arch/powerpc/kernel/setup_64.c

   1 /*
   2  *
   3  * Common boot and setup code.
   4  *
   5  * Copyright (C) 2001 PPC64 Team, IBM Corp
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #undef DEBUG
  14
  15 #include <linux/module.h>
  16 #include <linux/string.h>
  17 #include <linux/sched.h>
  18 #include <linux/init.h>
  19 #include <linux/kernel.h>
  20 #include <linux/reboot.h>
  21 #include <linux/delay.h>
  22 #include <linux/initrd.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/ioport.h>
  25 #include <linux/console.h>
  26 #include <linux/utsname.h>
  27 #include <linux/tty.h>
  28 #include <linux/root_dev.h>
  29 #include <linux/notifier.h>
  30 #include <linux/cpu.h>
  31 #include <linux/unistd.h>
  32 #include <linux/serial.h>
  33 #include <linux/serial_8250.h>
  34 #include <linux/bootmem.h>
  35 #include <linux/pci.h>
  36 #include <linux/lockdep.h>
  37 #include <linux/lmb.h>
  38 #include <asm/io.h>
  39 #include <asm/kdump.h>
  40 #include <asm/prom.h>
  41 #include <asm/processor.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/smp.h>
  44 #include <asm/elf.h>
  45 #include <asm/machdep.h>
  46 #include <asm/paca.h>
  47 #include <asm/time.h>
  48 #include <asm/cputable.h>
  49 #include <asm/sections.h>
  50 #include <asm/btext.h>
  51 #include <asm/nvram.h>
  52 #include <asm/setup.h>
  53 #include <asm/system.h>
  54 #include <asm/rtas.h>
  55 #include <asm/iommu.h>
  56 #include <asm/serial.h>
  57 #include <asm/cache.h>
  58 #include <asm/page.h>
  59 #include <asm/mmu.h>
  60 #include <asm/firmware.h>
  61 #include <asm/xmon.h>
  62 #include <asm/udbg.h>
  63 #include <asm/kexec.h>
  64
  65 #include "setup.h"
  66
  67 #ifdef DEBUG
  68 #define DBG(fmt...) udbg_printf(fmt)
  69 #else
  70 #define DBG(fmt...)
  71 #endif
  72
  73 int have_of = 1;
  74 int boot_cpuid = 0;
  75 u64 ppc64_pft_size;
  76
  77 /* Pick defaults since we might want to patch instructions
  78  * before we've read this from the device tree.
  79  */
  80 struct ppc64_caches ppc64_caches = {
  81         .dline_size = 0x40,
  82         .log_dline_size = 6,
  83         .iline_size = 0x40,
  84         .log_iline_size = 6
  85 };
  86 EXPORT_SYMBOL_GPL(ppc64_caches);
  87
  88 /*
  89  * These are used in binfmt_elf.c to put aux entries on the stack
  90  * for each elf executable being started.
  91  */
  92 int dcache_bsize;
  93 int icache_bsize;
  94 int ucache_bsize;
  95
  96 #ifdef CONFIG_SMP
  97
  98 static int smt_enabled_cmdline;
  99
 100 /* Look for ibm,smt-enabled OF option */
 101 static void check_smt_enabled(void)
 102 {
 103         struct device_node *dn;
 104         const char *smt_option;
 105
 106         /* Allow the command line to overrule the OF option */
 107         if (smt_enabled_cmdline)
 108                 return;
 109
 110         dn = of_find_node_by_path("/options");
 111
 112         if (dn) {
 113                 smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
 114
 115                 if (smt_option) {
 116                         if (!strcmp(smt_option, "on"))
 117                                 smt_enabled_at_boot = 1;
 118                         else if (!strcmp(smt_option, "off"))
 119                                 smt_enabled_at_boot = 0;
 120                 }
 121         }
 122 }
 123
 124 /* Look for smt-enabled= cmdline option */
 125 static int __init early_smt_enabled(char *p)
 126 {
 127         smt_enabled_cmdline = 1;
 128
 129         if (!p)
 130                 return 0;
 131
 132         if (!strcmp(p, "on") || !strcmp(p, "1"))
 133                 smt_enabled_at_boot = 1;
 134         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 135                 smt_enabled_at_boot = 0;
 136
 137         return 0;
 138 }
 139 early_param("smt-enabled", early_smt_enabled);
 140
 141 #else
 142 #define check_smt_enabled()
 143 #endif /* CONFIG_SMP */
 144
 145 /* Put the paca pointer into r13 and SPRG3 */
 146 void __init setup_paca(int cpu)
 147 {
 148         local_paca = &paca[cpu];
 149         mtspr(SPRN_SPRG3, local_paca);
 150 }
 151
 152 /*
 153  * Early initialization entry point. This is called by head.S
 154  * with MMU translation disabled. We rely on the "feature" of
 155  * the CPU that ignores the top 2 bits of the address in real
 156  * mode so we can access kernel globals normally provided we
 157  * only toy with things in the RMO region. From here, we do
 158  * some early parsing of the device-tree to setup out LMB
 159  * data structures, and allocate & initialize the hash table
 160  * and segment tables so we can start running with translation
 161  * enabled.
