arch/powerpc/platforms/powernv/setup.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * PowerNV setup code.
   4  *
   5  * Copyright 2011 IBM Corp.
   6  */
   7
   8 #undef DEBUG
   9
  10 #include <linux/cpu.h>
  11 #include <linux/errno.h>
  12 #include <linux/sched.h>
  13 #include <linux/kernel.h>
  14 #include <linux/tty.h>
  15 #include <linux/reboot.h>
  16 #include <linux/init.h>
  17 #include <linux/console.h>
  18 #include <linux/delay.h>
  19 #include <linux/irq.h>
  20 #include <linux/seq_file.h>
  21 #include <linux/of.h>
  22 #include <linux/of_fdt.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/bug.h>
  25 #include <linux/pci.h>
  26 #include <linux/cpufreq.h>
  27 #include <linux/memblock.h>
  28
  29 #include <asm/machdep.h>
  30 #include <asm/firmware.h>
  31 #include <asm/xics.h>
  32 #include <asm/xive.h>
  33 #include <asm/opal.h>
  34 #include <asm/kexec.h>
  35 #include <asm/smp.h>
  36 #include <asm/tm.h>
  37 #include <asm/setup.h>
  38 #include <asm/security_features.h>
  39
  40 #include "powernv.h"
  41
  42
  43 static bool fw_feature_is(const char *state, const char *name,
  44                           struct device_node *fw_features)
  45 {
  46         struct device_node *np;
  47         bool rc = false;
  48
  49         np = of_get_child_by_name(fw_features, name);
  50         if (np) {
  51                 rc = of_property_read_bool(np, state);
  52                 of_node_put(np);
  53         }
  54
  55         return rc;
  56 }
  57
  58 static void init_fw_feat_flags(struct device_node *np)
  59 {
  60         if (fw_feature_is("enabled", "inst-spec-barrier-ori31,31,0", np))
  61                 security_ftr_set(SEC_FTR_SPEC_BAR_ORI31);
  62
  63         if (fw_feature_is("enabled", "fw-bcctrl-serialized", np))
  64                 security_ftr_set(SEC_FTR_BCCTRL_SERIALISED);
  65
  66         if (fw_feature_is("enabled", "inst-l1d-flush-ori30,30,0", np))
  67                 security_ftr_set(SEC_FTR_L1D_FLUSH_ORI30);
  68
  69         if (fw_feature_is("enabled", "inst-l1d-flush-trig2", np))
  70                 security_ftr_set(SEC_FTR_L1D_FLUSH_TRIG2);
  71
  72         if (fw_feature_is("enabled", "fw-l1d-thread-split", np))
  73                 security_ftr_set(SEC_FTR_L1D_THREAD_PRIV);
  74
  75         if (fw_feature_is("enabled", "fw-count-cache-disabled", np))
  76                 security_ftr_set(SEC_FTR_COUNT_CACHE_DISABLED);
  77
  78         if (fw_feature_is("enabled", "fw-count-cache-flush-bcctr2,0,0", np))
  79                 security_ftr_set(SEC_FTR_BCCTR_FLUSH_ASSIST);
  80
  81         if (fw_feature_is("enabled", "needs-count-cache-flush-on-context-switch", np))
  82                 security_ftr_set(SEC_FTR_FLUSH_COUNT_CACHE);
  83
  84         /*
  85          * The features below are enabled by default, so we instead look to see
  86          * if firmware has *disabled* them, and clear them if so.
