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 /* Pick defaults since we might want to patch instructions
81 * before we've read this from the device tree.
83 struct ppc64_caches ppc64_caches = {
89 EXPORT_SYMBOL_GPL(ppc64_caches);
92 * These are used in binfmt_elf.c to put aux entries on the stack
93 * for each elf executable being started.
99 #if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
100 void __init setup_tlb_core_data(void)
104 BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
106 for_each_possible_cpu(cpu) {
107 int first = cpu_first_thread_sibling(cpu);
110 * If we boot via kdump on a non-primary thread,
111 * make sure we point at the thread that actually
114 if (cpu_first_thread_sibling(boot_cpuid) == first)
117 paca[cpu].tcd_ptr = &paca[first].tcd;
120 * If we have threads, we need either tlbsrx.
121 * or e6500 tablewalk mode, or else TLB handlers
122 * will be racy and could produce duplicate entries.
124 if (smt_enabled_at_boot >= 2 &&
125 !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
126 book3e_htw_mode != PPC_HTW_E6500) {
127 /* Should we panic instead? */
128 WARN_ONCE("%s: unsupported MMU configuration -- expect problems\n",
137 static char *smt_enabled_cmdline;
139 /* Look for ibm,smt-enabled OF option */
140 void __init check_smt_enabled(void)
142 struct device_node *dn;
143 const char *smt_option;
145 /* Default to enabling all threads */
146 smt_enabled_at_boot = threads_per_core;
148 /* Allow the command line to overrule the OF option */
149 if (smt_enabled_cmdline) {
150 if (!strcmp(smt_enabled_cmdline, "on"))
151 smt_enabled_at_boot = threads_per_core;
152 else if (!strcmp(smt_enabled_cmdline, "off"))
153 smt_enabled_at_boot = 0;
158 rc = kstrtoint(smt_enabled_cmdline, 10, &smt);
160 smt_enabled_at_boot =
161 min(threads_per_core, smt);
164 dn = of_find_node_by_path("/options");
166 smt_option = of_get_property(dn, "ibm,smt-enabled",
170 if (!strcmp(smt_option, "on"))
171 smt_enabled_at_boot = threads_per_core;
172 else if (!strcmp(smt_option, "off"))
173 smt_enabled_at_boot = 0;
181 /* Look for smt-enabled= cmdline option */
182 static int __init early_smt_enabled(char *p)
184 smt_enabled_cmdline = p;
187 early_param("smt-enabled", early_smt_enabled);
189 #endif /* CONFIG_SMP */
191 /** Fix up paca fields required for the boot cpu */
192 static void __init fixup_boot_paca(void)
194 /* The boot cpu is started */
195 get_paca()->cpu_start = 1;
196 /* Allow percpu accesses to work until we setup percpu data */
197 get_paca()->data_offset = 0;
200 static void __init configure_exceptions(void)
203 * Setup the trampolines from the lowmem exception vectors
204 * to the kdump kernel when not using a relocatable kernel.
206 setup_kdump_trampoline();
208 /* Under a PAPR hypervisor, we need hypercalls */
209 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
210 /* Enable AIL if possible */
211 pseries_enable_reloc_on_exc();
214 * Tell the hypervisor that we want our exceptions to
215 * be taken in little endian mode.
217 * We don't call this for big endian as our calling convention
218 * makes us always enter in BE, and the call may fail under
219 * some circumstances with kdump.
221 #ifdef __LITTLE_ENDIAN__
222 pseries_little_endian_exceptions();
225 /* Set endian mode using OPAL */
226 if (firmware_has_feature(FW_FEATURE_OPAL))
227 opal_configure_cores();
229 /* Enable AIL if supported, and we are in hypervisor mode */
230 if (early_cpu_has_feature(CPU_FTR_HVMODE) &&
231 early_cpu_has_feature(CPU_FTR_ARCH_207S)) {
232 unsigned long lpcr = mfspr(SPRN_LPCR);
233 mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
238 static void cpu_ready_for_interrupts(void)
240 /* Set IR and DR in PACA MSR */
241 get_paca()->kernel_msr = MSR_KERNEL;
245 * Early initialization entry point. This is called by head.S
246 * with MMU translation disabled. We rely on the "feature" of
247 * the CPU that ignores the top 2 bits of the address in real
248 * mode so we can access kernel globals normally provided we
249 * only toy with things in the RMO region. From here, we do
250 * some early parsing of the device-tree to setup out MEMBLOCK
251 * data structures, and allocate & initialize the hash table
252 * and segment tables so we can start running with translation
255 * It is this function which will call the probe() callback of
256 * the various platform types and copy the matching one to the
257 * global ppc_md structure. Your platform can eventually do
258 * some very early initializations from the probe() routine, but
259 * this is not recommended, be very careful as, for example, the
260 * device-tree is not accessible via normal means at this point.
