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_rfi_flush(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 enable = security_ftr_enabled(SEC_FTR_FAVOUR_SECURITY) && \ 126 (security_ftr_enabled(SEC_FTR_L1D_FLUSH_PR) || \ 127 security_ftr_enabled(SEC_FTR_L1D_FLUSH_HV)); 128 129 setup_rfi_flush(type, enable); 130 setup_count_cache_flush(); 131 } 132 133 static void __init pnv_check_guarded_cores(void) 134 { 135 struct device_node *dn; 136 int bad_count = 0; 137 138 for_each_node_by_type(dn, "cpu") { 139 if (of_property_match_string(dn, "status", "bad") >= 0) 140 bad_count++; 141 }; 142 143 if (bad_count) { 144 printk(" _ _______________\n"); 145 pr_cont(" | | / \\\n"); 146 pr_cont(" | | | WARNING! |\n"); 147 pr_cont(" | | | |\n"); 148 pr_cont(" | | | It looks like |\n"); 149 pr_cont(" |_| | you have %*d |\n", 3, bad_count); 150 pr_cont(" _ | guarded cores |\n"); 151 pr_cont(" (_) \\_______________/\n"); 152 } 153 } 154 155 static void __init pnv_setup_arch(void) 156 { 157 set_arch_panic_timeout(10, ARCH_PANIC_TIMEOUT); 158 159 pnv_setup_rfi_flush(); 160 setup_stf_barrier(); 161 162 /* Initialize SMP */ 163 pnv_smp_init(); 164 165 /* Setup PCI */ 166 pnv_pci_init(); 167 168 /* Setup RTC and NVRAM callbacks */ 169 if (firmware_has_feature(FW_FEATURE_OPAL)) 170 opal_nvram_init(); 171 172 /* Enable NAP mode */ 173 powersave_nap = 1; 174 175 pnv_check_guarded_cores(); 176 177 /* XXX PMCS */ 178 } 179 180 static void __init pnv_init(void) 181 { 182 /* 183 * Initialize the LPC bus now so that legacy serial 184 * ports can be found on it 185 */ 186 opal_lpc_init(); 187 188 #ifdef CONFIG_HVC_OPAL 189 if (firmware_has_feature(FW_FEATURE_OPAL)) 190 hvc_opal_init_early(); 191 else 192 #endif 193 add_preferred_console("hvc", 0, NULL); 194 195 if (!radix_enabled()) { 196 int i; 197 198 /* Allocate per cpu area to save old slb contents during MCE */ 199 for_each_possible_cpu(i) 200 paca_ptrs[i]->mce_faulty_slbs = memblock_alloc_node(mmu_slb_size, __alignof__(*paca_ptrs[i]->mce_faulty_slbs), cpu_to_node(i)); 201 } 202 } 203 204 static void __init pnv_init_IRQ(void) 205 { 206 /* Try using a XIVE if available, otherwise use a XICS */ 207 if (!xive_native_init()) 208 xics_init(); 209 210 WARN_ON(!ppc_md.get_irq); 211 } 212 213 static void pnv_show_cpuinfo(struct seq_file *m) 214 { 215 struct device_node *root; 216 const char *model = ""; 217 218 root = of_find_node_by_path("/"); 219 if (root) 220 model = of_get_property(root, "model", NULL); 221 seq_printf(m, "machine\t\t: PowerNV %s\n", model); 222 if (firmware_has_feature(FW_FEATURE_OPAL)) 223 seq_printf(m, "firmware\t: OPAL\n"); 224 else 225 seq_printf(m, "firmware\t: BML\n"); 226 of_node_put(root); 227 if (radix_enabled()) 228 seq_printf(m, "MMU\t\t: Radix\n"); 229 else 230 seq_printf(m, "MMU\t\t: Hash\n"); 231 } 232 233 static void pnv_prepare_going_down(void) 234 { 235 /* 236 * Disable all notifiers from OPAL, we can't 237 * service interrupts anymore anyway 238 */ 239 opal_event_shutdown(); 240 241 /* Print flash update message if one is scheduled. */ 242 opal_flash_update_print_message(); 243 244 smp_send_stop(); 245 246 hard_irq_disable(); 247 } 248 249 static void __noreturn pnv_restart(char *cmd) 250 { 251 long rc; 252 253 pnv_prepare_going_down(); 254 255 do { 256 if (!cmd || !strlen(cmd)) 257 rc = opal_cec_reboot(); 258 else if (strcmp(cmd, "full") == 0) 259 rc = opal_cec_reboot2(OPAL_REBOOT_FULL_IPL, NULL); 260 else if (strcmp(cmd, "mpipl") == 0) 261 rc = opal_cec_reboot2(OPAL_REBOOT_MPIPL, NULL); 262 else if (strcmp(cmd, "error") == 0) 263 rc = opal_cec_reboot2(OPAL_REBOOT_PLATFORM_ERROR, NULL); 264 else if (strcmp(cmd, "fast") == 0) 265 rc = opal_cec_reboot2(OPAL_REBOOT_FAST, NULL); 266 else 267 rc = OPAL_UNSUPPORTED; 268 269 if (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) { 270 /* Opal is busy wait for some time and retry */ 271 opal_poll_events(NULL); 272 mdelay(10); 273 274 } else if (cmd && rc) { 275 /* Unknown error while issuing reboot */ 276 if (rc == OPAL_UNSUPPORTED) 277 pr_err("Unsupported '%s' reboot.