arch/powerpc/platforms/pseries/setup.c

   1 /*
   2  *  64-bit pSeries and RS/6000 setup code.
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *  Adapted from 'alpha' version by Gary Thomas
   6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
   7  *  Modified by PPC64 Team, IBM Corp
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License
  11  * as published by the Free Software Foundation; either version
  12  * 2 of the License, or (at your option) any later version.
  13  */
  14
  15 /*
  16  * bootup setup stuff..
  17  */
  18
  19 #undef DEBUG
  20
  21 #include <linux/cpu.h>
  22 #include <linux/errno.h>
  23 #include <linux/sched.h>
  24 #include <linux/kernel.h>
  25 #include <linux/mm.h>
  26 #include <linux/stddef.h>
  27 #include <linux/unistd.h>
  28 #include <linux/slab.h>
  29 #include <linux/user.h>
  30 #include <linux/a.out.h>
  31 #include <linux/tty.h>
  32 #include <linux/major.h>
  33 #include <linux/interrupt.h>
  34 #include <linux/reboot.h>
  35 #include <linux/init.h>
  36 #include <linux/ioport.h>
  37 #include <linux/console.h>
  38 #include <linux/pci.h>
  39 #include <linux/utsname.h>
  40 #include <linux/adb.h>
  41 #include <linux/module.h>
  42 #include <linux/delay.h>
  43 #include <linux/irq.h>
  44 #include <linux/seq_file.h>
  45 #include <linux/root_dev.h>
  46
  47 #include <asm/mmu.h>
  48 #include <asm/processor.h>
  49 #include <asm/io.h>
  50 #include <asm/pgtable.h>
  51 #include <asm/prom.h>
  52 #include <asm/rtas.h>
  53 #include <asm/pci-bridge.h>
  54 #include <asm/iommu.h>
  55 #include <asm/dma.h>
  56 #include <asm/machdep.h>
  57 #include <asm/irq.h>
  58 #include <asm/kexec.h>
  59 #include <asm/time.h>
  60 #include <asm/nvram.h>
  61 #include "xics.h"
  62 #include <asm/pmc.h>
  63 #include <asm/mpic.h>
  64 #include <asm/ppc-pci.h>
  65 #include <asm/i8259.h>
  66 #include <asm/udbg.h>
  67 #include <asm/smp.h>
  68
  69 #include "plpar_wrappers.h"
  70 #include "ras.h"
  71 #include "firmware.h"
  72
  73 #ifdef DEBUG
  74 #define DBG(fmt...) udbg_printf(fmt)
  75 #else
  76 #define DBG(fmt...)
  77 #endif
  78
  79 /* move those away to a .h */
  80 extern void smp_init_pseries_mpic(void);
  81 extern void smp_init_pseries_xics(void);
  82 extern void find_udbg_vterm(void);
  83
  84 int fwnmi_active;  /* TRUE if an FWNMI handler is present */
  85
  86 static void pseries_shared_idle_sleep(void);
  87 static void pseries_dedicated_idle_sleep(void);
  88
  89 static struct device_node *pSeries_mpic_node;
  90
  91 static void pSeries_show_cpuinfo(struct seq_file *m)
  92 {
  93         struct device_node *root;
  94         const char *model = "";
  95
  96         root = of_find_node_by_path("/");
  97         if (root)
  98                 model = get_property(root, "model", NULL);
  99         seq_printf(m, "machine\t\t: CHRP %s\n", model);
 100         of_node_put(root);
 101 }
 102
 103 /* Initialize firmware assisted non-maskable interrupts if
 104  * the firmware supports this feature.
