3 * Common boot and setup code.
5 * Copyright (C) 2001 PPC64 Team, IBM Corp
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.
15 #include <linux/export.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/memblock.h>
38 #include <linux/memory.h>
39 #include <linux/nmi.h>
42 #include <asm/kdump.h>
44 #include <asm/processor.h>
45 #include <asm/pgtable.h>
48 #include <asm/machdep.h>
51 #include <asm/cputable.h>
52 #include <asm/sections.h>
53 #include <asm/btext.h>
54 #include <asm/nvram.h>
55 #include <asm/setup.h>
57 #include <asm/iommu.h>
58 #include <asm/serial.h>
59 #include <asm/cache.h>
62 #include <asm/firmware.h>
65 #include <asm/kexec.h>
66 #include <asm/code-patching.h>
67 #include <asm/livepatch.h>
69 #include <asm/cputhreads.h>
72 #define DBG(fmt...) udbg_printf(fmt)
77 int spinning_secondaries;
80 struct ppc64_caches ppc64_caches = {
90 EXPORT_SYMBOL_GPL(ppc64_caches);
92 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
93 void __init setup_tlb_core_data(void)
97 BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
99 for_each_possible_cpu(cpu) {
100 int first = cpu_first_thread_sibling(cpu);
103 * If we boot via kdump on a non-primary thread,
104 * make sure we point at the thread that actually
107 if (cpu_first_thread_sibling(boot_cpuid) == first)
110 paca[cpu].tcd_ptr = &paca[first].tcd;
113 * If we have threads, we need either tlbsrx.
114 * or e6500 tablewalk mode, or else TLB handlers
115 * will be racy and could produce duplicate entries.
117 if (smt_enabled_at_boot >= 2 &&
118 !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
119 book3e_htw_mode != PPC_HTW_E6500) {
120 /* Should we panic instead? */
121 WARN_ONCE("%s: unsupported MMU configuration -- expect problems\n",
130 static char *smt_enabled_cmdline;
132 /* Look for ibm,smt-enabled OF option */
133 void __init check_smt_enabled(void)
135 struct device_node *dn;
136 const char *smt_option;
138 /* Default to enabling all threads */
139 smt_enabled_at_boot = threads_per_core;
141 /* Allow the command line to overrule the OF option */
142 if (smt_enabled_cmdline) {
143 if (!strcmp(smt_enabled_cmdline, "on"))
144 smt_enabled_at_boot = threads_per_core;
145 else if (!strcmp(smt_enabled_cmdline, "off"))
146 smt_enabled_at_boot = 0;
151 rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
153 smt_enabled_at_boot =
154 min(threads_per_core, smt);
157 dn = of_find_node_by_path("/options");
159 smt_option = of_get_property(dn, "ibm,smt-enabled",
163 if (!strcmp(smt_option, "on"))
164 smt_enabled_at_boot = threads_per_core;
165 else if (!strcmp(smt_option, "off"))
166 smt_enabled_at_boot = 0;
174 /* Look for smt-enabled= cmdline option */
175 static int __init early_smt_enabled(char *p)
177 smt_enabled_cmdline = p;
180 early_param("smt-enabled", early_smt_enabled);
182 #endif /* CONFIG_SMP */
184 /** Fix up paca fields required for the boot cpu */
185 static void __init fixup_boot_paca(void)
187 /* The boot cpu is started */
188 get_paca()->cpu_start = 1;
189 /* Allow percpu accesses to work until we setup percpu data */
190 get_paca()->data_offset = 0;
193 static void __init configure_exceptions(void)
196 * Setup the trampolines from the lowmem exception vectors
197 * to the kdump kernel when not using a relocatable kernel.
199 setup_kdump_trampoline();
201 /* Under a PAPR hypervisor, we need hypercalls */
202 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
203 /* Enable AIL if possible */
204 pseries_enable_reloc_on_exc();
207 * Tell the hypervisor that we want our exceptions to
208 * be taken in little endian mode.
210 * We don't call this for big endian as our calling convention
211 * makes us always enter in BE, and the call may fail under
212 * some circumstances with kdump.
214 #ifdef __LITTLE_ENDIAN__
215 pseries_little_endian_exceptions();
218 /* Set endian mode using OPAL */
219 if (firmware_has_feature(FW_FEATURE_OPAL))
220 opal_configure_cores();
222 /* AIL on native is done in cpu_ready_for_interrupts() */
226 static void cpu_ready_for_interrupts(void)
229 * Enable AIL if supported, and we are in hypervisor mode. This
230 * is called once for every processor.
