1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * 4 * Common boot and setup code. 5 * 6 * Copyright (C) 2001 PPC64 Team, IBM Corp 7 */ 8 9 #include <linux/export.h> 10 #include <linux/string.h> 11 #include <linux/sched.h> 12 #include <linux/init.h> 13 #include <linux/kernel.h> 14 #include <linux/reboot.h> 15 #include <linux/delay.h> 16 #include <linux/initrd.h> 17 #include <linux/seq_file.h> 18 #include <linux/ioport.h> 19 #include <linux/console.h> 20 #include <linux/utsname.h> 21 #include <linux/tty.h> 22 #include <linux/root_dev.h> 23 #include <linux/notifier.h> 24 #include <linux/cpu.h> 25 #include <linux/unistd.h> 26 #include <linux/serial.h> 27 #include <linux/serial_8250.h> 28 #include <linux/memblock.h> 29 #include <linux/pci.h> 30 #include <linux/lockdep.h> 31 #include <linux/memory.h> 32 #include <linux/nmi.h> 33 #include <linux/pgtable.h> 34 #include <linux/of.h> 35 #include <linux/of_fdt.h> 36 37 #include <asm/asm-prototypes.h> 38 #include <asm/kvm_guest.h> 39 #include <asm/io.h> 40 #include <asm/kdump.h> 41 #include <asm/processor.h> 42 #include <asm/smp.h> 43 #include <asm/elf.h> 44 #include <asm/machdep.h> 45 #include <asm/paca.h> 46 #include <asm/time.h> 47 #include <asm/cputable.h> 48 #include <asm/dt_cpu_ftrs.h> 49 #include <asm/sections.h> 50 #include <asm/btext.h> 51 #include <asm/nvram.h> 52 #include <asm/setup.h> 53 #include <asm/rtas.h> 54 #include <asm/iommu.h> 55 #include <asm/serial.h> 56 #include <asm/cache.h> 57 #include <asm/page.h> 58 #include <asm/mmu.h> 59 #include <asm/firmware.h> 60 #include <asm/xmon.h> 61 #include <asm/udbg.h> 62 #include <asm/kexec.h> 63 #include <asm/code-patching.h> 64 #include <asm/ftrace.h> 65 #include <asm/opal.h> 66 #include <asm/cputhreads.h> 67 #include <asm/hw_irq.h> 68 #include <asm/feature-fixups.h> 69 #include <asm/kup.h> 70 #include <asm/early_ioremap.h> 71 #include <asm/pgalloc.h> 72 73 #include "setup.h" 74 75 int spinning_secondaries; 76 u64 ppc64_pft_size; 77 78 struct ppc64_caches ppc64_caches = { 79 .l1d = { 80 .block_size = 0x40, 81 .log_block_size = 6, 82 }, 83 .l1i = { 84 .block_size = 0x40, 85 .log_block_size = 6 86 }, 87 }; 88 EXPORT_SYMBOL_GPL(ppc64_caches); 89 90 #if defined(CONFIG_PPC_BOOK3E_64) && defined(CONFIG_SMP) 91 void __init setup_tlb_core_data(void) 92 { 93 int cpu; 94 95 BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0); 96 97 for_each_possible_cpu(cpu) { 98 int first = cpu_first_thread_sibling(cpu); 99 100 /* 101 * If we boot via kdump on a non-primary thread, 102 * make sure we point at the thread that actually 103 * set up this TLB. 104 */ 105 if (cpu_first_thread_sibling(boot_cpuid) == first) 106 first = boot_cpuid; 107 108 paca_ptrs[cpu]->tcd_ptr = &paca_ptrs[first]->tcd; 109 110 /* 111 * If we have threads, we need either tlbsrx. 112 * or e6500 tablewalk mode, or else TLB handlers 113 * will be racy and could produce duplicate entries. 114 * Should we panic instead? 115 */ 116 WARN_ONCE(smt_enabled_at_boot >= 2 && 117 book3e_htw_mode != PPC_HTW_E6500, 118 "%s: unsupported MMU configuration\n", __func__); 119 } 120 } 121 #endif 122 123 #ifdef CONFIG_SMP 124 125 static char *smt_enabled_cmdline; 126 127 /* Look for ibm,smt-enabled OF option */ 128 void __init check_smt_enabled(void) 129 { 130 struct device_node *dn; 131 const char *smt_option; 132 133 /* Default to enabling all threads */ 134 smt_enabled_at_boot = threads_per_core; 135 136 /* Allow the command line to overrule the OF option */ 137 if (smt_enabled_cmdline) { 138 if (!strcmp(smt_enabled_cmdline, "on")) 139 smt_enabled_at_boot = threads_per_core; 140 else if (!strcmp(smt_enabled_cmdline, "off")) 141 smt_enabled_at_boot = 0; 142 else { 143 int smt; 144 int rc; 145 146 rc = kstrtoint(smt_enabled_cmdline, 10, &smt); 147 if (!rc) 148 smt_enabled_at_boot = 149 min(threads_per_core, smt); 150 } 151 } else { 152 dn = of_find_node_by_path("/options"); 153 if (dn) { 154 smt_option = of_get_property(dn, "ibm,smt-enabled", 155 NULL); 156 157 if (smt_option) { 158 if (!strcmp(smt_option, "on")) 159 smt_enabled_at_boot = threads_per_core; 160 else if (!strcmp(smt_option, "off")) 161 smt_enabled_at_boot = 0; 162 } 163 164 of_node_put(dn); 165 } 166 } 167 } 168 169 /* Look for smt-enabled= cmdline option */ 170 static int __init early_smt_enabled(char *p) 171 { 172 smt_enabled_cmdline = p; 173 return 0; 174 } 175 early_param("smt-enabled", early_smt_enabled); 176 177 #endif /* CONFIG_SMP */ 178 179 /** Fix up paca fields required for the boot cpu */ 180 static void __init fixup_boot_paca(struct paca_struct *boot_paca) 181 { 182 /* The boot cpu is started */ 183 boot_paca->cpu_start = 1; 184 #ifdef CONFIG_PPC_BOOK3S_64 185 /* 186 * Give the early boot machine check stack somewhere to use, use 187 * half of the init stack. This is a bit hacky but there should not be 188 * deep stack usage in early init so shouldn't overflow it or overwrite 189 * things. 190 */ 191 boot_paca->mc_emergency_sp = (void *)&init_thread_union + 192 (THREAD_SIZE/2); 193 #endif 194 /* Allow percpu accesses to work until we setup percpu data */ 195 boot_paca->data_offset = 0; 196 /* Mark interrupts soft and hard disabled in PACA */ 197 boot_paca->irq_soft_mask = IRQS_DISABLED; 198 boot_paca->irq_happened = PACA_IRQ_HARD_DIS; 199 WARN_ON(mfmsr() & MSR_EE); 200 } 201 202 static void __init configure_exceptions(void) 203 { 204 /* 205 * Setup the trampolines from the lowmem exception vectors 206 * to the kdump kernel when not using a relocatable kernel. 207 */ 208 setup_kdump_trampoline(); 209 210 /* Under a PAPR hypervisor, we need hypercalls */ 211 if (firmware_has_feature(FW_FEATURE_SET_MODE)) { 212 /* 213 * - PR KVM does not support AIL mode interrupts in the host 214 * while a PR guest is running. 215 * 216 * - SCV system call interrupt vectors are only implemented for 217 * AIL mode interrupts. 218 * 219 * - On pseries, AIL mode can only be enabled and disabled 220 * system-wide so when a PR VM is created on a pseries host, 221 * all CPUs of the host are set to AIL=0 mode. 222 * 223 * - Therefore host CPUs must not execute scv while a PR VM 224 * exists. 225 * 226 * - SCV support can not be disabled dynamically because the 227 * feature is advertised to host userspace. Disabling the 228 * facility and emulating it would be possible but is not 229 * implemented. 230 * 231 * - So SCV support is blanket disabled if PR KVM could possibly 232 * run. That is, PR support compiled in, booting on pseries 233 * with hash MMU. 234 */ 235 if (IS_ENABLED(CONFIG_KVM_BOOK3S_PR_POSSIBLE) && !radix_enabled()) { 236 init_task.thread.fscr &= ~FSCR_SCV; 237 cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_SCV; 238 } 239 240 /* Enable AIL if possible */ 241 if (!pseries_enable_reloc_on_exc()) { 242 init_task.thread.fscr &= ~FSCR_SCV; 243 cur_cpu_spec->cpu_user_features2 &= ~PPC_FEATURE2_SCV; 244 } 245 246 /* 247 * Tell the hypervisor that we want our exceptions to 248 * be taken in little endian mode. 249 * 250 * We don't call this for big endian as our calling convention 251 * makes us always enter in BE, and the call may fail under 252 * some circumstances with kdump. 