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/hugetlb.h>
41 #include <asm/kdump.h>
43 #include <asm/processor.h>
44 #include <asm/pgtable.h>
47 #include <asm/machdep.h>
50 #include <asm/cputable.h>
51 #include <asm/sections.h>
52 #include <asm/btext.h>
53 #include <asm/nvram.h>
54 #include <asm/setup.h>
56 #include <asm/iommu.h>
57 #include <asm/serial.h>
58 #include <asm/cache.h>
61 #include <asm/firmware.h>
64 #include <asm/kexec.h>
65 #include <asm/mmu_context.h>
66 #include <asm/code-patching.h>
67 #include <asm/kvm_ppc.h>
68 #include <asm/hugetlb.h>
73 #define DBG(fmt...) udbg_printf(fmt)
79 int __initdata spinning_secondaries;
82 /* Pick defaults since we might want to patch instructions
83 * before we've read this from the device tree.
85 struct ppc64_caches ppc64_caches = {
91 EXPORT_SYMBOL_GPL(ppc64_caches);
94 * These are used in binfmt_elf.c to put aux entries on the stack
95 * for each elf executable being started.
103 static char *smt_enabled_cmdline;
105 /* Look for ibm,smt-enabled OF option */
106 static void check_smt_enabled(void)
108 struct device_node *dn;
109 const char *smt_option;
111 /* Default to enabling all threads */
112 smt_enabled_at_boot = threads_per_core;
114 /* Allow the command line to overrule the OF option */
115 if (smt_enabled_cmdline) {
116 if (!strcmp(smt_enabled_cmdline, "on"))
117 smt_enabled_at_boot = threads_per_core;
118 else if (!strcmp(smt_enabled_cmdline, "off"))
119 smt_enabled_at_boot = 0;
124 rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
126 smt_enabled_at_boot =
127 min(threads_per_core, (int)smt);
130 dn = of_find_node_by_path("/options");
132 smt_option = of_get_property(dn, "ibm,smt-enabled",
136 if (!strcmp(smt_option, "on"))
137 smt_enabled_at_boot = threads_per_core;
138 else if (!strcmp(smt_option, "off"))
139 smt_enabled_at_boot = 0;
147 /* Look for smt-enabled= cmdline option */
148 static int __init early_smt_enabled(char *p)
150 smt_enabled_cmdline = p;
153 early_param("smt-enabled", early_smt_enabled);
156 #define check_smt_enabled()
157 #endif /* CONFIG_SMP */
159 /** Fix up paca fields required for the boot cpu */
160 static void fixup_boot_paca(void)
162 /* The boot cpu is started */
163 get_paca()->cpu_start = 1;
164 /* Allow percpu accesses to work until we setup percpu data */
165 get_paca()->data_offset = 0;
169 * Early initialization entry point. This is called by head.S
170 * with MMU translation disabled. We rely on the "feature" of
171 * the CPU that ignores the top 2 bits of the address in real
172 * mode so we can access kernel globals normally provided we
173 * only toy with things in the RMO region. From here, we do
174 * some early parsing of the device-tree to setup out MEMBLOCK
175 * data structures, and allocate & initialize the hash table
176 * and segment tables so we can start running with translation
179 * It is this function which will call the probe() callback of
180 * the various platform types and copy the matching one to the
181 * global ppc_md structure. Your platform can eventually do
182 * some very early initializations from the probe() routine, but
183 * this is not recommended, be very careful as, for example, the
184 * device-tree is not accessible via normal means at this point.
187 void __init early_setup(unsigned long dt_ptr)
189 /* -------- printk is _NOT_ safe to use here ! ------- */
191 /* Identify CPU type */
192 identify_cpu(0, mfspr(SPRN_PVR));
194 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
195 initialise_paca(&boot_paca, 0);
196 setup_paca(&boot_paca);
199 /* Initialize lockdep early or else spinlocks will blow */
202 /* -------- printk is now safe to use ------- */
204 /* Enable early debugging if any specified (see udbg.h) */
207 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
210 * Do early initialization using the flattened device
211 * tree, such as retrieving the physical memory map or
212 * calculating/retrieving the hash table size.