232 * If we are not in hypervisor mode the job is done once for
233 * the whole partition in configure_exceptions().
235 if (early_cpu_has_feature(CPU_FTR_HVMODE) &&
236 early_cpu_has_feature(CPU_FTR_ARCH_207S)) {
237 unsigned long lpcr = mfspr(SPRN_LPCR);
238 mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
241 /* Set IR and DR in PACA MSR */
242 get_paca()->kernel_msr = MSR_KERNEL;
246 * Early initialization entry point. This is called by head.S
247 * with MMU translation disabled. We rely on the "feature" of
248 * the CPU that ignores the top 2 bits of the address in real
249 * mode so we can access kernel globals normally provided we
250 * only toy with things in the RMO region. From here, we do
251 * some early parsing of the device-tree to setup out MEMBLOCK
252 * data structures, and allocate & initialize the hash table
253 * and segment tables so we can start running with translation
256 * It is this function which will call the probe() callback of
257 * the various platform types and copy the matching one to the
258 * global ppc_md structure. Your platform can eventually do
259 * some very early initializations from the probe() routine, but
260 * this is not recommended, be very careful as, for example, the
261 * device-tree is not accessible via normal means at this point.
264 void __init early_setup(unsigned long dt_ptr)
266 static __initdata struct paca_struct boot_paca;
268 /* -------- printk is _NOT_ safe to use here ! ------- */
270 /* Identify CPU type */
271 identify_cpu(0, mfspr(SPRN_PVR));
273 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
274 initialise_paca(&boot_paca, 0);
275 setup_paca(&boot_paca);
278 /* -------- printk is now safe to use ------- */
280 /* Enable early debugging if any specified (see udbg.h) */
283 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
286 * Do early initialization using the flattened device
287 * tree, such as retrieving the physical memory map or
288 * calculating/retrieving the hash table size.
290 early_init_devtree(__va(dt_ptr));
292 /* Now we know the logical id of our boot cpu, setup the paca. */
293 setup_paca(&paca[boot_cpuid]);
297 * Configure exception handlers. This include setting up trampolines
298 * if needed, setting exception endian mode, etc...
300 configure_exceptions();
302 /* Apply all the dynamic patching */
303 apply_feature_fixups();
304 setup_feature_keys();
306 /* Initialize the hash table or TLB handling */
310 * At this point, we can let interrupts switch to virtual mode
311 * (the MMU has been setup), so adjust the MSR in the PACA to
312 * have IR and DR set and enable AIL if it exists
314 cpu_ready_for_interrupts();
316 DBG(" <- early_setup()\n");
318 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
320 * This needs to be done *last* (after the above DBG() even)
322 * Right after we return from this function, we turn on the MMU
323 * which means the real-mode access trick that btext does will
324 * no longer work, it needs to switch to using a real MMU
325 * mapping. This call will ensure that it does
328 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
332 void early_setup_secondary(void)
334 /* Mark interrupts disabled in PACA */
335 get_paca()->soft_enabled = 0;
337 /* Initialize the hash table or TLB handling */
338 early_init_mmu_secondary();
341 * At this point, we can let interrupts switch to virtual mode
342 * (the MMU has been setup), so adjust the MSR in the PACA to
343 * have IR and DR set.
345 cpu_ready_for_interrupts();
348 #endif /* CONFIG_SMP */
350 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC_CORE)
351 static bool use_spinloop(void)
353 if (!IS_ENABLED(CONFIG_PPC_BOOK3E))
357 * When book3e boots from kexec, the ePAPR spin table does
360 return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
363 void smp_release_cpus(void)
371 DBG(" -> smp_release_cpus()\n");
373 /* All secondary cpus are spinning on a common spinloop, release them
374 * all now so they can start to spin on their individual paca
375 * spinloops. For non SMP kernels, the secondary cpus never get out
376 * of the common spinloop.