253 */ 254 #ifdef __LITTLE_ENDIAN__ 255 pseries_little_endian_exceptions(); 256 #endif 257 } else { 258 /* Set endian mode using OPAL */ 259 if (firmware_has_feature(FW_FEATURE_OPAL)) 260 opal_configure_cores(); 261 262 /* AIL on native is done in cpu_ready_for_interrupts() */ 263 } 264 } 265 266 static void cpu_ready_for_interrupts(void) 267 { 268 /* 269 * Enable AIL if supported, and we are in hypervisor mode. This 270 * is called once for every processor. 271 * 272 * If we are not in hypervisor mode the job is done once for 273 * the whole partition in configure_exceptions(). 274 */ 275 if (cpu_has_feature(CPU_FTR_HVMODE)) { 276 unsigned long lpcr = mfspr(SPRN_LPCR); 277 unsigned long new_lpcr = lpcr; 278 279 if (cpu_has_feature(CPU_FTR_ARCH_31)) { 280 /* P10 DD1 does not have HAIL */ 281 if (pvr_version_is(PVR_POWER10) && 282 (mfspr(SPRN_PVR) & 0xf00) == 0x100) 283 new_lpcr |= LPCR_AIL_3; 284 else 285 new_lpcr |= LPCR_HAIL; 286 } else if (cpu_has_feature(CPU_FTR_ARCH_207S)) { 287 new_lpcr |= LPCR_AIL_3; 288 } 289 290 if (new_lpcr != lpcr) 291 mtspr(SPRN_LPCR, new_lpcr); 292 } 293 294 /* 295 * Set HFSCR:TM based on CPU features: 296 * In the special case of TM no suspend (P9N DD2.1), Linux is 297 * told TM is off via the dt-ftrs but told to (partially) use 298 * it via OPAL_REINIT_CPUS_TM_SUSPEND_DISABLED. So HFSCR[TM] 299 * will be off from dt-ftrs but we need to turn it on for the 300 * no suspend case. 301 */ 302 if (cpu_has_feature(CPU_FTR_HVMODE)) { 303 if (cpu_has_feature(CPU_FTR_TM_COMP)) 304 mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) | HFSCR_TM); 305 else 306 mtspr(SPRN_HFSCR, mfspr(SPRN_HFSCR) & ~HFSCR_TM); 307 } 308 309 /* Set IR and DR in PACA MSR */ 310 get_paca()->kernel_msr = MSR_KERNEL; 311 } 312 313 unsigned long spr_default_dscr = 0; 314 315 static void __init record_spr_defaults(void) 316 { 317 if (early_cpu_has_feature(CPU_FTR_DSCR)) 318 spr_default_dscr = mfspr(SPRN_DSCR); 319 } 320 321 /* 322 * Early initialization entry point. This is called by head.S 323 * with MMU translation disabled. We rely on the "feature" of 324 * the CPU that ignores the top 2 bits of the address in real 325 * mode so we can access kernel globals normally provided we 326 * only toy with things in the RMO region. From here, we do 327 * some early parsing of the device-tree to setup out MEMBLOCK 328 * data structures, and allocate & initialize the hash table 329 * and segment tables so we can start running with translation 330 * enabled. 331 * 332 * It is this function which will call the probe() callback of 333 * the various platform types and copy the matching one to the 334 * global ppc_md structure. Your platform can eventually do 335 * some very early initializations from the probe() routine, but 336 * this is not recommended, be very careful as, for example, the 337 * device-tree is not accessible via normal means at this point. 338 */ 339 340 void __init early_setup(unsigned long dt_ptr) 341 { 342 static __initdata struct paca_struct boot_paca; 343 344 /* -------- printk is _NOT_ safe to use here ! ------- */ 345 346 /* 347 * Assume we're on cpu 0 for now. 348 * 349 * We need to load a PACA very early for a few reasons. 350 * 351 * The stack protector canary is stored in the paca, so as soon as we 352 * call any stack protected code we need r13 pointing somewhere valid. 