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 #include <asm/swiotlb.h>
  65 #include <asm/mmu_context.h>
  66
  67 #include "setup.h"
  68
  69 #ifdef DEBUG
  70 #define DBG(fmt...) udbg_printf(fmt)
  71 #else
  72 #define DBG(fmt...)
  73 #endif
  74
  75 int boot_cpuid = 0;
  76 u64 ppc64_pft_size;
  77
  78 /* Pick defaults since we might want to patch instructions
  79  * before we've read this from the device tree.
  80  */
  81 struct ppc64_caches ppc64_caches = {
  82         .dline_size = 0x40,
  83         .log_dline_size = 6,
  84         .iline_size = 0x40,
  85         .log_iline_size = 6
  86 };
  87 EXPORT_SYMBOL_GPL(ppc64_caches);
  88
  89 /*
  90  * These are used in binfmt_elf.c to put aux entries on the stack
  91  * for each elf executable being started.
  92  */
  93 int dcache_bsize;
  94 int icache_bsize;
  95 int ucache_bsize;
  96
  97 #ifdef CONFIG_SMP
  98
  99 static int smt_enabled_cmdline;
 100
 101 /* Look for ibm,smt-enabled OF option */
 102 static void check_smt_enabled(void)
 103 {
 104         struct device_node *dn;
 105         const char *smt_option;
 106
 107         /* Allow the command line to overrule the OF option */
 108         if (smt_enabled_cmdline)
 109                 return;
 110
 111         dn = of_find_node_by_path("/options");
 112
 113         if (dn) {
 114                 smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
 115
 116                 if (smt_option) {
 117                         if (!strcmp(smt_option, "on"))
 118                                 smt_enabled_at_boot = 1;
 119                         else if (!strcmp(smt_option, "off"))
 120                                 smt_enabled_at_boot = 0;
 121                 }
 122         }
 123 }
 124
 125 /* Look for smt-enabled= cmdline option */
 126 static int __init early_smt_enabled(char *p)
 127 {
 128         smt_enabled_cmdline = 1;
 129
 130         if (!p)
 131                 return 0;
 132
 133         if (!strcmp(p, "on") || !strcmp(p, "1"))
 134                 smt_enabled_at_boot = 1;
 135         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 136                 smt_enabled_at_boot = 0;
 137
 138         return 0;
 139 }
 140 early_param("smt-enabled", early_smt_enabled);
 141
 142 #else
 143 #define check_smt_enabled()
 144 #endif /* CONFIG_SMP */
 145
 146 /* Put the paca pointer into r13 and SPRG_PACA */
 147 void __init setup_paca(int cpu)
 148 {
 149         local_paca = &paca[cpu];
 150         mtspr(SPRN_SPRG_PACA, local_paca);
 151 #ifdef CONFIG_PPC_BOOK3E
 152         mtspr(SPRN_SPRG_TLB_EXFRAME, local_paca->extlb);
 153 #endif
 154 }
 155
 156 /*
 157  * Early initialization entry point. This is called by head.S
 158  * with MMU translation disabled. We rely on the "feature" of
 159  * the CPU that ignores the top 2 bits of the address in real
 160  * mode so we can access kernel globals normally provided we
 161  * only toy with things in the RMO region. From here, we do
 162  * some early parsing of the device-tree to setup out LMB
 163  * data structures, and allocate & initialize the hash table
 164  * and segment tables so we can start running with translation
 165  * enabled.
 166  *
 167  * It is this function which will call the probe() callback of
 168  * the various platform types and copy the matching one to the
 169  * global ppc_md structure. Your platform can eventually do
 170  * some very early initializations from the probe() routine, but
 171  * this is not recommended, be very careful as, for example, the
 172  * device-tree is not accessible via normal means at this point.