  87          */
  88         if (fw_feature_is("disabled", "speculation-policy-favor-security", np))
  89                 security_ftr_clear(SEC_FTR_FAVOUR_SECURITY);
  90
  91         if (fw_feature_is("disabled", "needs-l1d-flush-msr-pr-0-to-1", np))
  92                 security_ftr_clear(SEC_FTR_L1D_FLUSH_PR);
  93
  94         if (fw_feature_is("disabled", "needs-l1d-flush-msr-hv-1-to-0", np))
  95                 security_ftr_clear(SEC_FTR_L1D_FLUSH_HV);
  96
  97         if (fw_feature_is("disabled", "needs-spec-barrier-for-bound-checks", np))
  98                 security_ftr_clear(SEC_FTR_BNDS_CHK_SPEC_BAR);
  99 }
 100
 101 static void pnv_setup_security_mitigations(void)
 102 {
 103         struct device_node *np, *fw_features;
 104         enum l1d_flush_type type;
 105         bool enable;
 106
 107         /* Default to fallback in case fw-features are not available */
 108         type = L1D_FLUSH_FALLBACK;
 109
 110         np = of_find_node_by_name(NULL, "ibm,opal");
 111         fw_features = of_get_child_by_name(np, "fw-features");
 112         of_node_put(np);
 113
 114         if (fw_features) {
 115                 init_fw_feat_flags(fw_features);
 116                 of_node_put(fw_features);
 117
 118                 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_TRIG2))
 119                         type = L1D_FLUSH_MTTRIG;
 120
 121                 if (security_ftr_enabled(SEC_FTR_L1D_FLUSH_ORI30))
 122                         type = L1D_FLUSH_ORI;
 123         }
 124
 125         /*
 126          * If we are non-Power9 bare metal, we don't need to flush on kernel
 127          * entry or after user access: they fix a P9 specific vulnerability.
 128          */
 129         if (!pvr_version_is(PVR_POWER9)) {
 130                 security_ftr_clear(SEC_FTR_L1D_FLUSH_ENTRY);
 131                 security_ftr_clear(SEC_FTR_L1D_FLUSH_UACCESS);
 132         }
 133
 134         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \
 135                  (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR)   || \
 136                   security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV));
 137
 138         setup_rfi_flush(type, enable);
 139         setup_count_cache_flush();
 140
 141         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 142                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_ENTRY);
 143         setup_entry_flush(enable);
 144
 145         enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) &&
 146                  security_ftr_enabled(SEC_FTR_L1D_FLUSH_UACCESS);
 147         setup_uaccess_flush(enable);
 148
 149         setup_stf_barrier();
 150 }
 151
 152 static void __init pnv_check_guarded_cores(void)
 153 {
 154         struct device_node *dn;
 155         int bad_count = 0;
 156
 157         for_each_node_by_type(dn, "cpu") {
 158                 if (of_property_match_string(dn, "status", "bad") >= 0)
 159                         bad_count++;
 160         };
 161
 162         if (bad_count) {
 163                 printk("  _     _______________\n");
 164                 pr_cont(" | |   /               \\\n");
 165                 pr_cont(" | |   |    WARNING!   |\n");
 166                 pr_cont(" | |   |               |\n");
 167                 pr_cont(" | |   | It looks like |\n");
 168                 pr_cont(" |_|   |  you have %*d |\n", 3, bad_count);
 169                 pr_cont("  _    | guarded cores |\n");
 170                 pr_cont(" (_)   \\_______________/\n");
 171         }
 172 }
 173
 174 static void __init pnv_setup_arch(void)
 175 {
 176         set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT);
 177
 178         pnv_setup_security_mitigations();
 179
 180         /* Initialize SMP */
 181         pnv_smp_init();
 182
 183         /* Setup PCI */
 184         pnv_pci_init();
 185
 186         /* Setup RTC and NVRAM callbacks */
 187         if (firmware_has_feature(FW_FEATURE_OPAL))
 188                 opal_nvram_init();
 189
 190         /* Enable NAP mode */
 191         powersave_nap = 1;
 192
 193         pnv_check_guarded_cores();
 194
 195         /* XXX PMCS */
 196
 197         pnv_rng_init();
 198 }
 199
 200 static void __init pnv_init(void)
 201 {
 202         /*
 203          * Initialize the LPC bus now so that legacy serial
 204          * ports can be found on it
 205          */
 206         opal_lpc_init();
 207
 208 #ifdef CONFIG_HVC_OPAL
 209         if (firmware_has_feature(FW_FEATURE_OPAL))
 210                 hvc_opal_init_early();
 211         else
 212 #endif
 213                 add_preferred_console("hvc", 0, NULL);
 214
 215         if (!radix_enabled()) {
 216                 size_t size = sizeof(struct slb_entry) * mmu_slb_size;
 217                 int i;
 218
 219                 /* Allocate per cpu area to save old slb contents during MCE */
 220                 for_each_possible_cpu(i) {