353 * 354 * If we are using kcov it will call in_task() in its instrumentation, 355 * which relies on the current task from the PACA. 356 * 357 * dt_cpu_ftrs_init() calls into generic OF/fdt code, as well as 358 * printk(), which can trigger both stack protector and kcov. 359 * 360 * percpu variables and spin locks also use the paca. 361 * 362 * So set up a temporary paca. It will be replaced below once we know 363 * what CPU we are on. 364 */ 365 initialise_paca(&boot_paca, 0); 366 fixup_boot_paca(&boot_paca); 367 WARN_ON(local_paca != 0); 368 setup_paca(&boot_paca); /* install the paca into registers */ 369 370 /* -------- printk is now safe to use ------- */ 371 372 if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && (mfmsr() & MSR_HV)) 373 enable_machine_check(); 374 375 /* Try new device tree based feature discovery ... */ 376 if (!dt_cpu_ftrs_init(__va(dt_ptr))) 377 /* Otherwise use the old style CPU table */ 378 identify_cpu(0, mfspr(SPRN_PVR)); 379 380 /* Enable early debugging if any specified (see udbg.h) */ 381 udbg_early_init(); 382 383 udbg_printf(" -> %s(), dt_ptr: 0x%lx\n", __func__, dt_ptr); 384 385 /* 386 * Do early initialization using the flattened device 387 * tree, such as retrieving the physical memory map or 388 * calculating/retrieving the hash table size. 389 */ 390 early_init_devtree(__va(dt_ptr)); 391 392 /* Now we know the logical id of our boot cpu, setup the paca. */ 393 if (boot_cpuid != 0) { 394 /* Poison paca_ptrs[0] again if it's not the boot cpu */ 395 memset(&paca_ptrs[0], 0x88, sizeof(paca_ptrs[0])); 396 } 397 fixup_boot_paca(paca_ptrs[boot_cpuid]); 398 setup_paca(paca_ptrs[boot_cpuid]); /* install the paca into registers */ 399 400 /* 401 * Configure exception handlers. This include setting up trampolines 402 * if needed, setting exception endian mode, etc... 403 */ 404 configure_exceptions(); 405 406 /* 407 * Configure Kernel Userspace Protection. This needs to happen before 408 * feature fixups for platforms that implement this using features. 409 */ 410 setup_kup(); 411 412 /* Apply all the dynamic patching */ 413 apply_feature_fixups(); 414 setup_feature_keys(); 415 416 /* Initialize the hash table or TLB handling */ 417 early_init_mmu(); 418 419 early_ioremap_setup(); 420 421 /* 422 * After firmware and early platform setup code has set things up, 423 * we note the SPR values for configurable control/performance 424 * registers, and use those as initial defaults. 425 */ 426 record_spr_defaults(); 427 428 /* 429 * At this point, we can let interrupts switch to virtual mode 430 * (the MMU has been setup), so adjust the MSR in the PACA to 431 * have IR and DR set and enable AIL if it exists 432 */ 433 cpu_ready_for_interrupts(); 434 435 /* 436 * We enable ftrace here, but since we only support DYNAMIC_FTRACE, it 437 * will only actually get enabled on the boot cpu much later once 438 * ftrace itself has been initialized. 