 173  */
 174
 175 void __init early_setup(unsigned long dt_ptr)
 176 {
 177         /* -------- printk is _NOT_ safe to use here ! ------- */
 178
 179         /* Fill in any unititialised pacas */
 180         initialise_pacas();
 181
 182         /* Identify CPU type */
 183         identify_cpu(0, mfspr(SPRN_PVR));
 184
 185         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 186         setup_paca(0);
 187
 188         /* Initialize lockdep early or else spinlocks will blow */
 189         lockdep_init();
 190
 191         /* -------- printk is now safe to use ------- */
 192
 193         /* Enable early debugging if any specified (see udbg.h) */
 194         udbg_early_init();
 195
 196         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 197
 198         /*
 199          * Do early initialization using the flattened device
 200          * tree, such as retrieving the physical memory map or
 201          * calculating/retrieving the hash table size.
 202          */
 203         early_init_devtree(__va(dt_ptr));
 204
 205         /* Now we know the logical id of our boot cpu, setup the paca. */
 206         setup_paca(boot_cpuid);
 207
 208         /* Fix up paca fields required for the boot cpu */
 209         get_paca()->cpu_start = 1;
 210
 211         /* Probe the machine type */
 212         probe_machine();
 213
 214         setup_kdump_trampoline();
 215
 216         DBG("Found, Initializing memory management...\n");
 217
 218         /* Initialize the hash table or TLB handling */
 219         early_init_mmu();
 220
 221         DBG(" <- early_setup()\n");
 222 }
 223
 224 #ifdef CONFIG_SMP
 225 void early_setup_secondary(void)
 226 {
 227         /* Mark interrupts enabled in PACA */
 228         get_paca()->soft_enabled = 0;
 229
 230         /* Initialize the hash table or TLB handling */
 231         early_init_mmu_secondary();
 232 }
 233
 234 #endif /* CONFIG_SMP */
 235
 236 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 237 void smp_release_cpus(void)
 238 {
 239         unsigned long *ptr;
 240
 241         DBG(" -> smp_release_cpus()\n");
 242
 243         /* All secondary cpus are spinning on a common spinloop, release them
 244          * all now so they can start to spin on their individual paca
 245          * spinloops. For non SMP kernels, the secondary cpus never get out
 246          * of the common spinloop.
 247          */
 248
 249         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 250                         - PHYSICAL_START);
 251         *ptr = __pa(generic_secondary_smp_init);
 252         mb();
 253
 254         DBG(" <- smp_release_cpus()\n");
 255 }
 256 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 257
 258 /*
 259  * Initialize some remaining members of the ppc64_caches and systemcfg
 260  * structures
 261  * (at least until we get rid of them completely). This is mostly some
 262  * cache informations about the CPU that will be used by cache flush
 263  * routines and/or provided to userland
 264  */
 265 static void __init initialize_cache_info(void)
 266 {
 267         struct device_node *np;
 268         unsigned long num_cpus = 0;
 269
 270         DBG(" -> initialize_cache_info()\n");
 271
 272         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 273                 num_cpus += 1;
 274
 275                 /* We're assuming *all* of the CPUs have the same
 276                  * d-cache and i-cache sizes... -Peter
 277                  */
 278
 279                 if ( num_cpus == 1 ) {
 280                         const u32 *sizep, *lsizep;
 281                         u32 size, lsize;
 282
 283                         size = 0;
 284                         lsize = cur_cpu_spec->dcache_bsize;
 285                         sizep = of_get_property(np, "d-cache-size", NULL);
 286                         if (sizep != NULL)
 287                                 size = *sizep;
 288                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
 289                         /* fallback if block size missing */
 290                         if (lsizep == NULL)
 291                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
 292                         if (lsizep != NULL)
 293                                 lsize = *lsizep;
 294                         if (sizep == 0 || lsizep == 0)
 295                                 DBG("Argh, can't find dcache properties ! "
 296                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 297
 298                         ppc64_caches.dsize = size;
 299                         ppc64_caches.dline_size = lsize;
 300                         ppc64_caches.log_dline_size = __ilog2(lsize);
 301                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 302
 303                         size = 0;
 304                         lsize = cur_cpu_spec->icache_bsize;
 305                         sizep = of_get_property(np, "i-cache-size", NULL);
 306                         if (sizep != NULL)
 307                                 size = *sizep;
 308                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
 309                         if (lsizep == NULL)
 310                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
 311                         if (lsizep != NULL)
 312                                 lsize = *lsizep;
 313                         if (sizep == 0 || lsizep == 0)
 314                                 DBG("Argh, can't find icache properties ! "
 315                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 316
 317                         ppc64_caches.isize = size;
 318                         ppc64_caches.iline_size = lsize;
 319                         ppc64_caches.log_iline_size = __ilog2(lsize);
 320                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 321                 }
 322         }
 323
 324         DBG(" <- initialize_cache_info()\n");
 325 }
 326
 327
 328 /*
 329  * Do some initial setup of the system.  The parameters are those which
 330  * were passed in from the bootloader.