263 void __init early_setup(unsigned long dt_ptr)
265 static __initdata struct paca_struct boot_paca;
267 /* -------- printk is _NOT_ safe to use here ! ------- */
269 /* Identify CPU type */
270 identify_cpu(0, mfspr(SPRN_PVR));
272 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
273 initialise_paca(&boot_paca, 0);
274 setup_paca(&boot_paca);
277 /* -------- printk is now safe to use ------- */
279 /* Enable early debugging if any specified (see udbg.h) */
282 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
285 * Do early initialization using the flattened device
286 * tree, such as retrieving the physical memory map or
287 * calculating/retrieving the hash table size.
289 early_init_devtree(__va(dt_ptr));
291 /* Now we know the logical id of our boot cpu, setup the paca. */
292 setup_paca(&paca[boot_cpuid]);
296 * Configure exception handlers. This include setting up trampolines
297 * if needed, setting exception endian mode, etc...
299 configure_exceptions();
301 /* Apply all the dynamic patching */
302 apply_feature_fixups();
303 setup_feature_keys();
305 /* Initialize the hash table or TLB handling */
309 * At this point, we can let interrupts switch to virtual mode
310 * (the MMU has been setup), so adjust the MSR in the PACA to
311 * have IR and DR set and enable AIL if it exists
313 cpu_ready_for_interrupts();
315 DBG(" <- early_setup()\n");
317 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
319 * This needs to be done *last* (after the above DBG() even)
321 * Right after we return from this function, we turn on the MMU
322 * which means the real-mode access trick that btext does will
323 * no longer work, it needs to switch to using a real MMU
324 * mapping. This call will ensure that it does
327 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
331 void early_setup_secondary(void)
333 /* Mark interrupts disabled in PACA */
334 get_paca()->soft_enabled = 0;
336 /* Initialize the hash table or TLB handling */
337 early_init_mmu_secondary();
340 * At this point, we can let interrupts switch to virtual mode
341 * (the MMU has been setup), so adjust the MSR in the PACA to
342 * have IR and DR set.
344 cpu_ready_for_interrupts();
347 #endif /* CONFIG_SMP */
349 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
350 static bool use_spinloop(void)
352 if (!IS_ENABLED(CONFIG_PPC_BOOK3E))
356 * When book3e boots from kexec, the ePAPR spin table does
359 return of_property_read_bool(of_chosen, "linux,booted-from-kexec");
362 void smp_release_cpus(void)
370 DBG(" -> smp_release_cpus()\n");
372 /* All secondary cpus are spinning on a common spinloop, release them
373 * all now so they can start to spin on their individual paca
374 * spinloops. For non SMP kernels, the secondary cpus never get out
375 * of the common spinloop.
378 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
380 *ptr = ppc_function_entry(generic_secondary_smp_init);
382 /* And wait a bit for them to catch up */
383 for (i = 0; i < 100000; i++) {
386 if (spinning_secondaries == 0)
390 DBG("spinning_secondaries = %d\n", spinning_secondaries);
392 DBG(" <- smp_release_cpus()\n");
394 #endif /* CONFIG_SMP || CONFIG_KEXEC */
397 * Initialize some remaining members of the ppc64_caches and systemcfg
399 * (at least until we get rid of them completely). This is mostly some
400 * cache informations about the CPU that will be used by cache flush
401 * routines and/or provided to userland
403 void __init initialize_cache_info(void)
405 struct device_node *np;
406 unsigned long num_cpus = 0;
408 DBG(" -> initialize_cache_info()\n");
410 for_each_node_by_type(np, "cpu") {
414 * We're assuming *all* of the CPUs have the same
415 * d-cache and i-cache sizes... -Peter
418 const __be32 *sizep, *lsizep;
422 lsize = cur_cpu_spec->dcache_bsize;
423 sizep = of_get_property(np, "d-cache-size", NULL);
425 size = be32_to_cpu(*sizep);
426 lsizep = of_get_property(np, "d-cache-block-size",
428 /* fallback if block size missing */
430 lsizep = of_get_property(np,
434 lsize = be32_to_cpu(*lsizep);
435 if (sizep == NULL || lsizep == NULL)
436 DBG("Argh, can't find dcache properties ! "
437 "sizep: %p, lsizep: %p\n", sizep, lsizep);
439 ppc64_caches.dsize = size;
440 ppc64_caches.dline_size = lsize;
441 ppc64_caches.log_dline_size = __ilog2(lsize);
442 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
445 lsize = cur_cpu_spec->icache_bsize;
446 sizep = of_get_property(np, "i-cache-size", NULL);
448 size = be32_to_cpu(*sizep);
449 lsizep = of_get_property(np, "i-cache-block-size",
452 lsizep = of_get_property(np,
456 lsize = be32_to_cpu(*lsizep);
457 if (sizep == NULL || lsizep == NULL)
458 DBG("Argh, can't find icache properties ! "
459 "sizep: %p, lsizep: %p\n", sizep, lsizep);
461 ppc64_caches.isize = size;
462 ppc64_caches.iline_size = lsize;
463 ppc64_caches.log_iline_size = __ilog2(lsize);
464 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
468 /* For use by binfmt_elf */
469 dcache_bsize = ppc64_caches.dline_size;
470 icache_bsize = ppc64_caches.iline_size;
472 DBG(" <- initialize_cache_info()\n");
475 /* This returns the limit below which memory accesses to the linear
476 * mapping are guarnateed not to cause a TLB or SLB miss. This is
477 * used to allocate interrupt or emergency stacks for which our
478 * exception entry path doesn't deal with being interrupted.