439 */ 440 this_cpu_enable_ftrace(); 441 442 udbg_printf(" <- %s()\n", __func__); 443 444 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX 445 /* 446 * This needs to be done *last* (after the above udbg_printf() even) 447 * 448 * Right after we return from this function, we turn on the MMU 449 * which means the real-mode access trick that btext does will 450 * no longer work, it needs to switch to using a real MMU 451 * mapping. This call will ensure that it does 452 */ 453 btext_map(); 454 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */ 455 } 456 457 #ifdef CONFIG_SMP 458 void early_setup_secondary(void) 459 { 460 /* Mark interrupts disabled in PACA */ 461 irq_soft_mask_set(IRQS_DISABLED); 462 463 /* Initialize the hash table or TLB handling */ 464 early_init_mmu_secondary(); 465 466 /* Perform any KUP setup that is per-cpu */ 467 setup_kup(); 468 469 /* 470 * At this point, we can let interrupts switch to virtual mode 471 * (the MMU has been setup), so adjust the MSR in the PACA to 472 * have IR and DR set. 473 */ 474 cpu_ready_for_interrupts(); 475 } 476 477 #endif /* CONFIG_SMP */ 478 479 void panic_smp_self_stop(void) 480 { 481 hard_irq_disable(); 482 spin_begin(); 483 while (1) 484 spin_cpu_relax(); 485 } 486 487 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE) 488 static bool use_spinloop(void) 489 { 490 if (IS_ENABLED(CONFIG_PPC_BOOK3S)) { 491 /* 492 * See comments in head_64.S -- not all platforms insert 493 * secondaries at __secondary_hold and wait at the spin 494 * loop. 495 */ 496 if (firmware_has_feature(FW_FEATURE_OPAL)) 497 return false; 498 return true; 499 } 500 501 /* 502 * When book3e boots from kexec, the ePAPR spin table does 503 * not get used. 504 */ 505 return of_property_read_bool(of_chosen, "linux,booted-from-kexec"); 506 } 507 508 void smp_release_cpus(void) 509 { 510 unsigned long *ptr; 511 int i; 512 513 if (!use_spinloop()) 514 return; 515 516 /* All secondary cpus are spinning on a common spinloop, release them 517 * all now so they can start to spin on their individual paca 518 * spinloops. For non SMP kernels, the secondary cpus never get out 519 * of the common spinloop. 520 */ 521 522 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop 523 - PHYSICAL_START); 524 *ptr = ppc_function_entry(generic_secondary_smp_init); 525 526 /* And wait a bit for them to catch up */ 527 for (i = 0; i < 100000; i++) { 528 mb(); 529 HMT_low(); 530 if (spinning_secondaries == 0) 531 break; 532 udelay(1); 533 } 534 pr_debug("spinning_secondaries = %d\n", spinning_secondaries); 535 } 536 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */ 537 538 /* 539 * Initialize some remaining members of the ppc64_caches and systemcfg 540 * structures 541 * (at least until we get rid of them completely). This is mostly some 542 * cache informations about the CPU that will be used by cache flush 543 * routines and/or provided to userland 544 */ 545 546 static void __init init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize, 547 u32 bsize, u32 sets) 548 { 549 info->size = size; 550 info->sets = sets; 551 info->line_size = lsize; 552 info->block_size = bsize; 553 info->log_block_size = __ilog2(bsize); 554 if (bsize) 555 info->blocks_per_page = PAGE_SIZE / bsize; 556 else 557 info->blocks_per_page = 0; 558 559 if (sets == 0) 560 info->assoc = 0xffff; 561 else 562 info->assoc = size / (sets * lsize); 563 } 564 565 static bool __init parse_cache_info(struct device_node *np, 566 bool icache, 567 struct ppc_cache_info *info) 568 { 569 static const char *ipropnames[] __initdata = { 570 "i-cache-size", 571 "i-cache-sets", 572 "i-cache-block-size", 573 "i-cache-line-size", 574 }; 575 static const char *dpropnames[] __initdata = { 576 "d-cache-size", 577 "d-cache-sets", 578 "d-cache-block-size", 579 "d-cache-line-size", 580 }; 581 const char **propnames = icache ? ipropnames : dpropnames; 582 const __be32 *sizep, *lsizep, *bsizep, *setsp; 583 u32 size, lsize, bsize, sets; 584 bool success = true; 585 586 size = 0; 587 sets = -1u; 588 lsize = bsize = cur_cpu_spec->dcache_bsize; 589 sizep = of_get_property(np, propnames[0], NULL); 590 if (sizep != NULL) 591 size = be32_to_cpu(*sizep); 592 setsp = of_get_property(np, propnames[1], NULL); 593 if (setsp != NULL) 594 sets = be32_to_cpu(*setsp); 595 bsizep = of_get_property(np, propnames[2], NULL); 596 lsizep = of_get_property(np, propnames[3], NULL); 597 if (bsizep == NULL) 598 bsizep = lsizep; 599 if (lsizep == NULL) 600 lsizep = bsizep; 601 if (lsizep != NULL) 602 lsize = be32_to_cpu(*lsizep); 603 if (bsizep != NULL) 604 bsize = be32_to_cpu(*bsizep); 605 if (sizep == NULL || bsizep == NULL || lsizep == NULL) 606 success = false; 607 608 /* 609 * OF is weird .. it represents fully associative caches 610 * as "1 way" which doesn't make much sense and doesn't 611 * leave room for direct mapped. We'll assume that 0 612 * in OF means direct mapped for that reason. 613 */ 614 if (sets == 1) 615 sets = 0; 616 else if (sets == 0) 617 sets = 1; 618 619 init_cache_info(info, size, lsize, bsize, sets); 620 621 return success; 622 } 623 624 void __init initialize_cache_info(void) 625 { 626 struct device_node *cpu = NULL, *l2, *l3 = NULL; 627 u32 pvr; 628 629 /* 630 * All shipping POWER8 machines have a firmware bug that 631 * puts incorrect information in the device-tree. This will 632 * be (hopefully) fixed for future chips but for now hard 633 * code the values if we are running on one of these 634 */ 635 pvr = PVR_VER(mfspr(SPRN_PVR)); 636 if (pvr == PVR_POWER8 || pvr == PVR_POWER8E || 637 pvr == PVR_POWER8NVL) { 638 /* size lsize blk sets */ 639 init_cache_info(&ppc64_caches.l1i, 0x8000, 128, 128, 32); 640 init_cache_info(&ppc64_caches.l1d, 0x10000, 128, 128, 64); 641 init_cache_info(&ppc64_caches.l2, 0x80000, 128, 0, 512); 642 init_cache_info(&ppc64_caches.l3, 0x800000, 128, 0, 8192); 643 } else 644 cpu = of_find_node_by_type(NULL, "cpu"); 645 646 /* 647 * We're assuming *all* of the CPUs have the same 648 * d-cache and i-cache sizes... -Peter 649 */ 650 if (cpu) { 651 if (!parse_cache_info(cpu, false, &ppc64_caches.l1d)) 652 pr_warn("Argh, can't find dcache properties !\n"); 653 654 if (!parse_cache_info(cpu, true, &ppc64_caches.l1i)) 655 pr_warn("Argh, can't find icache properties !\n"); 656 657 /* 658 * Try to find the L2 and L3 if any. Assume they are 659 * unified and use the D-side properties. 660 */ 661 l2 = of_find_next_cache_node(cpu); 662 of_node_put(cpu); 663 if (l2) { 664 parse_cache_info(l2, false, &ppc64_caches.l2); 665 l3 = of_find_next_cache_node(l2); 666 of_node_put(l2); 667 } 668 if (l3) { 669 parse_cache_info(l3, false, &ppc64_caches.l3); 670 of_node_put(l3); 671 } 672 } 673 674 /* For use by binfmt_elf */ 675 dcache_bsize = ppc64_caches.l1d.block_size; 676 icache_bsize = ppc64_caches.l1i.block_size; 677 678 cur_cpu_spec->dcache_bsize = dcache_bsize; 679 cur_cpu_spec->icache_bsize = icache_bsize; 680 } 681 682 /* 683 * This returns the limit below which memory accesses to the linear 684 * mapping are guarnateed not to cause an architectural exception (e.g., 685 * TLB or SLB miss fault). 686 * 687 * This is used to allocate PACAs and various interrupt stacks that 688 * that are accessed early in interrupt handlers that must not cause 689 * re-entrant interrupts. 