.h>
  54 #include <asm/cache.h>
  55 #include <asm/page.h>
  56 #include <asm/mmu.h>
  57 #include <asm/lmb.h>
  58 #include <asm/iseries/it_lp_naca.h>
  59 #include <asm/firmware.h>
  60 #include <asm/xmon.h>
  61 #include <asm/udbg.h>
  62 #include <asm/kexec.h>
  63
  64 #include "setup.h"
  65
  66 #ifdef DEBUG
  67 #define DBG(fmt...) udbg_printf(fmt)
  68 #else
  69 #define DBG(fmt...)
  70 #endif
  71
  72 /*
  73  * Here are some early debugging facilities. You can enable one
  74  * but your kernel will not boot on anything else if you do so
  75  */
  76
  77 /* This one is for use on LPAR machines that support an HVC console
  78  * on vterm 0
  79  */
  80 extern void udbg_init_debug_lpar(void);
  81 /* This one is for use on Apple G5 machines
  82  */
  83 extern void udbg_init_pmac_realmode(void);
  84 /* That's RTAS panel debug */
  85 extern void call_rtas_display_status_delay(unsigned char c);
  86 /* Here's maple real mode debug */
  87 extern void udbg_init_maple_realmode(void);
  88
  89 #define EARLY_DEBUG_INIT() do {} while(0)
  90
  91 #if 0
  92 #define EARLY_DEBUG_INIT() udbg_init_debug_lpar()
  93 #define EARLY_DEBUG_INIT() udbg_init_maple_realmode()
  94 #define EARLY_DEBUG_INIT() udbg_init_pmac_realmode()
  95 #define EARLY_DEBUG_INIT()                                              \
  96         do { udbg_putc = call_rtas_display_status_delay; } while(0)
  97 #endif
  98
  99 int have_of = 1;
 100 int boot_cpuid = 0;
 101 int boot_cpuid_phys = 0;
 102 dev_t boot_dev;
 103 u64 ppc64_pft_size;
 104
 105 /* Pick defaults since we might want to patch instructions
 106  * before we've read this from the device tree.
 107  */
 108 struct ppc64_caches ppc64_caches = {
 109         .dline_size = 0x80,
 110         .log_dline_size = 7,
 111         .iline_size = 0x80,
 112         .log_iline_size = 7
 113 };
 114 EXPORT_SYMBOL_GPL(ppc64_caches);
 115
 116 /*
 117  * These are used in binfmt_elf.c to put aux entries on the stack
 118  * for each elf executable being started.
 119  */
 120 int dcache_bsize;
 121 int icache_bsize;
 122 int ucache_bsize;
 123
 124 /* The main machine-dep calls structure
 125  */
 126 struct machdep_calls ppc_md;
 127 EXPORT_SYMBOL(ppc_md);
 128
 129 #ifdef CONFIG_MAGIC_SYSRQ
 130 unsigned long SYSRQ_KEY;
 131 #endif /* CONFIG_MAGIC_SYSRQ */
 132
 133
 134 static int ppc64_panic_event(struct notifier_block *, unsigned long, void *);
 135 static struct notifier_block ppc64_panic_block = {
 136         .notifier_call = ppc64_panic_event,
 137         .priority = INT_MIN /* may not return; must be done last */
 138 };
 139
 140 #ifdef CONFIG_SMP
 141
 142 static int smt_enabled_cmdline;
 143
 144 /* Look for ibm,smt-enabled OF option */
 145 static void check_smt_enabled(void)
 146 {
 147         struct device_node *dn;
 148         char *smt_option;
 149
 150         /* Allow the command line to overrule the OF option */
 151         if (smt_enabled_cmdline)
 152                 return;
 153
 154         dn = of_find_node_by_path("/options");
 155
 156         if (dn) {
 157                 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
 158
 159                 if (smt_option) {
 160                         if (!strcmp(smt_option, "on"))
 161                                 smt_enabled_at_boot = 1;
 162                         else if (!strcmp(smt_option, "off"))
 163                                 smt_enabled_at_boot = 0;
 164                 }
 165         }
 166 }
 167
 168 /* Look for smt-enabled= cmdline option */
 169 static int __init early_smt_enabled(char *p)
 170 {
 171         smt_enabled_cmdline = 1;
 172
 173         if (!p)
 174                 return 0;
 175
 176         if (!strcmp(p, "on") || !strcmp(p, "1"))
 177                 smt_enabled_at_boot = 1;
 178         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 179                 smt_enabled_at_boot = 0;
 180
 181         return 0;
 182 }
 183 early_param("smt-enabled", early_smt_enabled);
 184
 185 #else
 186 #define check_smt_enabled()
 187 #endif /* CONFIG_SMP */
 188
 189 extern struct machdep_calls pSeries_md;
 190 extern struct machdep_calls pmac_md;
 191 extern struct machdep_calls maple_md;
 192 extern struct machdep_calls cell_md;
 193 extern struct machdep_calls iseries_md;
 194
 195 /* Ultimately, stuff them in an elf section like initcalls... */
 196 static struct machdep_calls __initdata *machines[] = {
 197 #ifdef CONFIG_PPC_PSERIES
 198         &pSeries_md,
 199 #endif /* CONFIG_PPC_PSERIES */
 200 #ifdef CONFIG_PPC_PMAC