 221                         paca_ptrs[i]->mce_faulty_slbs =
 222                                         memblock_alloc_node(size,
 223                                                 __alignof__(struct slb_entry),
 224                                                 cpu_to_node(i));
 225                 }
 226         }
 227 }
 228
 229 static void __init pnv_init_IRQ(void)
 230 {
 231         /* Try using a XIVE if available, otherwise use a XICS */
 232         if (!xive_native_init())
 233                 xics_init();
 234
 235         WARN_ON(!ppc_md.get_irq);
 236 }
 237
 238 static void pnv_show_cpuinfo(struct seq_file *m)
 239 {
 240         struct device_node *root;
 241         const char *model = "";
 242
 243         root = of_find_node_by_path("/");
 244         if (root)
 245                 model = of_get_property(root, "model", NULL);
 246         seq_printf(m, "machine\t\t: PowerNV %s\n", model);
 247         if (firmware_has_feature(FW_FEATURE_OPAL))
 248                 seq_printf(m, "firmware\t: OPAL\n");
 249         else
 250                 seq_printf(m, "firmware\t: BML\n");
 251         of_node_put(root);
 252         if (radix_enabled())
 253                 seq_printf(m, "MMU\t\t: Radix\n");
 254         else
 255                 seq_printf(m, "MMU\t\t: Hash\n");
 256 }
 257
 258 static void pnv_prepare_going_down(void)
 259 {
 260         /*
 261          * Disable all notifiers from OPAL, we can't
 262          * service interrupts anymore anyway
 263          */
 264         opal_event_shutdown();
 265
 266         /* Print flash update message if one is scheduled. */
 267         opal_flash_update_print_message();
 268
 269         smp_send_stop();
 270
 271         hard_irq_disable();
 272 }
 273
 274 static void  __noreturn pnv_restart(char *cmd)
 275 {
 276         long rc;
 277
 278         pnv_prepare_going_down();
 279
 280         do {
 281                 if (!cmd || !strlen(cmd))
 282                         rc = opal_cec_reboot();
 283                 else if (strcmp(cmd, "full") == 0)
 284                         rc = opal_cec_reboot2(OPAL_REBOOT_FULL_IPL, NULL);
 285                 else if (strcmp(cmd, "mpipl") == 0)
 286                         rc = opal_cec_reboot2(OPAL_REBOOT_MPIPL, NULL);
 287                 else if (strcmp(cmd, "error") == 0)
 288                         rc = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, NULL);
 289                 else if (strcmp(cmd, "fast") == 0)
 290                         rc = opal_cec_reboot2(OPAL_REBOOT_FAST, NULL);
 291                 else
 292                         rc = OPAL_UNSUPPORTED;
 293
 294                 if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 295                         /* Opal is busy wait for some time and retry */
 296                         opal_poll_events(NULL);
 297                         mdelay(10);
 298
 299                 } else  if (cmd && rc) {
 300                         /* Unknown error while issuing reboot */
 301                         if (rc == OPAL_UNSUPPORTED)
 302                                 pr_err("Unsupported '%s' reboot.\n", cmd);
 303                         else
 304                                 pr_err("Unable to issue '%s' reboot. Err=%ld\n",
 305                                        cmd, rc);
 306                         pr_info("Forcing a cec-reboot\n");
 307                         cmd = NULL;
 308                         rc = OPAL_BUSY;
 309
 310                 } else if (rc != OPAL_SUCCESS) {
 311                         /* Unknown error while issuing cec-reboot */
 312                         pr_err("Unable to reboot. Err=%ld\n", rc);
 313                 }
 314
 315         } while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT);
 316
 317         for (;;)
 318                 opal_poll_events(NULL);
 319 }
 320
 321 static void __noreturn pnv_power_off(void)
 322 {
 323         long rc = OPAL_BUSY;
 324
 325         pnv_prepare_going_down();
 326
 327         while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) {
 328                 rc = opal_cec_power_down(0);
 329                 if (rc == OPAL_BUSY_EVENT)
 330                         opal_poll_events(NULL);
 331                 else
 332                         mdelay(10);
 333         }
 334         for (;;)
 335                 opal_poll_events(NULL);
 336 }
 337
 338 static void __noreturn pnv_halt(void)
 339 {
 340         pnv_power_off();
 341 }
 342
 343 static void pnv_progress(char *s, unsigned short hex)
 344 {
 345 }
 346
 347 static void pnv_shutdown(void)
 348 {
 349         /* Let the PCI code clear up IODA tables */
 350         pnv_pci_shutdown();
 351
 352         /*
 353          * Stop OPAL activity: Unregister all OPAL interrupts so they
 354          * don't fire up while we kexec and make sure all potentially
 355          * DMA'ing ops are complete (such as dump retrieval).