 162  *
 163  * It is this function which will call the probe() callback of
 164  * the various platform types and copy the matching one to the
 165  * global ppc_md structure. Your platform can eventually do
 166  * some very early initializations from the probe() routine, but
 167  * this is not recommended, be very careful as, for example, the
 168  * device-tree is not accessible via normal means at this point.
 169  */
 170
 171 void __init early_setup(unsigned long dt_ptr)
 172 {
 173         /* Fill in any unititialised pacas */
 174         initialise_pacas();
 175
 176         /* Identify CPU type */
 177         identify_cpu(0, mfspr(SPRN_PVR));
 178
 179         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 180         setup_paca(0);
 181
 182         /* Enable early debugging if any specified (see udbg.h) */
 183         udbg_early_init();
 184
 185         /* Initialize lockdep early or else spinlocks will blow */
 186         lockdep_init();
 187
 188         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 189
 190         /*
 191          * Do early initialization using the flattened device
 192          * tree, such as retrieving the physical memory map or
 193          * calculating/retrieving the hash table size.
 194          */
 195         early_init_devtree(__va(dt_ptr));
 196
 197         /* Now we know the logical id of our boot cpu, setup the paca. */
 198         setup_paca(boot_cpuid);
 199
 200         /* Fix up paca fields required for the boot cpu */
 201         get_paca()->cpu_start = 1;
 202         get_paca()->stab_real = __pa((u64)&initial_stab);
 203         get_paca()->stab_addr = (u64)&initial_stab;
 204
 205         /* Probe the machine type */
 206         probe_machine();
 207
 208         setup_kdump_trampoline();
 209
 210         DBG("Found, Initializing memory management...\n");
 211
 212         /*
 213          * Initialize the MMU Hash table and create the linear mapping
 214          * of memory. Has to be done before stab/slb initialization as
 215          * this is currently where the page size encoding is obtained
 216          */
 217         htab_initialize();
 218
 219         /*
 220          * Initialize stab / SLB management except on iSeries
 221          */
 222         if (cpu_has_feature(CPU_FTR_SLB))
 223                 slb_initialize();
 224         else if (!firmware_has_feature(FW_FEATURE_ISERIES))
 225                 stab_initialize(get_paca()->stab_real);
 226
 227         DBG(" <- early_setup()\n");
 228 }
 229
 230 #ifdef CONFIG_SMP
 231 void early_setup_secondary(void)
 232 {
 233         struct paca_struct *lpaca = get_paca();
 234
 235         /* Mark interrupts enabled in PACA */
 236         lpaca->soft_enabled = 0;
 237
 238         /* Initialize hash table for that CPU */
 239         htab_initialize_secondary();
 240
 241         /* Initialize STAB/SLB. We use a virtual address as it works
 242          * in real mode on pSeries and we want a virutal address on
 243          * iSeries anyway
 244          */
 245         if (cpu_has_feature(CPU_FTR_SLB))
 246                 slb_initialize();
 247         else
 248                 stab_initialize(lpaca->stab_addr);
 249 }
 250
 251 #endif /* CONFIG_SMP */
 252
 253 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 254 void smp_release_cpus(void)
 255 {
 256         extern unsigned long __secondary_hold_spinloop;
 257         unsigned long *ptr;
 258
 259         DBG(" -> smp_release_cpus()\n");
 260
 261         /* All secondary cpus are spinning on a common spinloop, release them
 262          * all now so they can start to spin on their individual paca
 263          * spinloops. For non SMP kernels, the secondary cpus never get out
 264          * of the common spinloop.