 105  */
 106 static void __init fwnmi_init(void)
 107 {
 108         unsigned long system_reset_addr, machine_check_addr;
 109
 110         int ibm_nmi_register = rtas_token("ibm,nmi-register");
 111         if (ibm_nmi_register == RTAS_UNKNOWN_SERVICE)
 112                 return;
 113
 114         /* If the kernel's not linked at zero we point the firmware at low
 115          * addresses anyway, and use a trampoline to get to the real code. */
 116         system_reset_addr  = __pa(system_reset_fwnmi) - PHYSICAL_START;
 117         machine_check_addr = __pa(machine_check_fwnmi) - PHYSICAL_START;
 118
 119         if (0 == rtas_call(ibm_nmi_register, 2, 1, NULL, system_reset_addr,
 120                                 machine_check_addr))
 121                 fwnmi_active = 1;
 122 }
 123
 124 void pseries_8259_cascade(unsigned int irq, struct irq_desc *desc,
 125                           struct pt_regs *regs)
 126 {
 127         unsigned int cascade_irq = i8259_irq(regs);
 128         if (cascade_irq != NO_IRQ)
 129                 generic_handle_irq(cascade_irq, regs);
 130         desc->chip->eoi(irq);
 131 }
 132
 133 static void __init pseries_mpic_init_IRQ(void)
 134 {
 135         struct device_node *np, *old, *cascade = NULL;
 136         const unsigned int *addrp;
 137         unsigned long intack = 0;
 138         const unsigned int *opprop;
 139         unsigned long openpic_addr = 0;
 140         unsigned int cascade_irq;
 141         int naddr, n, i, opplen;
 142         struct mpic *mpic;
 143
 144         np = of_find_node_by_path("/");
 145         naddr = prom_n_addr_cells(np);
 146         opprop = get_property(np, "platform-open-pic", &opplen);
 147         if (opprop != 0) {
 148                 openpic_addr = of_read_number(opprop, naddr);
 149                 printk(KERN_DEBUG "OpenPIC addr: %lx\n", openpic_addr);
 150         }
 151         of_node_put(np);
 152
 153         BUG_ON(openpic_addr == 0);
 154
 155         /* Setup the openpic driver */
 156         mpic = mpic_alloc(pSeries_mpic_node, openpic_addr,
 157                           MPIC_PRIMARY,
 158                           16, 250, /* isu size, irq count */
 159                           " MPIC     ");
 160         BUG_ON(mpic == NULL);
 161
 162         /* Add ISUs */
 163         opplen /= sizeof(u32);
 164         for (n = 0, i = naddr; i < opplen; i += naddr, n++) {
 165                 unsigned long isuaddr = of_read_number(opprop + i, naddr);
 166                 mpic_assign_isu(mpic, n, isuaddr);
 167         }
 168
 169         /* All ISUs are setup, complete initialization */
 170         mpic_init(mpic);
 171
 172         /* Look for cascade */
 173         for_each_node_by_type(np, "interrupt-controller")
 174                 if (device_is_compatible(np, "chrp,iic")) {
 175                         cascade = np;
 176                         break;
 177                 }
 178         if (cascade == NULL)
 179                 return;
 180
 181         cascade_irq = irq_of_parse_and_map(cascade, 0);
 182         if (cascade == NO_IRQ) {
 183                 printk(KERN_ERR "xics: failed to map cascade interrupt");
 184                 return;
 185         }
 186
 187         /* Check ACK type */
 188         for (old = of_node_get(cascade); old != NULL ; old = np) {
 189                 np = of_get_parent(old);
 190                 of_node_put(old);
 191                 if (np == NULL)
 192                         break;
 193                 if (strcmp(np->name, "pci") != 0)
 194                         continue;
 195                 addrp = get_property(np, "8259-interrupt-acknowledge",
 196                                             NULL);
 197                 if (addrp == NULL)
 198                         continue;
 199                 naddr = prom_n_addr_cells(np);
 200                 intack = addrp[naddr-1];
 201                 if (naddr > 1)
 202                         intack |= ((unsigned long)addrp[naddr-2]) << 32;
 203         }
 204         if (intack)
 205                 printk(KERN_DEBUG "mpic: PCI 8259 intack at 0x%016lx\n",
 206                        intack);
 207         i8259_init(cascade, intack);
 208         of_node_put(cascade);
 209         set_irq_chained_handler(cascade_irq, pseries_8259_cascade);
 210 }
 211
 212 static void pseries_lpar_enable_pmcs(void)
 213 {
 214         unsigned long set, reset;
 215
 216         power4_enable_pmcs();
 217
 218         set = 1UL << 63;
 219         reset = 0;