 201         &pmac_md,
 202 #endif /* CONFIG_PPC_PMAC */
 203 #ifdef CONFIG_PPC_MAPLE
 204         &maple_md,
 205 #endif /* CONFIG_PPC_MAPLE */
 206 #ifdef CONFIG_PPC_CELL
 207         &cell_md,
 208 #endif
 209 #ifdef CONFIG_PPC_ISERIES
 210         &iseries_md,
 211 #endif
 212         NULL
 213 };
 214
 215 /*
 216  * Early initialization entry point. This is called by head.S
 217  * with MMU translation disabled. We rely on the "feature" of
 218  * the CPU that ignores the top 2 bits of the address in real
 219  * mode so we can access kernel globals normally provided we
 220  * only toy with things in the RMO region. From here, we do
 221  * some early parsing of the device-tree to setup out LMB
 222  * data structures, and allocate & initialize the hash table
 223  * and segment tables so we can start running with translation
 224  * enabled.
 225  *
 226  * It is this function which will call the probe() callback of
 227  * the various platform types and copy the matching one to the
 228  * global ppc_md structure. Your platform can eventually do
 229  * some very early initializations from the probe() routine, but
 230  * this is not recommended, be very careful as, for example, the
 231  * device-tree is not accessible via normal means at this point.
 232  */
 233
 234 void __init early_setup(unsigned long dt_ptr)
 235 {
 236         struct paca_struct *lpaca = get_paca();
 237         static struct machdep_calls **mach;
 238
 239         /*
 240          * Enable early debugging if any specified (see top of
 241          * this file)
 242          */
 243         EARLY_DEBUG_INIT();
 244
 245         DBG(" -> early_setup()\n");
 246
 247         /*
 248          * Do early initializations using the flattened device
 249          * tree, like retreiving the physical memory map or
 250          * calculating/retreiving the hash table size
 251          */
 252         early_init_devtree(__va(dt_ptr));
 253
 254         /*
 255          * Iterate all ppc_md structures until we find the proper
 256          * one for the current machine type
 257          */
 258         DBG("Probing machine type for platform %x...\n", _machine);
 259
 260         for (mach = machines; *mach; mach++) {
 261                 if ((*mach)->probe(_machine))
 262                         break;
 263         }
 264         /* What can we do if we didn't find ? */
 265         if (*mach == NULL) {
 266                 DBG("No suitable machine found !\n");
 267                 for (;;);
 268         }
 269         ppc_md = **mach;
 270
 271         DBG("Found, Initializing memory management...\n");
 272
 273         /*
 274          * Initialize the MMU Hash table and create the linear mapping
 275          * of memory. Has to be done before stab/slb initialization as
 276          * this is currently where the page size encoding is obtained
 277          */
 278         htab_initialize();
 279
 280         /*
 281          * Initialize stab / SLB management except on iSeries
 282          */
 283         if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
 284                 if (cpu_has_feature(CPU_FTR_SLB))
 285                         slb_initialize();
 286                 else
 287                         stab_initialize(lpaca->stab_real);
 288         }
 289
 290         DBG(" <- early_setup()\n");
 291 }
 292
 293 #ifdef CONFIG_SMP
 294 void early_setup_secondary(void)
 295 {
 296         struct paca_struct *lpaca = get_paca();
 297
 298         /* Mark enabled in PACA */
 299         lpaca->proc_enabled = 0;
 300
 301         /* Initialize hash table for that CPU */
 302         htab_initialize_secondary();
 303
 304         /* Initialize STAB/SLB. We use a virtual address as it works
 305          * in real mode on pSeries and we want a virutal address on
 306          * iSeries anyway
 307          */
 308         if (cpu_has_feature(CPU_FTR_SLB))
 309                 slb_initialize();
 310         else
 311                 stab_initialize(lpaca->stab_addr);
 312 }
 313
 314 #endif /* CONFIG_SMP */
 315
 316 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 317 void smp_release_cpus(void)
 318 {
 319         extern unsigned long __secondary_hold_spinloop;
 320
 321         DBG(" -> smp_release_cpus()\n");
 322
 323         /* All secondary cpus are spinning on a common spinloop, release them
 324          * all now so they can start to spin on their individual paca
 325          * spinloops. For non SMP kernels, the secondary cpus never get out
 326          * of the common spinloop.