214 early_init_devtree(__va(dt_ptr));
216 /* Now we know the logical id of our boot cpu, setup the paca. */
217 setup_paca(&paca[boot_cpuid]);
220 /* Probe the machine type */
223 setup_kdump_trampoline();
225 DBG("Found, Initializing memory management...\n");
227 /* Initialize the hash table or TLB handling */
231 * Reserve any gigantic pages requested on the command line.
232 * memblock needs to have been initialized by the time this is
233 * called since this will reserve memory.
235 reserve_hugetlb_gpages();
237 DBG(" <- early_setup()\n");
241 void early_setup_secondary(void)
243 /* Mark interrupts enabled in PACA */
244 get_paca()->soft_enabled = 0;
246 /* Initialize the hash table or TLB handling */
247 early_init_mmu_secondary();
250 #endif /* CONFIG_SMP */
252 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
253 void smp_release_cpus(void)
258 DBG(" -> smp_release_cpus()\n");
260 /* All secondary cpus are spinning on a common spinloop, release them
261 * all now so they can start to spin on their individual paca
262 * spinloops. For non SMP kernels, the secondary cpus never get out
263 * of the common spinloop.
266 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
268 *ptr = __pa(generic_secondary_smp_init);
270 /* And wait a bit for them to catch up */
271 for (i = 0; i < 100000; i++) {
274 if (spinning_secondaries == 0)
278 DBG("spinning_secondaries = %d\n", spinning_secondaries);
280 DBG(" <- smp_release_cpus()\n");
282 #endif /* CONFIG_SMP || CONFIG_KEXEC */
285 * Initialize some remaining members of the ppc64_caches and systemcfg
287 * (at least until we get rid of them completely). This is mostly some
288 * cache informations about the CPU that will be used by cache flush
289 * routines and/or provided to userland
291 static void __init initialize_cache_info(void)
293 struct device_node *np;
294 unsigned long num_cpus = 0;
296 DBG(" -> initialize_cache_info()\n");
298 for_each_node_by_type(np, "cpu") {
302 * We're assuming *all* of the CPUs have the same
303 * d-cache and i-cache sizes... -Peter
306 const u32 *sizep, *lsizep;
310 lsize = cur_cpu_spec->dcache_bsize;
311 sizep = of_get_property(np, "d-cache-size", NULL);
314 lsizep = of_get_property(np, "d-cache-block-size",
316 /* fallback if block size missing */
318 lsizep = of_get_property(np,
323 if (sizep == 0 || lsizep == 0)
324 DBG("Argh, can't find dcache properties ! "
325 "sizep: %p, lsizep: %p\n", sizep, lsizep);
327 ppc64_caches.dsize = size;
328 ppc64_caches.dline_size = lsize;
329 ppc64_caches.log_dline_size = __ilog2(lsize);
330 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
333 lsize = cur_cpu_spec->icache_bsize;
334 sizep = of_get_property(np, "i-cache-size", NULL);
337 lsizep = of_get_property(np, "i-cache-block-size",
340 lsizep = of_get_property(np,
345 if (sizep == 0 || lsizep == 0)
346 DBG("Argh, can't find icache properties ! "
347 "sizep: %p, lsizep: %p\n", sizep, lsizep);
349 ppc64_caches.isize = size;
350 ppc64_caches.iline_size = lsize;
351 ppc64_caches.log_iline_size = __ilog2(lsize);
352 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
356 DBG(" <- initialize_cache_info()\n");
361 * Do some initial setup of the system. The parameters are those which
362 * were passed in from the bootloader.
364 void __init setup_system(void)
366 DBG(" -> setup_system()\n");
368 /* Apply the CPUs-specific and firmware specific fixups to kernel
369 * text (nop out sections not relevant to this CPU or this firmware)
371 do_feature_fixups(cur_cpu_spec->cpu_features,
372 &__start___ftr_fixup, &__stop___ftr_fixup);
373 do_feature_fixups(cur_cpu_spec->mmu_features,
374 &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
375 do_feature_fixups(powerpc_firmware_features,
376 &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
377 do_lwsync_fixups(cur_cpu_spec->cpu_features,
378 &__start___lwsync_fixup, &__stop___lwsync_fixup);
382 * Unflatten the device-tree passed by prom_init or kexec
384 unflatten_device_tree();
387 * Fill the ppc64_caches & systemcfg structures with informations
388 * retrieved from the device-tree.