 220         plpar_hcall_norets(H_PERFMON, set, reset);
 221
 222         /* instruct hypervisor to maintain PMCs */
 223         if (firmware_has_feature(FW_FEATURE_SPLPAR))
 224                 get_lppaca()->pmcregs_in_use = 1;
 225 }
 226
 227 #ifdef CONFIG_KEXEC
 228 static void pseries_kexec_cpu_down_mpic(int crash_shutdown, int secondary)
 229 {
 230         mpic_teardown_this_cpu(secondary);
 231 }
 232
 233 static void pseries_kexec_cpu_down_xics(int crash_shutdown, int secondary)
 234 {
 235         /* Don't risk a hypervisor call if we're crashing */
 236         if (firmware_has_feature(FW_FEATURE_SPLPAR) && !crash_shutdown) {
 237                 unsigned long vpa = __pa(get_lppaca());
 238
 239                 if (unregister_vpa(hard_smp_processor_id(), vpa)) {
 240                         printk("VPA deregistration of cpu %u (hw_cpu_id %d) "
 241                                         "failed\n", smp_processor_id(),
 242                                         hard_smp_processor_id());
 243                 }
 244         }
 245         xics_teardown_cpu(secondary);
 246 }
 247 #endif /* CONFIG_KEXEC */
 248
 249 static void __init pseries_discover_pic(void)
 250 {
 251         struct device_node *np;
 252         const char *typep;
 253
 254         for (np = NULL; (np = of_find_node_by_name(np,
 255                                                    "interrupt-controller"));) {
 256                 typep = get_property(np, "compatible", NULL);
 257                 if (strstr(typep, "open-pic")) {
 258                         pSeries_mpic_node = of_node_get(np);
 259                         ppc_md.init_IRQ       = pseries_mpic_init_IRQ;
 260                         ppc_md.get_irq        = mpic_get_irq;
 261 #ifdef CONFIG_KEXEC
 262                         ppc_md.kexec_cpu_down = pseries_kexec_cpu_down_mpic;
 263 #endif
 264 #ifdef CONFIG_SMP
 265                         smp_init_pseries_mpic();
 266 #endif
 267                         return;
 268                 } else if (strstr(typep, "ppc-xicp")) {
 269                         ppc_md.init_IRQ       = xics_init_IRQ;
 270 #ifdef CONFIG_KEXEC
 271                         ppc_md.kexec_cpu_down = pseries_kexec_cpu_down_xics;
 272 #endif
 273 #ifdef CONFIG_SMP
 274                         smp_init_pseries_xics();
 275 #endif
 276                         return;
 277                 }
 278         }
 279         printk(KERN_ERR "pSeries_discover_pic: failed to recognize"
 280                " interrupt-controller\n");
 281 }
 282
 283 static void __init pSeries_setup_arch(void)
 284 {
 285         /* Discover PIC type and setup ppc_md accordingly */
 286         pseries_discover_pic();
 287
 288         /* openpic global configuration register (64-bit format). */
 289         /* openpic Interrupt Source Unit pointer (64-bit format). */
 290         /* python0 facility area (mmio) (64-bit format) REAL address. */
 291
 292         /* init to some ~sane value until calibrate_delay() runs */
 293         loops_per_jiffy = 50000000;
 294
 295         if (ROOT_DEV == 0) {
 296                 printk("No ramdisk, default root is /dev/sda2\n");
 297                 ROOT_DEV = Root_SDA2;
 298         }
 299
 300         fwnmi_init();
 301
 302         /* Find and initialize PCI host bridges */
 303         init_pci_config_tokens();
 304         find_and_init_phbs();
 305         eeh_init();
 306
 307         pSeries_nvram_init();
 308
 309         /* Choose an idle loop */
 310         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 311                 vpa_init(boot_cpuid);
 312                 if (get_lppaca()->shared_proc) {
 313                         printk(KERN_DEBUG "Using shared processor idle loop\n");
 314                         ppc_md.power_save = pseries_shared_idle_sleep;
 315                 } else {
 316                         printk(KERN_DEBUG "Using dedicated idle loop\n");
 317                         ppc_md.power_save = pseries_dedicated_idle_sleep;
 318                 }
 319         } else {
 320                 printk(KERN_DEBUG "Using default idle loop\n");
 321         }
 322
 323         if (firmware_has_feature(FW_FEATURE_LPAR))
 324                 ppc_md.enable_pmcs = pseries_lpar_enable_pmcs;
 325         else
 326                 ppc_md.enable_pmcs = power4_enable_pmcs;
 327 }
 328
 329 static int __init pSeries_init_panel(void)
 330 {
 331         /* Manually leave the kernel version on the panel. */
 332         ppc_md.progress("Linux ppc64\n", 0);
 333         ppc_md.progress(system_utsname.release, 0);