 331  */
 332 void __init setup_system(void)
 333 {
 334         DBG(" -> setup_system()\n");
 335
 336         /* Apply the CPUs-specific and firmware specific fixups to kernel
 337          * text (nop out sections not relevant to this CPU or this firmware)
 338          */
 339         do_feature_fixups(cur_cpu_spec->cpu_features,
 340                           &__start___ftr_fixup, &__stop___ftr_fixup);
 341         do_feature_fixups(cur_cpu_spec->mmu_features,
 342                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
 343         do_feature_fixups(powerpc_firmware_features,
 344                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 345         do_lwsync_fixups(cur_cpu_spec->cpu_features,
 346                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
 347
 348         /*
 349          * Unflatten the device-tree passed by prom_init or kexec
 350          */
 351         unflatten_device_tree();
 352
 353         /*
 354          * Fill the ppc64_caches & systemcfg structures with informations
 355          * retrieved from the device-tree.
 356          */
 357         initialize_cache_info();
 358
 359         /*
 360          * Initialize irq remapping subsystem
 361          */
 362         irq_early_init();
 363
 364 #ifdef CONFIG_PPC_RTAS
 365         /*
 366          * Initialize RTAS if available
 367          */
 368         rtas_initialize();
 369 #endif /* CONFIG_PPC_RTAS */
 370
 371         /*
 372          * Check if we have an initrd provided via the device-tree
 373          */
 374         check_for_initrd();
 375
 376         /*
 377          * Do some platform specific early initializations, that includes
 378          * setting up the hash table pointers. It also sets up some interrupt-mapping
 379          * related options that will be used by finish_device_tree()
 380          */
 381         if (ppc_md.init_early)
 382                 ppc_md.init_early();
 383
 384         /*
 385          * We can discover serial ports now since the above did setup the
 386          * hash table management for us, thus ioremap works. We do that early
 387          * so that further code can be debugged
 388          */
 389         find_legacy_serial_ports();
 390
 391         /*
 392          * Register early console
 393          */
 394         register_early_udbg_console();
 395
 396         /*
 397          * Initialize xmon
 398          */
 399         xmon_setup();
 400
 401         check_smt_enabled();
 402         smp_setup_cpu_maps();
 403
 404 #ifdef CONFIG_SMP
 405         /* Release secondary cpus out of their spinloops at 0x60 now that
 406          * we can map physical -> logical CPU ids
 407          */
 408         smp_release_cpus();
 409 #endif
 410
 411         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 412
 413         printk("-----------------------------------------------------\n");
 414         printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
 415         printk("physicalMemorySize            = 0x%llx\n", lmb_phys_mem_size());
 416         if (ppc64_caches.dline_size != 0x80)
 417                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 418                        ppc64_caches.dline_size);
 419         if (ppc64_caches.iline_size != 0x80)
 420                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 421                        ppc64_caches.iline_size);
 422 #ifdef CONFIG_PPC_STD_MMU_64
 423         if (htab_address)
 424                 printk("htab_address                  = 0x%p\n", htab_address);
 425         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 426 #endif /* CONFIG_PPC_STD_MMU_64 */
 427         if (PHYSICAL_START > 0)
 428                 printk("physical_start                = 0x%llx\n",
 429                        (unsigned long long)PHYSICAL_START);
 430         printk("-----------------------------------------------------\n");
 431
 432         DBG(" <- setup_system()\n");
 433 }
 434
 435 #ifdef CONFIG_IRQSTACKS
 436 static void __init irqstack_early_init(void)
 437 {
 438         unsigned int i;
 439
 440         /*
 441          * interrupt stacks must be under 256MB, we cannot afford to take
 442          * SLB misses on them.