 265          * This is useless but harmless on iSeries, secondaries are already
 266          * waiting on their paca spinloops. */
 267
 268         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 269                         - PHYSICAL_START);
 270         *ptr = 1;
 271         mb();
 272
 273         DBG(" <- smp_release_cpus()\n");
 274 }
 275 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 276
 277 /*
 278  * Initialize some remaining members of the ppc64_caches and systemcfg
 279  * structures
 280  * (at least until we get rid of them completely). This is mostly some
 281  * cache informations about the CPU that will be used by cache flush
 282  * routines and/or provided to userland
 283  */
 284 static void __init initialize_cache_info(void)
 285 {
 286         struct device_node *np;
 287         unsigned long num_cpus = 0;
 288
 289         DBG(" -> initialize_cache_info()\n");
 290
 291         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 292                 num_cpus += 1;
 293
 294                 /* We're assuming *all* of the CPUs have the same
 295                  * d-cache and i-cache sizes... -Peter
 296                  */
 297
 298                 if ( num_cpus == 1 ) {
 299                         const u32 *sizep, *lsizep;
 300                         u32 size, lsize;
 301
 302                         size = 0;
 303                         lsize = cur_cpu_spec->dcache_bsize;
 304                         sizep = of_get_property(np, "d-cache-size", NULL);
 305                         if (sizep != NULL)
 306                                 size = *sizep;
 307                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
 308                         /* fallback if block size missing */
 309                         if (lsizep == NULL)
 310                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
 311                         if (lsizep != NULL)
 312                                 lsize = *lsizep;
 313                         if (sizep == 0 || lsizep == 0)
 314                                 DBG("Argh, can't find dcache properties ! "
 315                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 316
 317                         ppc64_caches.dsize = size;
 318                         ppc64_caches.dline_size = lsize;
 319                         ppc64_caches.log_dline_size = __ilog2(lsize);
 320                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 321
 322                         size = 0;
 323                         lsize = cur_cpu_spec->icache_bsize;
 324                         sizep = of_get_property(np, "i-cache-size", NULL);
 325                         if (sizep != NULL)
 326                                 size = *sizep;
 327                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
 328                         if (lsizep == NULL)
 329                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
 330                         if (lsizep != NULL)
 331                                 lsize = *lsizep;
 332                         if (sizep == 0 || lsizep == 0)
 333                                 DBG("Argh, can't find icache properties ! "
 334                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 335
 336                         ppc64_caches.isize = size;
 337                         ppc64_caches.iline_size = lsize;
 338                         ppc64_caches.log_iline_size = __ilog2(lsize);
 339                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 340                 }
 341         }
 342
 343         DBG(" <- initialize_cache_info()\n");
 344 }
 345
 346
 347 /*
 348  * Do some initial setup of the system.  The parameters are those which
 349  * were passed in from the bootloader.
 350  */
 351 void __init setup_system(void)
 352 {
 353         DBG(" -> setup_system()\n");
 354
 355         /* Apply the CPUs-specific and firmware specific fixups to kernel
 356          * text (nop out sections not relevant to this CPU or this firmware)
 357          */
 358         do_feature_fixups(cur_cpu_spec->cpu_features,
 359                           &__start___ftr_fixup, &__stop___ftr_fixup);
 360         do_feature_fixups(powerpc_firmware_features,
 361                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 362
 363         /*
 364          * Unflatten the device-tree passed by prom_init or kexec
 365          */
 366         unflatten_device_tree();
 367
 368         /*
 369          * Fill the ppc64_caches & systemcfg structures with informations
 370          * retrieved from the device-tree.