379 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
381 *ptr = ppc_function_entry(generic_secondary_smp_init);
383 /* And wait a bit for them to catch up */
384 for (i = 0; i < 100000; i++) {
387 if (spinning_secondaries == 0)
391 DBG("spinning_secondaries = %d\n", spinning_secondaries);
393 DBG(" <- smp_release_cpus()\n");
395 #endif /* CONFIG_SMP || CONFIG_KEXEC_CORE */
398 * Initialize some remaining members of the ppc64_caches and systemcfg
400 * (at least until we get rid of them completely). This is mostly some
401 * cache informations about the CPU that will be used by cache flush
402 * routines and/or provided to userland
405 static void init_cache_info(struct ppc_cache_info *info, u32 size, u32 lsize,
410 info->line_size = lsize;
411 info->block_size = bsize;
412 info->log_block_size = __ilog2(bsize);
413 info->blocks_per_page = PAGE_SIZE / bsize;
416 info->assoc = 0xffff;
418 info->assoc = size / (sets * lsize);
421 static bool __init parse_cache_info(struct device_node *np,
423 struct ppc_cache_info *info)
425 static const char *ipropnames[] __initdata = {
428 "i-cache-block-size",
431 static const char *dpropnames[] __initdata = {
434 "d-cache-block-size",
437 const char **propnames = icache ? ipropnames : dpropnames;
438 const __be32 *sizep, *lsizep, *bsizep, *setsp;
439 u32 size, lsize, bsize, sets;
444 lsize = bsize = cur_cpu_spec->dcache_bsize;
445 sizep = of_get_property(np, propnames[0], NULL);
447 size = be32_to_cpu(*sizep);
448 setsp = of_get_property(np, propnames[1], NULL);
450 sets = be32_to_cpu(*setsp);
451 bsizep = of_get_property(np, propnames[2], NULL);
452 lsizep = of_get_property(np, propnames[3], NULL);
456 lsize = be32_to_cpu(*lsizep);
458 bsize = be32_to_cpu(*bsizep);
459 if (sizep == NULL || bsizep == NULL || lsizep == NULL)
463 * OF is weird .. it represents fully associative caches
464 * as "1 way" which doesn't make much sense and doesn't
465 * leave room for direct mapped. We'll assume that 0
466 * in OF means direct mapped for that reason.
473 init_cache_info(info, size, lsize, bsize, sets);
478 void __init initialize_cache_info(void)
480 struct device_node *cpu = NULL, *l2, *l3 = NULL;
483 DBG(" -> initialize_cache_info()\n");
486 * All shipping POWER8 machines have a firmware bug that
487 * puts incorrect information in the device-tree. This will
488 * be (hopefully) fixed for future chips but for now hard
489 * code the values if we are running on one of these
491 pvr = PVR_VER(mfspr(SPRN_PVR));
492 if (pvr == PVR_POWER8 || pvr == PVR_POWER8E ||
493 pvr == PVR_POWER8NVL) {
494 /* size lsize blk sets */
495 init_cache_info(&ppc64_caches.l1i, 0x8000, 128, 128, 32);
496 init_cache_info(&ppc64_caches.l1d, 0x10000, 128, 128, 64);
497 init_cache_info(&ppc64_caches.l2, 0x80000, 128, 0, 512);
498 init_cache_info(&ppc64_caches.l3, 0x800000, 128, 0, 8192);
500 cpu = of_find_node_by_type(NULL, "cpu");
503 * We're assuming *all* of the CPUs have the same
504 * d-cache and i-cache sizes... -Peter
507 if (!parse_cache_info(cpu, false, &ppc64_caches.l1d))
508 DBG("Argh, can't find dcache properties !\n");
510 if (!parse_cache_info(cpu, true, &ppc64_caches.l1i))
511 DBG("Argh, can't find icache properties !\n");
514 * Try to find the L2 and L3 if any. Assume they are
515 * unified and use the D-side properties.
517 l2 = of_find_next_cache_node(cpu);
520 parse_cache_info(l2, false, &ppc64_caches.l2);
521 l3 = of_find_next_cache_node(l2);
525 parse_cache_info(l3, false, &ppc64_caches.l3);
530 /* For use by binfmt_elf */
531 dcache_bsize = ppc64_caches.l1d.block_size;
532 icache_bsize = ppc64_caches.l1i.block_size;
534 DBG(" <- initialize_cache_info()\n");
537 /* This returns the limit below which memory accesses to the linear
538 * mapping are guarnateed not to cause a TLB or SLB miss. This is
539 * used to allocate interrupt or emergency stacks for which our
540 * exception entry path doesn't deal with being interrupted.