 443          */
 444         for_each_possible_cpu(i) {
 445                 softirq_ctx[i] = (struct thread_info *)
 446                         __va(lmb_alloc_base(THREAD_SIZE,
 447                                             THREAD_SIZE, 0x10000000));
 448                 hardirq_ctx[i] = (struct thread_info *)
 449                         __va(lmb_alloc_base(THREAD_SIZE,
 450                                             THREAD_SIZE, 0x10000000));
 451         }
 452 }
 453 #else
 454 #define irqstack_early_init()
 455 #endif
 456
 457 #ifdef CONFIG_PPC_BOOK3E
 458 static void __init exc_lvl_early_init(void)
 459 {
 460         unsigned int i;
 461
 462         for_each_possible_cpu(i) {
 463                 critirq_ctx[i] = (struct thread_info *)
 464                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 465                 dbgirq_ctx[i] = (struct thread_info *)
 466                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 467                 mcheckirq_ctx[i] = (struct thread_info *)
 468                         __va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
 469         }
 470 }
 471 #else
 472 #define exc_lvl_early_init()
 473 #endif
 474
 475 /*
 476  * Stack space used when we detect a bad kernel stack pointer, and
 477  * early in SMP boots before relocation is enabled.
 478  */
 479 static void __init emergency_stack_init(void)
 480 {
 481         unsigned long limit;
 482         unsigned int i;
 483
 484         /*
 485          * Emergency stacks must be under 256MB, we cannot afford to take
 486          * SLB misses on them. The ABI also requires them to be 128-byte
 487          * aligned.
 488          *
 489          * Since we use these as temporary stacks during secondary CPU
 490          * bringup, we need to get at them in real mode. This means they
 491          * must also be within the RMO region.
 492          */
 493         limit = min(0x10000000ULL, lmb.rmo_size);
 494
 495         for_each_possible_cpu(i) {
 496                 unsigned long sp;
 497                 sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
 498                 sp += THREAD_SIZE;
 499                 paca[i].emergency_sp = __va(sp);
 500         }
 501 }
 502
 503 /*
 504  * Called into from start_kernel, after lock_kernel has been called.
 505  * Initializes bootmem, which is unsed to manage page allocation until
 506  * mem_init is called.
 507  */
 508 void __init setup_arch(char **cmdline_p)
 509 {
 510         ppc64_boot_msg(0x12, "Setup Arch");
 511
 512         *cmdline_p = cmd_line;
 513
 514         /*
 515          * Set cache line size based on type of cpu as a default.
 516          * Systems with OF can look in the properties on the cpu node(s)
 517          * for a possibly more accurate value.