\n", cmd); 278 else 279 pr_err("Unable to issue '%s' reboot. Err=%ld\n", 280 cmd, rc); 281 pr_info("Forcing a cec-reboot\n"); 282 cmd = NULL; 283 rc = OPAL_BUSY; 284 285 } else if (rc != OPAL_SUCCESS) { 286 /* Unknown error while issuing cec-reboot */ 287 pr_err("Unable to reboot. Err=%ld\n", rc); 288 } 289 290 } while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT); 291 292 for (;;) 293 opal_poll_events(NULL); 294 } 295 296 static void __noreturn pnv_power_off(void) 297 { 298 long rc = OPAL_BUSY; 299 300 pnv_prepare_going_down(); 301 302 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) { 303 rc = opal_cec_power_down(0); 304 if (rc == OPAL_BUSY_EVENT) 305 opal_poll_events(NULL); 306 else 307 mdelay(10); 308 } 309 for (;;) 310 opal_poll_events(NULL); 311 } 312 313 static void __noreturn pnv_halt(void) 314 { 315 pnv_power_off(); 316 } 317 318 static void pnv_progress(char *s, unsigned short hex) 319 { 320 } 321 322 static void pnv_shutdown(void) 323 { 324 /* Let the PCI code clear up IODA tables */ 325 pnv_pci_shutdown(); 326 327 /* 328 * Stop OPAL activity: Unregister all OPAL interrupts so they 329 * don't fire up while we kexec and make sure all potentially 330 * DMA'ing ops are complete (such as dump retrieval). 331 */ 332 opal_shutdown(); 333 } 334 335 #ifdef CONFIG_KEXEC_CORE 336 static void pnv_kexec_wait_secondaries_down(void) 337 { 338 int my_cpu, i, notified = -1; 339 340 my_cpu = get_cpu(); 341 342 for_each_online_cpu(i) { 343 uint8_t status; 344 int64_t rc, timeout = 1000; 345 346 if (i == my_cpu) 347 continue; 348 349 for (;;) { 350 rc = opal_query_cpu_status(get_hard_smp_processor_id(i), 351 &status); 352 if (rc != OPAL_SUCCESS || status != OPAL_THREAD_STARTED) 353 break; 354 barrier(); 355 if (i != notified) { 356 printk(KERN_INFO "kexec: waiting for cpu %d " 357 "(physical %d) to enter OPAL\n", 358 i, paca_ptrs[i]->hw_cpu_id); 359 notified = i; 360 } 361 362 /* 363 * On crash secondaries might be unreachable or hung, 364 * so timeout if we've waited too long 365 * */ 366 mdelay(1); 367 if (timeout-- == 0) { 368 printk(KERN_ERR "kexec: timed out waiting for " 369 "cpu %d (physical %d) to enter OPAL\n", 370 i, paca_ptrs[i]->hw_cpu_id); 371 break; 372 } 373 } 374 } 375 } 376 377 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary) 378 { 379 u64 reinit_flags; 380 381 if (xive_enabled()) 382 xive_teardown_cpu(); 383 else 384 xics_kexec_teardown_cpu(secondary); 385 386 /* On OPAL, we return all CPUs to firmware */ 387 if (!firmware_has_feature(FW_FEATURE_OPAL)) 388 return; 389 390 if (secondary) { 391 /* Return secondary CPUs to firmware on OPAL v3 */ 392 mb(); 393 get_paca()->kexec_state = KEXEC_STATE_REAL_MODE; 394 mb(); 395 396 /* Return the CPU to OPAL */ 397 opal_return_cpu(); 398 } else { 399 /* Primary waits for the secondaries to have reached OPAL */ 400 pnv_kexec_wait_secondaries_down(); 401 402 /* Switch XIVE back to emulation mode */ 403 if (xive_enabled()) 404 xive_shutdown(); 405 406 /* 407 * We might be running as little-endian - now that interrupts 408 * are disabled, reset the HILE bit to big-endian so we don't 409 * take interrupts in the wrong endian later 410 * 411 * We reinit to enable both radix and hash on P9 to ensure 412 * the mode used by the next kernel is always supported. 