542 static __init u64 safe_stack_limit(void)
544 #ifdef CONFIG_PPC_BOOK3E
545 /* Freescale BookE bolts the entire linear mapping */
546 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
547 return linear_map_top;
548 /* Other BookE, we assume the first GB is bolted */
551 /* BookS, the first segment is bolted */
552 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
553 return 1UL << SID_SHIFT_1T;
554 return 1UL << SID_SHIFT;
558 void __init irqstack_early_init(void)
560 u64 limit = safe_stack_limit();
564 * Interrupt stacks must be in the first segment since we
565 * cannot afford to take SLB misses on them.
567 for_each_possible_cpu(i) {
568 softirq_ctx[i] = (struct thread_info *)
569 __va(memblock_alloc_base(THREAD_SIZE,
570 THREAD_SIZE, limit));
571 hardirq_ctx[i] = (struct thread_info *)
572 __va(memblock_alloc_base(THREAD_SIZE,
573 THREAD_SIZE, limit));
577 #ifdef CONFIG_PPC_BOOK3E
578 void __init exc_lvl_early_init(void)
583 for_each_possible_cpu(i) {
584 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
585 critirq_ctx[i] = (struct thread_info *)__va(sp);
586 paca[i].crit_kstack = __va(sp + THREAD_SIZE);
588 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
589 dbgirq_ctx[i] = (struct thread_info *)__va(sp);
590 paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
592 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
593 mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
594 paca[i].mc_kstack = __va(sp + THREAD_SIZE);
597 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
598 patch_exception(0x040, exc_debug_debug_book3e);
603 * Stack space used when we detect a bad kernel stack pointer, and
604 * early in SMP boots before relocation is enabled. Exclusive emergency
605 * stack for machine checks.
607 void __init emergency_stack_init(void)
613 * Emergency stacks must be under 256MB, we cannot afford to take
614 * SLB misses on them. The ABI also requires them to be 128-byte
617 * Since we use these as temporary stacks during secondary CPU
618 * bringup, we need to get at them in real mode. This means they
619 * must also be within the RMO region.
621 limit = min(safe_stack_limit(), ppc64_rma_size);
623 for_each_possible_cpu(i) {
624 struct thread_info *ti;
625 ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
626 klp_init_thread_info(ti);
627 paca[i].emergency_sp = (void *)ti + THREAD_SIZE;
629 #ifdef CONFIG_PPC_BOOK3S_64
630 /* emergency stack for machine check exception handling. */
631 ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
632 klp_init_thread_info(ti);
633 paca[i].mc_emergency_sp = (void *)ti + THREAD_SIZE;
639 #define PCPU_DYN_SIZE ()
641 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
643 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
644 __pa(MAX_DMA_ADDRESS));
647 static void __init pcpu_fc_free(void *ptr, size_t size)
649 free_bootmem(__pa(ptr), size);
652 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
654 if (cpu_to_node(from) == cpu_to_node(to))
655 return LOCAL_DISTANCE;
657 return REMOTE_DISTANCE;
660 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
661 EXPORT_SYMBOL(__per_cpu_offset);
663 void __init setup_per_cpu_areas(void)
665 const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
672 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
673 * to group units. For larger mappings, use 1M atom which
674 * should be large enough to contain a number of units.
676 if (mmu_linear_psize == MMU_PAGE_4K)
677 atom_size = PAGE_SIZE;
681 rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
682 pcpu_fc_alloc, pcpu_fc_free);
684 panic("cannot initialize percpu area (err=%d)", rc);
686 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
687 for_each_possible_cpu(cpu) {
688 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
689 paca[cpu].data_offset = __per_cpu_offset[cpu];
694 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
695 unsigned long memory_block_size_bytes(void)
697 if (ppc_md.memory_block_size)
698 return ppc_md.memory_block_size();
700 return MIN_MEMORY_BLOCK_SIZE;
704 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
705 struct ppc_pci_io ppc_pci_io;
706 EXPORT_SYMBOL(ppc_pci_io);
709 #ifdef CONFIG_HARDLOCKUP_DETECTOR
710 u64 hw_nmi_get_sample_period(int watchdog_thresh)
712 return ppc_proc_freq * watchdog_thresh;
716 * The hardlockup detector breaks PMU event based branches and is likely
717 * to get false positives in KVM guests, so disable it by default.
719 static int __init disable_hardlockup_detector(void)
721 hardlockup_detector_disable();
725 early_initcall(disable_hardlockup_detector);