 371          */
 372         initialize_cache_info();
 373
 374         /*
 375          * Initialize irq remapping subsystem
 376          */
 377         irq_early_init();
 378
 379 #ifdef CONFIG_PPC_RTAS
 380         /*
 381          * Initialize RTAS if available
 382          */
 383         rtas_initialize();
 384 #endif /* CONFIG_PPC_RTAS */
 385
 386         /*
 387          * Check if we have an initrd provided via the device-tree
 388          */
 389         check_for_initrd();
 390
 391         /*
 392          * Do some platform specific early initializations, that includes
 393          * setting up the hash table pointers. It also sets up some interrupt-mapping
 394          * related options that will be used by finish_device_tree()
 395          */
 396         if (ppc_md.init_early)
 397                 ppc_md.init_early();
 398
 399         /*
 400          * We can discover serial ports now since the above did setup the
 401          * hash table management for us, thus ioremap works. We do that early
 402          * so that further code can be debugged
 403          */
 404         find_legacy_serial_ports();
 405
 406         /*
 407          * Register early console
 408          */
 409         register_early_udbg_console();
 410
 411         /*
 412          * Initialize xmon
 413          */
 414         xmon_setup();
 415
 416         check_smt_enabled();
 417         smp_setup_cpu_maps();
 418
 419 #ifdef CONFIG_SMP
 420         /* Release secondary cpus out of their spinloops at 0x60 now that
 421          * we can map physical -> logical CPU ids
 422          */
 423         smp_release_cpus();
 424 #endif
 425
 426         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 427
 428         printk("-----------------------------------------------------\n");
 429         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 430         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 431         if (ppc64_caches.dline_size != 0x80)
 432                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 433                        ppc64_caches.dline_size);
 434         if (ppc64_caches.iline_size != 0x80)
 435                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 436                        ppc64_caches.iline_size);
 437         if (htab_address)
 438                 printk("htab_address                  = 0x%p\n", htab_address);
 439         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 440 #if PHYSICAL_START > 0
 441         printk("physical_start                = 0x%lx\n", PHYSICAL_START);
 442 #endif
 443         printk("-----------------------------------------------------\n");
 444
 445         DBG(" <- setup_system()\n");
 446 }
 447
 448 #ifdef CONFIG_IRQSTACKS
 449 static void __init irqstack_early_init(void)
 450 {
 451         unsigned int i;
 452
 453         /*
 454          * interrupt stacks must be under 256MB, we cannot afford to take
 455          * SLB misses on them.
 456          */
 457         for_each_possible_cpu(i) {
 458                 softirq_ctx[i] = (struct thread_info *)
 459                         __va(lmb_alloc_base(THREAD_SIZE,
 460                                             THREAD_SIZE, 0x10000000));
 461                 hardirq_ctx[i] = (struct thread_info *)
 462                         __va(lmb_alloc_base(THREAD_SIZE,
 463                                             THREAD_SIZE, 0x10000000));
 464         }
 465 }
 466 #else
 467 #define irqstack_early_init()
 468 #endif
 469
 470 /*
 471  * Stack space used when we detect a bad kernel stack pointer, and
 472  * early in SMP boots before relocation is enabled.
 473  */
 474 static void __init emergency_stack_init(void)
 475 {
 476         unsigned long limit;
 477         unsigned int i;
 478
 479         /*
 480          * Emergency stacks must be under 256MB, we cannot afford to take
 481          * SLB misses on them. The ABI also requires them to be 128-byte
 482          * aligned.
 483          *
 484          * Since we use these as temporary stacks during secondary CPU
 485          * bringup, we need to get at them in real mode. This means they
 486          * must also be within the RMO region.
 487          */
 488         limit = min(0x10000000UL, lmb.rmo_size);
 489
 490         for_each_possible_cpu(i)
 491                 paca[i].emergency_sp =
 492                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 493 }
 494
 495 /*
 496  * Called into from start_kernel, after lock_kernel has been called.