413 */ 414 reinit_flags = OPAL_REINIT_CPUS_HILE_BE; 415 if (cpu_has_feature(CPU_FTR_ARCH_300)) 416 reinit_flags |= OPAL_REINIT_CPUS_MMU_RADIX | 417 OPAL_REINIT_CPUS_MMU_HASH; 418 opal_reinit_cpus(reinit_flags); 419 } 420 } 421 #endif /* CONFIG_KEXEC_CORE */ 422 423 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE 424 static unsigned long pnv_memory_block_size(void) 425 { 426 /* 427 * We map the kernel linear region with 1GB large pages on radix. For 428 * memory hot unplug to work our memory block size must be at least 429 * this size. 430 */ 431 if (radix_enabled()) 432 return radix_mem_block_size; 433 else 434 return 256UL * 1024 * 1024; 435 } 436 #endif 437 438 static void __init pnv_setup_machdep_opal(void) 439 { 440 ppc_md.get_boot_time = opal_get_boot_time; 441 ppc_md.restart = pnv_restart; 442 pm_power_off = pnv_power_off; 443 ppc_md.halt = pnv_halt; 444 /* ppc_md.system_reset_exception gets filled in by pnv_smp_init() */ 445 ppc_md.machine_check_exception = opal_machine_check; 446 ppc_md.mce_check_early_recovery = opal_mce_check_early_recovery; 447 if (opal_check_token(OPAL_HANDLE_HMI2)) 448 ppc_md.hmi_exception_early = opal_hmi_exception_early2; 449 else 450 ppc_md.hmi_exception_early = opal_hmi_exception_early; 451 ppc_md.handle_hmi_exception = opal_handle_hmi_exception; 452 } 453 454 static int __init pnv_probe(void) 455 { 456 if (!of_machine_is_compatible("ibm,powernv")) 457 return 0; 458 459 if (firmware_has_feature(FW_FEATURE_OPAL)) 460 pnv_setup_machdep_opal(); 461 462 pr_debug("PowerNV detected !\n"); 463 464 pnv_init(); 465 466 return 1; 467 } 468 469 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 470 void __init pnv_tm_init(void) 471 { 472 if (!firmware_has_feature(FW_FEATURE_OPAL) || 473 !pvr_version_is(PVR_POWER9) || 474 early_cpu_has_feature(CPU_FTR_TM)) 475 return; 476 477 if (opal_reinit_cpus(OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED) != OPAL_SUCCESS) 478 return; 479 480 pr_info("Enabling TM (Transactional Memory) with Suspend Disabled\n"); 481 cur_cpu_spec->cpu_features |= CPU_FTR_TM; 482 /* Make sure "normal" HTM is off (it should be) */ 483 cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_HTM; 484 /* Turn on no suspend mode, and HTM no SC */ 485 cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_HTM_NO_SUSPEND | \ 486 PPC_FEATURE2_HTM_NOSC; 487 tm_suspend_disabled = true; 488 } 489 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */ 490 491 /* 492 * Returns the cpu frequency for 'cpu' in Hz. This is used by 493 * /proc/cpuinfo 494 */ 495 static unsigned long pnv_get_proc_freq(unsigned int cpu) 496 { 497 unsigned long ret_freq; 498 499 ret_freq = cpufreq_get(cpu) * 1000ul; 500 501 /* 502 * If the backend cpufreq driver does not exist, 503 * then fallback to old way of reporting the clockrate. 504 */ 505 if (!ret_freq) 506 ret_freq = ppc_proc_freq; 507 return ret_freq; 508 } 509 510 static long pnv_machine_check_early(struct pt_regs *regs) 511 { 512 long handled = 0; 513 514 if (cur_cpu_spec && cur_cpu_spec->machine_check_early) 515 handled = cur_cpu_spec->machine_check_early(regs); 516 517 return handled; 518 } 519 520 define_machine(powernv) { 521 .name = "PowerNV", 522 .probe = pnv_probe, 523 .setup_arch = pnv_setup_arch, 524 .init_IRQ = pnv_init_IRQ, 525 .show_cpuinfo = pnv_show_cpuinfo, 526 .get_proc_freq = pnv_get_proc_freq, 527 .progress = pnv_progress, 528 .machine_shutdown = pnv_shutdown, 529 .power_save = NULL, 530 .calibrate_decr = generic_calibrate_decr, 531 .machine_check_early = pnv_machine_check_early, 532 #ifdef CONFIG_KEXEC_CORE 533 .kexec_cpu_down = pnv_kexec_cpu_down, 534 #endif 535 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE 536 .memory_block_size = pnv_memory_block_size, 537 #endif 538 }; 539