690 */ 691 __init u64 ppc64_bolted_size(void) 692 { 693 #ifdef CONFIG_PPC_BOOK3E_64 694 /* Freescale BookE bolts the entire linear mapping */ 695 /* XXX: BookE ppc64_rma_limit setup seems to disagree? */ 696 if (early_mmu_has_feature(MMU_FTR_TYPE_FSL_E)) 697 return linear_map_top; 698 /* Other BookE, we assume the first GB is bolted */ 699 return 1ul << 30; 700 #else 701 /* BookS radix, does not take faults on linear mapping */ 702 if (early_radix_enabled()) 703 return ULONG_MAX; 704 705 /* BookS hash, the first segment is bolted */ 706 if (early_mmu_has_feature(MMU_FTR_1T_SEGMENT)) 707 return 1UL << SID_SHIFT_1T; 708 return 1UL << SID_SHIFT; 709 #endif 710 } 711 712 static void *__init alloc_stack(unsigned long limit, int cpu) 713 { 714 void *ptr; 715 716 BUILD_BUG_ON(STACK_INT_FRAME_SIZE % 16); 717 718 ptr = memblock_alloc_try_nid(THREAD_SIZE, THREAD_ALIGN, 719 MEMBLOCK_LOW_LIMIT, limit, 720 early_cpu_to_node(cpu)); 721 if (!ptr) 722 panic("cannot allocate stacks"); 723 724 return ptr; 725 } 726 727 void __init irqstack_early_init(void) 728 { 729 u64 limit = ppc64_bolted_size(); 730 unsigned int i; 731 732 /* 733 * Interrupt stacks must be in the first segment since we 734 * cannot afford to take SLB misses on them. They are not 735 * accessed in realmode. 736 */ 737 for_each_possible_cpu(i) { 738 softirq_ctx[i] = alloc_stack(limit, i); 739 hardirq_ctx[i] = alloc_stack(limit, i); 740 } 741 } 742 743 #ifdef CONFIG_PPC_BOOK3E_64 744 void __init exc_lvl_early_init(void) 745 { 746 unsigned int i; 747 748 for_each_possible_cpu(i) { 749 void *sp; 750 751 sp = alloc_stack(ULONG_MAX, i); 752 critirq_ctx[i] = sp; 753 paca_ptrs[i]->crit_kstack = sp + THREAD_SIZE; 754 755 sp = alloc_stack(ULONG_MAX, i); 756 dbgirq_ctx[i] = sp; 757 paca_ptrs[i]->dbg_kstack = sp + THREAD_SIZE; 758 759 sp = alloc_stack(ULONG_MAX, i); 760 mcheckirq_ctx[i] = sp; 761 paca_ptrs[i]->mc_kstack = sp + THREAD_SIZE; 762 } 763 764 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC)) 765 patch_exception(0x040, exc_debug_debug_book3e); 766 } 767 #endif 768 769 /* 770 * Stack space used when we detect a bad kernel stack pointer, and 771 * early in SMP boots before relocation is enabled. Exclusive emergency 772 * stack for machine checks. 773 */ 774 void __init emergency_stack_init(void) 775 { 776 u64 limit, mce_limit; 777 unsigned int i; 778 779 /* 780 * Emergency stacks must be under 256MB, we cannot afford to take 781 * SLB misses on them. The ABI also requires them to be 128-byte 782 * aligned. 783 * 784 * Since we use these as temporary stacks during secondary CPU 785 * bringup, machine check, system reset, and HMI, we need to get 786 * at them in real mode. This means they must also be within the RMO 787 * region. 788 * 789 * The IRQ stacks allocated elsewhere in this file are zeroed and 790 * initialized in kernel/irq.c. These are initialized here in order 791 * to have emergency stacks available as early as possible. 792 */ 793 limit = mce_limit = min(ppc64_bolted_size(), ppc64_rma_size); 794 795 /* 796 * Machine check on pseries calls rtas, but can't use the static 797 * rtas_args due to a machine check hitting while the lock is held. 798 * rtas args have to be under 4GB, so the machine check stack is 799 * limited to 4GB so args can be put on stack. 