 356          */
 357         opal_shutdown();
 358 }
 359
 360 #ifdef CONFIG_KEXEC_CORE
 361 static void pnv_kexec_wait_secondaries_down(void)
 362 {
 363         int my_cpu, i, notified = -1;
 364
 365         my_cpu = get_cpu();
 366
 367         for_each_online_cpu(i) {
 368                 uint8_t status;
 369                 int64_t rc, timeout = 1000;
 370
 371                 if (i == my_cpu)
 372                         continue;
 373
 374                 for (;;) {
 375                         rc = opal_query_cpu_status(get_hard_smp_processor_id(i),
 376                                                    &status);
 377                         if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED)
 378                                 break;
 379                         barrier();
 380                         if (i != notified) {
 381                                 printk(KERN_INFO "kexec: waiting for cpu %d "
 382                                        "(physical %d) to enter OPAL\n",
 383                                        i, paca_ptrs[i]->hw_cpu_id);
 384                                 notified = i;
 385                         }
 386
 387                         /*
 388                          * On crash secondaries might be unreachable or hung,
 389                          * so timeout if we've waited too long
 390                          * */
 391                         mdelay(1);
 392                         if (timeout-- == 0) {
 393                                 printk(KERN_ERR "kexec: timed out waiting for "
 394                                        "cpu %d (physical %d) to enter OPAL\n",
 395                                        i, paca_ptrs[i]->hw_cpu_id);
 396                                 break;
 397                         }
 398                 }
 399         }
 400 }
 401
 402 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
 403 {
 404         u64 reinit_flags;
 405
 406         if (xive_enabled())
 407                 xive_teardown_cpu();
 408         else
 409                 xics_kexec_teardown_cpu(secondary);
 410
 411         /* On OPAL, we return all CPUs to firmware */
 412         if (!firmware_has_feature(FW_FEATURE_OPAL))
 413                 return;
 414
 415         if (secondary) {
 416                 /* Return secondary CPUs to firmware on OPAL v3 */
 417                 mb();
 418                 get_paca()->kexec_state = KEXEC_STATE_REAL_MODE;
 419                 mb();
 420
 421                 /* Return the CPU to OPAL */
 422                 opal_return_cpu();
 423         } else {
 424                 /* Primary waits for the secondaries to have reached OPAL */
 425                 pnv_kexec_wait_secondaries_down();
 426
 427                 /* Switch XIVE back to emulation mode */
 428                 if (xive_enabled())
 429                         xive_shutdown();
 430
 431                 /*
 432                  * We might be running as little-endian - now that interrupts
 433                  * are disabled, reset the HILE bit to big-endian so we don't
 434                  * take interrupts in the wrong endian later
 435                  *
 436                  * We reinit to enable both radix and hash on P9 to ensure
 437                  * the mode used by the next kernel is always supported.
 438                  */
 439                 reinit_flags = OPAL_REINIT_CPUS_HILE_BE;
 440                 if (cpu_has_feature(CPU_FTR_ARCH_300))
 441                         reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX |
 442                                 OPAL_REINIT_CPUS_MMU_HASH;
 443                 opal_reinit_cpus(reinit_flags);
 444         }
 445 }
 446 #endif /* CONFIG_KEXEC_CORE */
 447
 448 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 449 static unsigned long pnv_memory_block_size(void)
 450 {
 451         /*
 452          * We map the kernel linear region with 1GB large pages on radix. For
 453          * memory hot unplug to work our memory block size must be at least
 454          * this size.