390 initialize_cache_info();
392 #ifdef CONFIG_PPC_RTAS
394 * Initialize RTAS if available
397 #endif /* CONFIG_PPC_RTAS */
400 * Check if we have an initrd provided via the device-tree
405 * Do some platform specific early initializations, that includes
406 * setting up the hash table pointers. It also sets up some interrupt-mapping
407 * related options that will be used by finish_device_tree()
409 if (ppc_md.init_early)
413 * We can discover serial ports now since the above did setup the
414 * hash table management for us, thus ioremap works. We do that early
415 * so that further code can be debugged
417 find_legacy_serial_ports();
420 * Register early console
422 register_early_udbg_console();
429 smp_setup_cpu_maps();
433 /* Release secondary cpus out of their spinloops at 0x60 now that
434 * we can map physical -> logical CPU ids
439 printk("Starting Linux PPC64 %s\n", init_utsname()->version);
441 printk("-----------------------------------------------------\n");
442 printk("ppc64_pft_size = 0x%llx\n", ppc64_pft_size);
443 printk("physicalMemorySize = 0x%llx\n", memblock_phys_mem_size());
444 if (ppc64_caches.dline_size != 0x80)
445 printk("ppc64_caches.dcache_line_size = 0x%x\n",
446 ppc64_caches.dline_size);
447 if (ppc64_caches.iline_size != 0x80)
448 printk("ppc64_caches.icache_line_size = 0x%x\n",
449 ppc64_caches.iline_size);
450 #ifdef CONFIG_PPC_STD_MMU_64
452 printk("htab_address = 0x%p\n", htab_address);
453 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask);
454 #endif /* CONFIG_PPC_STD_MMU_64 */
455 if (PHYSICAL_START > 0)
456 printk("physical_start = 0x%llx\n",
457 (unsigned long long)PHYSICAL_START);
458 printk("-----------------------------------------------------\n");
460 DBG(" <- setup_system()\n");
463 /* This returns the limit below which memory accesses to the linear
464 * mapping are guarnateed not to cause a TLB or SLB miss. This is
465 * used to allocate interrupt or emergency stacks for which our
466 * exception entry path doesn't deal with being interrupted.
468 static u64 safe_stack_limit(void)
470 #ifdef CONFIG_PPC_BOOK3E
471 /* Freescale BookE bolts the entire linear mapping */
472 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
473 return linear_map_top;
474 /* Other BookE, we assume the first GB is bolted */
477 /* BookS, the first segment is bolted */
478 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
479 return 1UL << SID_SHIFT_1T;
480 return 1UL << SID_SHIFT;
484 static void __init irqstack_early_init(void)
486 u64 limit = safe_stack_limit();
490 * Interrupt stacks must be in the first segment since we
491 * cannot afford to take SLB misses on them.
493 for_each_possible_cpu(i) {
494 softirq_ctx[i] = (struct thread_info *)
495 __va(memblock_alloc_base(THREAD_SIZE,
496 THREAD_SIZE, limit));
497 hardirq_ctx[i] = (struct thread_info *)
498 __va(memblock_alloc_base(THREAD_SIZE,
499 THREAD_SIZE, limit));
503 #ifdef CONFIG_PPC_BOOK3E
504 static void __init exc_lvl_early_init(void)
506 extern unsigned int interrupt_base_book3e;
507 extern unsigned int exc_debug_debug_book3e;
511 for_each_possible_cpu(i) {
512 critirq_ctx[i] = (struct thread_info *)
513 __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
514 dbgirq_ctx[i] = (struct thread_info *)
515 __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
516 mcheckirq_ctx[i] = (struct thread_info *)
517 __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
520 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
521 patch_branch(&interrupt_base_book3e + (0x040 / 4) + 1,
522 (unsigned long)&exc_debug_debug_book3e, 0);
525 #define exc_lvl_early_init()
529 * Stack space used when we detect a bad kernel stack pointer, and
530 * early in SMP boots before relocation is enabled.