 327          * This is useless but harmless on iSeries, secondaries are already
 328          * waiting on their paca spinloops. */
 329
 330         __secondary_hold_spinloop = 1;
 331         mb();
 332
 333         DBG(" <- smp_release_cpus()\n");
 334 }
 335 #else
 336 #define smp_release_cpus()
 337 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 338
 339 /*
 340  * Initialize some remaining members of the ppc64_caches and systemcfg
 341  * structures
 342  * (at least until we get rid of them completely). This is mostly some
 343  * cache informations about the CPU that will be used by cache flush
 344  * routines and/or provided to userland
 345  */
 346 static void __init initialize_cache_info(void)
 347 {
 348         struct device_node *np;
 349         unsigned long num_cpus = 0;
 350
 351         DBG(" -> initialize_cache_info()\n");
 352
 353         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 354                 num_cpus += 1;
 355
 356                 /* We're assuming *all* of the CPUs have the same
 357                  * d-cache and i-cache sizes... -Peter
 358                  */
 359
 360                 if ( num_cpus == 1 ) {
 361                         u32 *sizep, *lsizep;
 362                         u32 size, lsize;
 363                         const char *dc, *ic;
 364
 365                         /* Then read cache informations */
 366                         if (_machine == PLATFORM_POWERMAC) {
 367                                 dc = "d-cache-block-size";
 368                                 ic = "i-cache-block-size";
 369                         } else {
 370                                 dc = "d-cache-line-size";
 371                                 ic = "i-cache-line-size";
 372                         }
 373
 374                         size = 0;
 375                         lsize = cur_cpu_spec->dcache_bsize;
 376                         sizep = (u32 *)get_property(np, "d-cache-size", NULL);
 377                         if (sizep != NULL)
 378                                 size = *sizep;
 379                         lsizep = (u32 *) get_property(np, dc, NULL);
 380                         if (lsizep != NULL)
 381                                 lsize = *lsizep;
 382                         if (sizep == 0 || lsizep == 0)
 383                                 DBG("Argh, can't find dcache properties ! "
 384                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 385
 386                         ppc64_caches.dsize = size;
 387                         ppc64_caches.dline_size = lsize;
 388                         ppc64_caches.log_dline_size = __ilog2(lsize);
 389                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 390
 391                         size = 0;
 392                         lsize = cur_cpu_spec->icache_bsize;
 393                         sizep = (u32 *)get_property(np, "i-cache-size", NULL);
 394                         if (sizep != NULL)
 395                                 size = *sizep;
 396                         lsizep = (u32 *)get_property(np, ic, NULL);
 397                         if (lsizep != NULL)
 398                                 lsize = *lsizep;
 399                         if (sizep == 0 || lsizep == 0)
 400                                 DBG("Argh, can't find icache properties ! "
 401                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 402
 403                         ppc64_caches.isize = size;
 404                         ppc64_caches.iline_size = lsize;
 405                         ppc64_caches.log_iline_size = __ilog2(lsize);
 406                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 407                 }
 408         }
 409
 410         DBG(" <- initialize_cache_info()\n");
 411 }
 412
 413
 414 /*
 415  * Do some initial setup of the system.  The parameters are those which
 416  * were passed in from the bootloader.