 455          */
 456         if (radix_enabled())
 457                 return radix_mem_block_size;
 458         else
 459                 return 256UL * 1024 * 1024;
 460 }
 461 #endif
 462
 463 static void __init pnv_setup_machdep_opal(void)
 464 {
 465         ppc_md.get_boot_time = opal_get_boot_time;
 466         ppc_md.restart = pnv_restart;
 467         pm_power_off = pnv_power_off;
 468         ppc_md.halt = pnv_halt;
 469         /* ppc_md.system_reset_exception gets filled in by pnv_smp_init() */
 470         ppc_md.machine_check_exception = opal_machine_check;
 471         ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery;
 472         if (opal_check_token(OPAL_HANDLE_HMI2))
 473                 ppc_md.hmi_exception_early = opal_hmi_exception_early2;
 474         else
 475                 ppc_md.hmi_exception_early = opal_hmi_exception_early;
 476         ppc_md.handle_hmi_exception = opal_handle_hmi_exception;
 477 }
 478
 479 static int __init pnv_probe(void)
 480 {
 481         if (!of_machine_is_compatible("ibm,powernv"))
 482                 return 0;
 483
 484         if (firmware_has_feature(FW_FEATURE_OPAL))
 485                 pnv_setup_machdep_opal();
 486
 487         pr_debug("PowerNV detected !\n");
 488
 489         pnv_init();
 490
 491         return 1;
 492 }
 493
 494 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 495 void __init pnv_tm_init(void)
 496 {
 497         if (!firmware_has_feature(FW_FEATURE_OPAL) ||
 498             !pvr_version_is(PVR_POWER9) ||
 499             early_cpu_has_feature(CPU_FTR_TM))
 500                 return;
 501
 502         if (opal_reinit_cpus(OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED) != OPAL_SUCCESS)
 503                 return;
 504
 505         pr_info("Enabling TM (Transactional Memory) with Suspend Disabled\n");
 506         cur_cpu_spec->cpu_features |= CPU_FTR_TM;
 507         /* Make sure "normal" HTM is off (it should be) */
 508         cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_HTM;
 509         /* Turn on no suspend mode, and HTM no SC */
 510         cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_HTM_NO_SUSPEND | \
 511                                             PPC_FEATURE2_HTM_NOSC;
 512         tm_suspend_disabled = true;
 513 }
 514 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
 515
 516 /*
 517  * Returns the cpu frequency for 'cpu' in Hz. This is used by
 518  * /proc/cpuinfo
 519  */
 520 static unsigned long pnv_get_proc_freq(unsigned int cpu)
 521 {
 522         unsigned long ret_freq;
 523
 524         ret_freq = cpufreq_get(cpu) * 1000ul;
 525
 526         /*
 527          * If the backend cpufreq driver does not exist,
 528          * then fallback to old way of reporting the clockrate.
 529          */
 530         if (!ret_freq)
 531                 ret_freq = ppc_proc_freq;
 532         return ret_freq;
 533 }
 534
 535 static long pnv_machine_check_early(struct pt_regs *regs)
 536 {
 537         long handled = 0;
 538
 539         if (cur_cpu_spec && cur_cpu_spec->machine_check_early)
 540                 handled = cur_cpu_spec->machine_check_early(regs);
 541
 542         return handled;
 543 }
 544
 545 define_machine(powernv) {
 546         .name                   = "PowerNV",
 547         .probe                  = pnv_probe,
 548         .setup_arch             = pnv_setup_arch,
 549         .init_IRQ               = pnv_init_IRQ,
 550         .show_cpuinfo           = pnv_show_cpuinfo,
 551         .get_proc_freq          = pnv_get_proc_freq,
 552         .progress               = pnv_progress,
 553         .machine_shutdown       = pnv_shutdown,
 554         .power_save             = NULL,
 555         .calibrate_decr         = generic_calibrate_decr,
 556         .machine_check_early    = pnv_machine_check_early,
 557 #ifdef CONFIG_KEXEC_CORE
 558         .kexec_cpu_down         = pnv_kexec_cpu_down,
 559 #endif
 560 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 561         .memory_block_size      = pnv_memory_block_size,
 562 #endif
 563 };