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