800 */ 801 if (firmware_has_feature(FW_FEATURE_LPAR) && mce_limit > SZ_4G) 802 mce_limit = SZ_4G; 803 804 for_each_possible_cpu(i) { 805 paca_ptrs[i]->emergency_sp = alloc_stack(limit, i) + THREAD_SIZE; 806 807 #ifdef CONFIG_PPC_BOOK3S_64 808 /* emergency stack for NMI exception handling. */ 809 paca_ptrs[i]->nmi_emergency_sp = alloc_stack(limit, i) + THREAD_SIZE; 810 811 /* emergency stack for machine check exception handling. */ 812 paca_ptrs[i]->mc_emergency_sp = alloc_stack(mce_limit, i) + THREAD_SIZE; 813 #endif 814 } 815 } 816 817 #ifdef CONFIG_SMP 818 static int pcpu_cpu_distance(unsigned int from, unsigned int to) 819 { 820 if (early_cpu_to_node(from) == early_cpu_to_node(to)) 821 return LOCAL_DISTANCE; 822 else 823 return REMOTE_DISTANCE; 824 } 825 826 static __init int pcpu_cpu_to_node(int cpu) 827 { 828 return early_cpu_to_node(cpu); 829 } 830 831 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly; 832 EXPORT_SYMBOL(__per_cpu_offset); 833 834 void __init setup_per_cpu_areas(void) 835 { 836 const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE; 837 size_t atom_size; 838 unsigned long delta; 839 unsigned int cpu; 840 int rc = -EINVAL; 841 842 /* 843 * BookE and BookS radix are historical values and should be revisited. 844 */ 845 if (IS_ENABLED(CONFIG_PPC_BOOK3E_64)) { 846 atom_size = SZ_1M; 847 } else if (radix_enabled()) { 848 atom_size = PAGE_SIZE; 849 } else if (IS_ENABLED(CONFIG_PPC_64S_HASH_MMU)) { 850 /* 851 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need 852 * to group units. For larger mappings, use 1M atom which 853 * should be large enough to contain a number of units. 854 */ 855 if (mmu_linear_psize == MMU_PAGE_4K) 856 atom_size = PAGE_SIZE; 857 else 858 atom_size = SZ_1M; 859 } 860 861 if (pcpu_chosen_fc != PCPU_FC_PAGE) { 862 rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance, 863 pcpu_cpu_to_node); 864 if (rc) 865 pr_warn("PERCPU: %s allocator failed (%d), " 866 "falling back to page size\n", 867 pcpu_fc_names[pcpu_chosen_fc], rc); 868 } 869 870 if (rc < 0) 871 rc = pcpu_page_first_chunk(0, pcpu_cpu_to_node); 872 if (rc < 0) 873 panic("cannot initialize percpu area (err=%d)", rc); 874 875 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start; 876 for_each_possible_cpu(cpu) { 877 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu]; 878 paca_ptrs[cpu]->data_offset = __per_cpu_offset[cpu]; 879 } 880 } 881 #endif 882 883 #ifdef CONFIG_MEMORY_HOTPLUG 884 unsigned long memory_block_size_bytes(void) 885 { 886 if (ppc_md.memory_block_size) 887 return ppc_md.memory_block_size(); 888 889 return MIN_MEMORY_BLOCK_SIZE; 890 } 891 #endif 892 893 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO) 894 struct ppc_pci_io ppc_pci_io; 895 EXPORT_SYMBOL(ppc_pci_io); 896 #endif 897 898 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF 899 u64 hw_nmi_get_sample_period(int watchdog_thresh) 900 { 901 return ppc_proc_freq * watchdog_thresh; 902 } 903 #endif 904 905 /* 906 * The perf based hardlockup detector breaks PMU event based branches, so 907 * disable it by default. Book3S has a soft-nmi hardlockup detector based 908 * on the decrementer interrupt, so it does not suffer from this problem. 909 * 910 * It is likely to get false positives in KVM guests, so disable it there 911 * by default too. PowerVM will not stop or arbitrarily oversubscribe 912 * CPUs, but give a minimum regular allotment even with SPLPAR, so enable 913 * the detector for non-KVM guests, assume PowerVM. 914 */ 915 static int __init disable_hardlockup_detector(void) 916 { 917 #ifdef CONFIG_HARDLOCKUP_DETECTOR_PERF 918 hardlockup_detector_disable(); 919 #else 920 if (firmware_has_feature(FW_FEATURE_LPAR)) { 921 if (is_kvm_guest()) 922 hardlockup_detector_disable(); 923 } 924 #endif 925 926 return 0; 927 } 928 early_initcall(disable_hardlockup_detector); 929