 334
 335         return 0;
 336 }
 337 arch_initcall(pSeries_init_panel);
 338
 339 static void pSeries_mach_cpu_die(void)
 340 {
 341         local_irq_disable();
 342         idle_task_exit();
 343         xics_teardown_cpu(0);
 344         rtas_stop_self();
 345         /* Should never get here... */
 346         BUG();
 347         for(;;);
 348 }
 349
 350 static int pseries_set_dabr(unsigned long dabr)
 351 {
 352         return plpar_hcall_norets(H_SET_DABR, dabr);
 353 }
 354
 355 static int pseries_set_xdabr(unsigned long dabr)
 356 {
 357         /* We want to catch accesses from kernel and userspace */
 358         return plpar_hcall_norets(H_SET_XDABR, dabr,
 359                         H_DABRX_KERNEL | H_DABRX_USER);
 360 }
 361
 362 /*
 363  * Early initialization.  Relocation is on but do not reference unbolted pages
 364  */
 365 static void __init pSeries_init_early(void)
 366 {
 367         DBG(" -> pSeries_init_early()\n");
 368
 369         fw_feature_init();
 370
 371         if (firmware_has_feature(FW_FEATURE_LPAR))
 372                 find_udbg_vterm();
 373
 374         if (firmware_has_feature(FW_FEATURE_DABR))
 375                 ppc_md.set_dabr = pseries_set_dabr;
 376         else if (firmware_has_feature(FW_FEATURE_XDABR))
 377                 ppc_md.set_dabr = pseries_set_xdabr;
 378
 379         iommu_init_early_pSeries();
 380
 381         DBG(" <- pSeries_init_early()\n");
 382 }
 383
 384
 385 static int pSeries_check_legacy_ioport(unsigned int baseport)
 386 {
 387         struct device_node *np;
 388
 389 #define I8042_DATA_REG  0x60
 390 #define FDC_BASE        0x3f0
 391
 392
 393         switch(baseport) {
 394         case I8042_DATA_REG:
 395                 np = of_find_node_by_type(NULL, "8042");
 396                 if (np == NULL)
 397                         return -ENODEV;
 398                 of_node_put(np);
 399                 break;
 400         case FDC_BASE:
 401                 np = of_find_node_by_type(NULL, "fdc");
 402                 if (np == NULL)
 403                         return -ENODEV;
 404                 of_node_put(np);
 405                 break;
 406         }
 407         return 0;
 408 }
 409
 410 /*
 411  * Called very early, MMU is off, device-tree isn't unflattened
 412  */
 413
 414 static int __init pSeries_probe_hypertas(unsigned long node,
 415                                          const char *uname, int depth,
 416                                          void *data)
 417 {
 418         if (depth != 1 ||
 419             (strcmp(uname, "rtas") != 0 && strcmp(uname, "rtas@0") != 0))
 420                 return 0;
 421
 422         if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL) != NULL)
 423                 powerpc_firmware_features |= FW_FEATURE_LPAR;
 424
 425         if (firmware_has_feature(FW_FEATURE_LPAR))
 426                 hpte_init_lpar();
 427         else
 428                 hpte_init_native();
 429
 430         return 1;
 431 }
 432
 433 static int __init pSeries_probe(void)
 434 {
 435         unsigned long root = of_get_flat_dt_root();
 436         char *dtype = of_get_flat_dt_prop(of_get_flat_dt_root(),
 437                                           "device_type", NULL);
 438         if (dtype == NULL)
 439                 return 0;
 440         if (strcmp(dtype, "chrp"))
 441                 return 0;
 442
 443         /* Cell blades firmware claims to be chrp while it's not. Until this
 444          * is fixed, we need to avoid those here.
 445          */
 446         if (of_flat_dt_is_compatible(root, "IBM,CPBW-1.0") ||
 447             of_flat_dt_is_compatible(root, "IBM,CBEA"))
 448                 return 0;
 449
 450         DBG("pSeries detected, looking for LPAR capability...\n");
 451
 452         /* Now try to figure out if we are running on LPAR */
 453         of_scan_flat_dt(pSeries_probe_hypertas, NULL);
 454
 455         DBG("Machine is%s LPAR !\n",
 456             (powerpc_firmware_features & FW_FEATURE_LPAR) ? "" : " not");
 457
 458         return 1;
 459 }
 460
 461
 462 DECLARE_PER_CPU(unsigned long, smt_snooze_delay);
 463
 464 static void pseries_dedicated_idle_sleep(void)
 465 {
 466         unsigned int cpu = smp_processor_id();
 467         unsigned long start_snooze;
 468         unsigned long *smt_snooze_delay = &__get_cpu_var(smt_snooze_delay);
 469
 470         /*
 471          * Indicate to the HV that we are idle. Now would be
 472          * a good time to find other work to dispatch.