 497  * Initializes bootmem, which is unsed to manage page allocation until
 498  * mem_init is called.
 499  */
 500 void __init setup_arch(char **cmdline_p)
 501 {
 502         ppc64_boot_msg(0x12, "Setup Arch");
 503
 504         *cmdline_p = cmd_line;
 505
 506         /*
 507          * Set cache line size based on type of cpu as a default.
 508          * Systems with OF can look in the properties on the cpu node(s)
 509          * for a possibly more accurate value.
 510          */
 511         dcache_bsize = ppc64_caches.dline_size;
 512         icache_bsize = ppc64_caches.iline_size;
 513
 514         /* reboot on panic */
 515         panic_timeout = 180;
 516
 517         if (ppc_md.panic)
 518                 setup_panic();
 519
 520         init_mm.start_code = (unsigned long)_stext;
 521         init_mm.end_code = (unsigned long) _etext;
 522         init_mm.end_data = (unsigned long) _edata;
 523         init_mm.brk = klimit;
 524
 525         irqstack_early_init();
 526         emergency_stack_init();
 527
 528         stabs_alloc();
 529
 530         /* set up the bootmem stuff with available memory */
 531         do_init_bootmem();
 532         sparse_init();
 533
 534 #ifdef CONFIG_DUMMY_CONSOLE
 535         conswitchp = &dummy_con;
 536 #endif
 537
 538         if (ppc_md.setup_arch)
 539                 ppc_md.setup_arch();
 540
 541         paging_init();
 542         ppc64_boot_msg(0x15, "Setup Done");
 543 }
 544
 545
 546 /* ToDo: do something useful if ppc_md is not yet setup. */
 547 #define PPC64_LINUX_FUNCTION 0x0f000000
 548 #define PPC64_IPL_MESSAGE 0xc0000000
 549 #define PPC64_TERM_MESSAGE 0xb0000000
 550
 551 static void ppc64_do_msg(unsigned int src, const char *msg)
 552 {
 553         if (ppc_md.progress) {
 554                 char buf[128];
 555
 556                 sprintf(buf, "%08X\n", src);
 557                 ppc_md.progress(buf, 0);
 558                 snprintf(buf, 128, "%s", msg);
 559                 ppc_md.progress(buf, 0);
 560         }
 561 }
 562
 563 /* Print a boot progress message. */
 564 void ppc64_boot_msg(unsigned int src, const char *msg)
 565 {
 566         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 567         printk("[boot]%04x %s\n", src, msg);
 568 }
 569
 570 /* Print a termination message (print only -- does not stop the kernel) */
 571 void ppc64_terminate_msg(unsigned int src, const char *msg)
 572 {
 573         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 574         printk("[terminate]%04x %s\n", src, msg);
 575 }
 576
 577 void cpu_die(void)
 578 {
 579         if (ppc_md.cpu_die)
 580                 ppc_md.cpu_die();
 581 }
 582
 583 #ifdef CONFIG_SMP
 584 void __init setup_per_cpu_areas(void)
 585 {
 586         int i;
 587         unsigned long size;
 588         char *ptr;
 589
 590         /* Copy section for each CPU (we discard the original) */
 591         size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
 592 #ifdef CONFIG_MODULES
 593         if (size < PERCPU_ENOUGH_ROOM)
 594                 size = PERCPU_ENOUGH_ROOM;
 595 #endif
 596
 597         for_each_possible_cpu(i) {
 598                 ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 599                 if (!ptr)
 600                         panic("Cannot allocate cpu data for CPU %d\n", i);
 601
 602                 paca[i].data_offset = ptr - __per_cpu_start;
 603                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 604         }
 605
 606         /* Now that per_cpu is setup, initialize cpu_sibling_map */
 607         smp_setup_cpu_sibling_map();
 608 }
 609 #endif
 610
 611
 612 #ifdef CONFIG_PPC_INDIRECT_IO
 613 struct ppc_pci_io ppc_pci_io;
 614 EXPORT_SYMBOL(ppc_pci_io);
 615 #endif /* CONFIG_PPC_INDIRECT_IO */
 616