 417  */
 418 void __init setup_system(void)
 419 {
 420         DBG(" -> setup_system()\n");
 421
 422         /*
 423          * Unflatten the device-tree passed by prom_init or kexec
 424          */
 425         unflatten_device_tree();
 426
 427 #ifdef CONFIG_KEXEC
 428         kexec_setup();  /* requires unflattened device tree. */
 429 #endif
 430
 431         /*
 432          * Fill the ppc64_caches & systemcfg structures with informations
 433          * retreived from the device-tree. Need to be called before
 434          * finish_device_tree() since the later requires some of the
 435          * informations filled up here to properly parse the interrupt
 436          * tree.
 437          * It also sets up the cache line sizes which allows to call
 438          * routines like flush_icache_range (used by the hash init
 439          * later on).
 440          */
 441         initialize_cache_info();
 442
 443 #ifdef CONFIG_PPC_RTAS
 444         /*
 445          * Initialize RTAS if available
 446          */
 447         rtas_initialize();
 448 #endif /* CONFIG_PPC_RTAS */
 449
 450         /*
 451          * Check if we have an initrd provided via the device-tree
 452          */
 453         check_for_initrd();
 454
 455         /*
 456          * Do some platform specific early initializations, that includes
 457          * setting up the hash table pointers. It also sets up some interrupt-mapping
 458          * related options that will be used by finish_device_tree()
 459          */
 460         ppc_md.init_early();
 461
 462         /*
 463          * We can discover serial ports now since the above did setup the
 464          * hash table management for us, thus ioremap works. We do that early
 465          * so that further code can be debugged
 466          */
 467 #ifdef CONFIG_PPC_MULTIPLATFORM
 468         find_legacy_serial_ports();
 469 #endif
 470
 471         /*
 472          * "Finish" the device-tree, that is do the actual parsing of
 473          * some of the properties like the interrupt map
 474          */
 475         finish_device_tree();
 476
 477 #ifdef CONFIG_BOOTX_TEXT
 478         init_boot_display();
 479 #endif
 480
 481         /*
 482          * Initialize xmon
 483          */
 484 #ifdef CONFIG_XMON_DEFAULT
 485         xmon_init(1);
 486 #endif
 487         /*
 488          * Register early console
 489          */
 490         register_early_udbg_console();
 491
 492         /* Save unparsed command line copy for /proc/cmdline */
 493         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
 494
 495         parse_early_param();
 496
 497         check_smt_enabled();
 498         smp_setup_cpu_maps();
 499
 500         /* Release secondary cpus out of their spinloops at 0x60 now that
 501          * we can map physical -> logical CPU ids
 502          */
 503         smp_release_cpus();
 504
 505         printk("Starting Linux PPC64 %s\n", system_utsname.version);
 506
 507         printk("-----------------------------------------------------\n");
 508         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 509         printk("ppc64_interrupt_controller    = 0x%ld\n",
 510                ppc64_interrupt_controller);
 511         printk("platform                      = 0x%x\n", _machine);
 512         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 513         printk("ppc64_caches.dcache_line_size = 0x%x\n",
 514                ppc64_caches.dline_size);
 515         printk("ppc64_caches.icache_line_size = 0x%x\n",
 516                ppc64_caches.iline_size);
 517         printk("htab_address                  = 0x%p\n", htab_address);
 518         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 519         printk("-----------------------------------------------------\n");
 520
 521         mm_init_ppc64();
 522
 523         DBG(" <- setup_system()\n");
 524 }
 525
 526 static int ppc64_panic_event(struct notifier_block *this,
 527                              unsigned long event, void *ptr)
 528 {
 529         ppc_md.panic((char *)ptr);  /* May not return */
 530         return NOTIFY_DONE;
 531 }
 532
 533 #ifdef CONFIG_IRQSTACKS
 534 static void __init irqstack_early_init(void)
 535 {
 536         unsigned int i;
 537
 538         /*
 539          * interrupt stacks must be under 256MB, we cannot afford to take
 540          * SLB misses on them.
 541          */
 542         for_each_cpu(i) {
 543                 softirq_ctx[i] = (struct thread_info *)
 544                         __va(lmb_alloc_base(THREAD_SIZE,
 545                                             THREAD_SIZE, 0x10000000));
 546                 hardirq_ctx[i] = (struct thread_info *)
 547                         __va(lmb_alloc_base(THREAD_SIZE,
 548                                             THREAD_SIZE, 0x10000000));
 549         }
 550 }
 551 #else
 552 #define irqstack_early_init()
 553 #endif
 554
 555 /*
 556  * Stack space used when we detect a bad kernel stack pointer, and
 557  * early in SMP boots before relocation is enabled.