 473          */
 474         get_lppaca()->idle = 1;
 475
 476         /*
 477          * We come in with interrupts disabled, and need_resched()
 478          * has been checked recently.  If we should poll for a little
 479          * while, do so.
 480          */
 481         if (*smt_snooze_delay) {
 482                 start_snooze = get_tb() +
 483                         *smt_snooze_delay * tb_ticks_per_usec;
 484                 local_irq_enable();
 485                 set_thread_flag(TIF_POLLING_NRFLAG);
 486
 487                 while (get_tb() < start_snooze) {
 488                         if (need_resched() || cpu_is_offline(cpu))
 489                                 goto out;
 490                         ppc64_runlatch_off();
 491                         HMT_low();
 492                         HMT_very_low();
 493                 }
 494
 495                 HMT_medium();
 496                 clear_thread_flag(TIF_POLLING_NRFLAG);
 497                 smp_mb();
 498                 local_irq_disable();
 499                 if (need_resched() || cpu_is_offline(cpu))
 500                         goto out;
 501         }
 502
 503         /*
 504          * If not SMT, cede processor.  If CPU is running SMT
 505          * cede if the other thread is not idle, so that it can
 506          * go single-threaded.  If the other thread is idle,
 507          * we ask the hypervisor if it has pending work it
 508          * wants to do and cede if it does.  Otherwise we keep
 509          * polling in order to reduce interrupt latency.
 510          *
 511          * Doing the cede when the other thread is active will
 512          * result in this thread going dormant, meaning the other
 513          * thread gets to run in single-threaded (ST) mode, which
 514          * is slightly faster than SMT mode with this thread at
 515          * very low priority.  The cede enables interrupts, which
 516          * doesn't matter here.
 517          */
 518         if (!cpu_has_feature(CPU_FTR_SMT) || !lppaca[cpu ^ 1].idle
 519             || poll_pending() == H_PENDING)
 520                 cede_processor();
 521
 522 out:
 523         HMT_medium();
 524         get_lppaca()->idle = 0;
 525 }
 526
 527 static void pseries_shared_idle_sleep(void)
 528 {
 529         /*
 530          * Indicate to the HV that we are idle. Now would be
 531          * a good time to find other work to dispatch.
 532          */
 533         get_lppaca()->idle = 1;
 534
 535         /*
 536          * Yield the processor to the hypervisor.  We return if
 537          * an external interrupt occurs (which are driven prior
 538          * to returning here) or if a prod occurs from another
 539          * processor. When returning here, external interrupts
 540          * are enabled.
 541          */
 542         cede_processor();
 543
 544         get_lppaca()->idle = 0;
 545 }
 546
 547 static int pSeries_pci_probe_mode(struct pci_bus *bus)
 548 {
 549         if (firmware_has_feature(FW_FEATURE_LPAR))
 550                 return PCI_PROBE_DEVTREE;
 551         return PCI_PROBE_NORMAL;
 552 }
 553
 554 define_machine(pseries) {
 555         .name                   = "pSeries",
 556         .probe                  = pSeries_probe,
 557         .setup_arch             = pSeries_setup_arch,
 558         .init_early             = pSeries_init_early,
 559         .show_cpuinfo           = pSeries_show_cpuinfo,
 560         .log_error              = pSeries_log_error,
 561         .pcibios_fixup          = pSeries_final_fixup,
 562         .pci_probe_mode         = pSeries_pci_probe_mode,
 563         .irq_bus_setup          = pSeries_irq_bus_setup,
 564         .restart                = rtas_restart,
 565         .power_off              = rtas_power_off,
 566         .halt                   = rtas_halt,
 567         .panic                  = rtas_os_term,
 568         .cpu_die                = pSeries_mach_cpu_die,
 569         .get_boot_time          = rtas_get_boot_time,
 570         .get_rtc_time           = rtas_get_rtc_time,
 571         .set_rtc_time           = rtas_set_rtc_time,
 572         .calibrate_decr         = generic_calibrate_decr,
 573         .progress               = rtas_progress,
 574         .check_legacy_ioport    = pSeries_check_legacy_ioport,
 575         .system_reset_exception = pSeries_system_reset_exception,
 576         .machine_check_exception = pSeries_machine_check_exception,
 577 #ifdef CONFIG_KEXEC
 578         .machine_kexec          = default_machine_kexec,
 579         .machine_kexec_prepare  = default_machine_kexec_prepare,
 580         .machine_crash_shutdown = default_machine_crash_shutdown,
 581 #endif
 582 };