 518          */
 519         dcache_bsize = ppc64_caches.dline_size;
 520         icache_bsize = ppc64_caches.iline_size;
 521
 522         /* reboot on panic */
 523         panic_timeout = 180;
 524
 525         if (ppc_md.panic)
 526                 setup_panic();
 527
 528         init_mm.start_code = (unsigned long)_stext;
 529         init_mm.end_code = (unsigned long) _etext;
 530         init_mm.end_data = (unsigned long) _edata;
 531         init_mm.brk = klimit;
 532
 533         irqstack_early_init();
 534         exc_lvl_early_init();
 535         emergency_stack_init();
 536
 537 #ifdef CONFIG_PPC_STD_MMU_64
 538         stabs_alloc();
 539 #endif
 540         /* set up the bootmem stuff with available memory */
 541         do_init_bootmem();
 542         sparse_init();
 543
 544 #ifdef CONFIG_DUMMY_CONSOLE
 545         conswitchp = &dummy_con;
 546 #endif
 547
 548         if (ppc_md.setup_arch)
 549                 ppc_md.setup_arch();
 550
 551 #ifdef CONFIG_SWIOTLB
 552         if (ppc_swiotlb_enable)
 553                 swiotlb_init();
 554 #endif
 555
 556         paging_init();
 557
 558         /* Initialize the MMU context management stuff */
 559         mmu_context_init();
 560
 561         ppc64_boot_msg(0x15, "Setup Done");
 562 }
 563
 564
 565 /* ToDo: do something useful if ppc_md is not yet setup. */
 566 #define PPC64_LINUX_FUNCTION 0x0f000000
 567 #define PPC64_IPL_MESSAGE 0xc0000000
 568 #define PPC64_TERM_MESSAGE 0xb0000000
 569
 570 static void ppc64_do_msg(unsigned int src, const char *msg)
 571 {
 572         if (ppc_md.progress) {
 573                 char buf[128];
 574
 575                 sprintf(buf, "%08X\n", src);
 576                 ppc_md.progress(buf, 0);
 577                 snprintf(buf, 128, "%s", msg);
 578                 ppc_md.progress(buf, 0);
 579         }
 580 }
 581
 582 /* Print a boot progress message. */
 583 void ppc64_boot_msg(unsigned int src, const char *msg)
 584 {
 585         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 586         printk("[boot]%04x %s\n", src, msg);
 587 }
 588
 589 void cpu_die(void)
 590 {
 591         if (ppc_md.cpu_die)
 592                 ppc_md.cpu_die();
 593 }
 594
 595 #ifdef CONFIG_SMP
 596 #define PCPU_DYN_SIZE           ()
 597
 598 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
 599 {
 600         return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
 601                                     __pa(MAX_DMA_ADDRESS));
 602 }
 603
 604 static void __init pcpu_fc_free(void *ptr, size_t size)
 605 {
 606         free_bootmem(__pa(ptr), size);
 607 }
 608
 609 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
 610 {
 611         if (cpu_to_node(from) == cpu_to_node(to))
 612                 return LOCAL_DISTANCE;
 613         else
 614                 return REMOTE_DISTANCE;
 615 }
 616
 617 void __init setup_per_cpu_areas(void)
 618 {
 619         const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
 620         size_t atom_size;
 621         unsigned long delta;
 622         unsigned int cpu;
 623         int rc;
 624
 625         /*
 626          * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
 627          * to group units.  For larger mappings, use 1M atom which
 628          * should be large enough to contain a number of units.
 629          */
 630         if (mmu_linear_psize == MMU_PAGE_4K)
 631                 atom_size = PAGE_SIZE;
 632         else
 633                 atom_size = 1 << 20;
 634
 635         rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
 636                                     pcpu_fc_alloc, pcpu_fc_free);
 637         if (rc < 0)
 638                 panic("cannot initialize percpu area (err=%d)", rc);
 639
 640         delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
 641         for_each_possible_cpu(cpu)
 642                 paca[cpu].data_offset = delta + pcpu_unit_offsets[cpu];
 643 }
 644 #endif
 645
 646
 647 #ifdef CONFIG_PPC_INDIRECT_IO
 648 struct ppc_pci_io ppc_pci_io;
 649 EXPORT_SYMBOL(ppc_pci_io);
 650 #endif /* CONFIG_PPC_INDIRECT_IO */
 651