 558  */
 559 static void __init emergency_stack_init(void)
 560 {
 561         unsigned long limit;
 562         unsigned int i;
 563
 564         /*
 565          * Emergency stacks must be under 256MB, we cannot afford to take
 566          * SLB misses on them. The ABI also requires them to be 128-byte
 567          * aligned.
 568          *
 569          * Since we use these as temporary stacks during secondary CPU
 570          * bringup, we need to get at them in real mode. This means they
 571          * must also be within the RMO region.
 572          */
 573         limit = min(0x10000000UL, lmb.rmo_size);
 574
 575         for_each_cpu(i)
 576                 paca[i].emergency_sp =
 577                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 578 }
 579
 580 /*
 581  * Called into from start_kernel, after lock_kernel has been called.
 582  * Initializes bootmem, which is unsed to manage page allocation until
 583  * mem_init is called.
 584  */
 585 void __init setup_arch(char **cmdline_p)
 586 {
 587         extern void do_init_bootmem(void);
 588
 589         ppc64_boot_msg(0x12, "Setup Arch");
 590
 591         *cmdline_p = cmd_line;
 592
 593         /*
 594          * Set cache line size based on type of cpu as a default.
 595          * Systems with OF can look in the properties on the cpu node(s)
 596          * for a possibly more accurate value.
 597          */
 598         dcache_bsize = ppc64_caches.dline_size;
 599         icache_bsize = ppc64_caches.iline_size;
 600
 601         /* reboot on panic */
 602         panic_timeout = 180;
 603
 604         if (ppc_md.panic)
 605                 notifier_chain_register(&panic_notifier_list, &ppc64_panic_block);
 606
 607         init_mm.start_code = PAGE_OFFSET;
 608         init_mm.end_code = (unsigned long) _etext;
 609         init_mm.end_data = (unsigned long) _edata;
 610         init_mm.brk = klimit;
 611
 612         irqstack_early_init();
 613         emergency_stack_init();
 614
 615         stabs_alloc();
 616
 617         /* set up the bootmem stuff with available memory */
 618         do_init_bootmem();
 619         sparse_init();
 620
 621 #ifdef CONFIG_DUMMY_CONSOLE
 622         conswitchp = &dummy_con;
 623 #endif
 624
 625         ppc_md.setup_arch();
 626
 627         /* Use the default idle loop if the platform hasn't provided one. */
 628         if (NULL == ppc_md.idle_loop) {
 629                 ppc_md.idle_loop = default_idle;
 630                 printk(KERN_INFO "Using default idle loop\n");
 631         }
 632
 633         paging_init();
 634         ppc64_boot_msg(0x15, "Setup Done");
 635 }
 636
 637
 638 /* ToDo: do something useful if ppc_md is not yet setup. */
 639 #define PPC64_LINUX_FUNCTION 0x0f000000
 640 #define PPC64_IPL_MESSAGE 0xc0000000
 641 #define PPC64_TERM_MESSAGE 0xb0000000
 642
 643 static void ppc64_do_msg(unsigned int src, const char *msg)
 644 {
 645         if (ppc_md.progress) {
 646                 char buf[128];
 647
 648                 sprintf(buf, "%08X\n", src);
 649                 ppc_md.progress(buf, 0);
 650                 snprintf(buf, 128, "%s", msg);
 651                 ppc_md.progress(buf, 0);
 652         }
 653 }
 654
 655 /* Print a boot progress message. */
 656 void ppc64_boot_msg(unsigned int src, const char *msg)
 657 {
 658         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 659         printk("[boot]%04x %s\n", src, msg);
 660 }
 661
 662 /* Print a termination message (print only -- does not stop the kernel) */
 663 void ppc64_terminate_msg(unsigned int src, const char *msg)
 664 {
 665         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 666         printk("[terminate]%04x %s\n", src, msg);
 667 }
 668
 669 int check_legacy_ioport(unsigned long base_port)
 670 {
 671         if (ppc_md.check_legacy_ioport == NULL)
 672                 return 0;
 673         return ppc_md.check_legacy_ioport(base_port);
 674 }
 675 EXPORT_SYMBOL(check_legacy_ioport);
 676
 677 void cpu_die(void)
 678 {
 679         if (ppc_md.cpu_die)
 680                 ppc_md.cpu_die();
 681 }