532 static void __init emergency_stack_init(void)
538 * Emergency stacks must be under 256MB, we cannot afford to take
539 * SLB misses on them. The ABI also requires them to be 128-byte
542 * Since we use these as temporary stacks during secondary CPU
543 * bringup, we need to get at them in real mode. This means they
544 * must also be within the RMO region.
546 limit = min(safe_stack_limit(), ppc64_rma_size);
548 for_each_possible_cpu(i) {
550 sp = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
552 paca[i].emergency_sp = __va(sp);
557 * Called into from start_kernel this initializes bootmem, which is used
558 * to manage page allocation until mem_init is called.
560 void __init setup_arch(char **cmdline_p)
562 ppc64_boot_msg(0x12, "Setup Arch");
564 *cmdline_p = cmd_line;
567 * Set cache line size based on type of cpu as a default.
568 * Systems with OF can look in the properties on the cpu node(s)
569 * for a possibly more accurate value.
571 dcache_bsize = ppc64_caches.dline_size;
572 icache_bsize = ppc64_caches.iline_size;
574 /* reboot on panic */
580 init_mm.start_code = (unsigned long)_stext;
581 init_mm.end_code = (unsigned long) _etext;
582 init_mm.end_data = (unsigned long) _edata;
583 init_mm.brk = klimit;
585 irqstack_early_init();
586 exc_lvl_early_init();
587 emergency_stack_init();
589 #ifdef CONFIG_PPC_STD_MMU_64
592 /* set up the bootmem stuff with available memory */
596 #ifdef CONFIG_DUMMY_CONSOLE
597 conswitchp = &dummy_con;
600 if (ppc_md.setup_arch)
605 /* Initialize the MMU context management stuff */
610 /* Interrupt code needs to be 64K-aligned */
611 if ((unsigned long)_stext & 0xffff)
612 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
613 (unsigned long)_stext);
615 ppc64_boot_msg(0x15, "Setup Done");
619 /* ToDo: do something useful if ppc_md is not yet setup. */
620 #define PPC64_LINUX_FUNCTION 0x0f000000
621 #define PPC64_IPL_MESSAGE 0xc0000000
622 #define PPC64_TERM_MESSAGE 0xb0000000
624 static void ppc64_do_msg(unsigned int src, const char *msg)
626 if (ppc_md.progress) {
629 sprintf(buf, "%08X\n", src);
630 ppc_md.progress(buf, 0);
631 snprintf(buf, 128, "%s", msg);
632 ppc_md.progress(buf, 0);
636 /* Print a boot progress message. */
637 void ppc64_boot_msg(unsigned int src, const char *msg)
639 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
640 printk("[boot]%04x %s\n", src, msg);
644 #define PCPU_DYN_SIZE ()
646 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
648 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
649 __pa(MAX_DMA_ADDRESS));
652 static void __init pcpu_fc_free(void *ptr, size_t size)
654 free_bootmem(__pa(ptr), size);
657 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
659 if (cpu_to_node(from) == cpu_to_node(to))
660 return LOCAL_DISTANCE;
662 return REMOTE_DISTANCE;
665 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
666 EXPORT_SYMBOL(__per_cpu_offset);
668 void __init setup_per_cpu_areas(void)
670 const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
677 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
678 * to group units. For larger mappings, use 1M atom which
679 * should be large enough to contain a number of units.
681 if (mmu_linear_psize == MMU_PAGE_4K)
682 atom_size = PAGE_SIZE;
686 rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
687 pcpu_fc_alloc, pcpu_fc_free);
689 panic("cannot initialize percpu area (err=%d)", rc);
691 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
692 for_each_possible_cpu(cpu) {
693 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
694 paca[cpu].data_offset = __per_cpu_offset[cpu];
700 #ifdef CONFIG_PPC_INDIRECT_IO
701 struct ppc_pci_io ppc_pci_io;
702 EXPORT_SYMBOL(ppc_pci_io);
703 #endif /* CONFIG_PPC_INDIRECT_IO */