480 static __init u64 safe_stack_limit(void)
482 #ifdef CONFIG_PPC_BOOK3E
483 /* Freescale BookE bolts the entire linear mapping */
484 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
485 return linear_map_top;
486 /* Other BookE, we assume the first GB is bolted */
489 /* BookS, the first segment is bolted */
490 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
491 return 1UL << SID_SHIFT_1T;
492 return 1UL << SID_SHIFT;
496 void __init irqstack_early_init(void)
498 u64 limit = safe_stack_limit();
502 * Interrupt stacks must be in the first segment since we
503 * cannot afford to take SLB misses on them.
505 for_each_possible_cpu(i) {
506 softirq_ctx[i] = (struct thread_info *)
507 __va(memblock_alloc_base(THREAD_SIZE,
508 THREAD_SIZE, limit));
509 hardirq_ctx[i] = (struct thread_info *)
510 __va(memblock_alloc_base(THREAD_SIZE,
511 THREAD_SIZE, limit));
515 #ifdef CONFIG_PPC_BOOK3E
516 void __init exc_lvl_early_init(void)
521 for_each_possible_cpu(i) {
522 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
523 critirq_ctx[i] = (struct thread_info *)__va(sp);
524 paca[i].crit_kstack = __va(sp + THREAD_SIZE);
526 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
527 dbgirq_ctx[i] = (struct thread_info *)__va(sp);
528 paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
530 sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
531 mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
532 paca[i].mc_kstack = __va(sp + THREAD_SIZE);
535 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
536 patch_exception(0x040, exc_debug_debug_book3e);
541 * Stack space used when we detect a bad kernel stack pointer, and
542 * early in SMP boots before relocation is enabled. Exclusive emergency
543 * stack for machine checks.
545 void __init emergency_stack_init(void)
551 * Emergency stacks must be under 256MB, we cannot afford to take
552 * SLB misses on them. The ABI also requires them to be 128-byte
555 * Since we use these as temporary stacks during secondary CPU
556 * bringup, we need to get at them in real mode. This means they
557 * must also be within the RMO region.
559 limit = min(safe_stack_limit(), ppc64_rma_size);
561 for_each_possible_cpu(i) {
562 struct thread_info *ti;
563 ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
564 klp_init_thread_info(ti);
565 paca[i].emergency_sp = (void *)ti + THREAD_SIZE;
567 #ifdef CONFIG_PPC_BOOK3S_64
568 /* emergency stack for machine check exception handling. */
569 ti = __va(memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit));
570 klp_init_thread_info(ti);
571 paca[i].mc_emergency_sp = (void *)ti + THREAD_SIZE;
577 #define PCPU_DYN_SIZE ()
579 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
581 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
582 __pa(MAX_DMA_ADDRESS));
585 static void __init pcpu_fc_free(void *ptr, size_t size)
587 free_bootmem(__pa(ptr), size);
590 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
592 if (cpu_to_node(from) == cpu_to_node(to))
593 return LOCAL_DISTANCE;
595 return REMOTE_DISTANCE;
598 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
599 EXPORT_SYMBOL(__per_cpu_offset);
601 void __init setup_per_cpu_areas(void)
603 const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
610 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
611 * to group units. For larger mappings, use 1M atom which
612 * should be large enough to contain a number of units.
614 if (mmu_linear_psize == MMU_PAGE_4K)
615 atom_size = PAGE_SIZE;
619 rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
620 pcpu_fc_alloc, pcpu_fc_free);
622 panic("cannot initialize percpu area (err=%d)", rc);
624 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
625 for_each_possible_cpu(cpu) {
626 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
627 paca[cpu].data_offset = __per_cpu_offset[cpu];
632 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
633 unsigned long memory_block_size_bytes(void)
635 if (ppc_md.memory_block_size)
636 return ppc_md.memory_block_size();
638 return MIN_MEMORY_BLOCK_SIZE;
642 #if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
643 struct ppc_pci_io ppc_pci_io;
644 EXPORT_SYMBOL(ppc_pci_io);
647 #ifdef CONFIG_HARDLOCKUP_DETECTOR
648 u64 hw_nmi_get_sample_period(int watchdog_thresh)
650 return ppc_proc_freq * watchdog_thresh;
654 * The hardlockup detector breaks PMU event based branches and is likely
655 * to get false positives in KVM guests, so disable it by default.
657 static int __init disable_hardlockup_detector(void)
659 hardlockup_detector_disable();
663 early_initcall(disable_hardlockup_detector);