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/module.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>
39 #include <asm/kdump.h>
41 #include <asm/processor.h>
42 #include <asm/pgtable.h>
45 #include <asm/machdep.h>
48 #include <asm/cputable.h>
49 #include <asm/sections.h>
50 #include <asm/btext.h>
51 #include <asm/nvram.h>
52 #include <asm/setup.h>
53 #include <asm/system.h>
55 #include <asm/iommu.h>
56 #include <asm/serial.h>
57 #include <asm/cache.h>
60 #include <asm/firmware.h>
63 #include <asm/kexec.h>
64 #include <asm/mmu_context.h>
69 #define DBG(fmt...) udbg_printf(fmt)
77 /* Pick defaults since we might want to patch instructions
78 * before we've read this from the device tree.
80 struct ppc64_caches ppc64_caches = {
86 EXPORT_SYMBOL_GPL(ppc64_caches);
89 * These are used in binfmt_elf.c to put aux entries on the stack
90 * for each elf executable being started.
98 static int smt_enabled_cmdline;
100 /* Look for ibm,smt-enabled OF option */
101 static void check_smt_enabled(void)
103 struct device_node *dn;
104 const char *smt_option;
106 /* Allow the command line to overrule the OF option */
107 if (smt_enabled_cmdline)
110 dn = of_find_node_by_path("/options");
113 smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
116 if (!strcmp(smt_option, "on"))
117 smt_enabled_at_boot = 1;
118 else if (!strcmp(smt_option, "off"))
119 smt_enabled_at_boot = 0;
124 /* Look for smt-enabled= cmdline option */
125 static int __init early_smt_enabled(char *p)
127 smt_enabled_cmdline = 1;
132 if (!strcmp(p, "on") || !strcmp(p, "1"))
133 smt_enabled_at_boot = 1;
134 else if (!strcmp(p, "off") || !strcmp(p, "0"))
135 smt_enabled_at_boot = 0;
139 early_param("smt-enabled", early_smt_enabled);
142 #define check_smt_enabled()
143 #endif /* CONFIG_SMP */
145 /* Put the paca pointer into r13 and SPRG_PACA */
146 static void __init setup_paca(struct paca_struct *new_paca)
148 local_paca = new_paca;
149 mtspr(SPRN_SPRG_PACA, local_paca);
150 #ifdef CONFIG_PPC_BOOK3E
151 mtspr(SPRN_SPRG_TLB_EXFRAME, local_paca->extlb);
156 * Early initialization entry point. This is called by head.S
157 * with MMU translation disabled. We rely on the "feature" of
158 * the CPU that ignores the top 2 bits of the address in real
159 * mode so we can access kernel globals normally provided we
160 * only toy with things in the RMO region. From here, we do
161 * some early parsing of the device-tree to setup out MEMBLOCK
162 * data structures, and allocate & initialize the hash table
163 * and segment tables so we can start running with translation
166 * It is this function which will call the probe() callback of
167 * the various platform types and copy the matching one to the
168 * global ppc_md structure. Your platform can eventually do
169 * some very early initializations from the probe() routine, but
170 * this is not recommended, be very careful as, for example, the
171 * device-tree is not accessible via normal means at this point.
174 void __init early_setup(unsigned long dt_ptr)
176 /* -------- printk is _NOT_ safe to use here ! ------- */
178 /* Identify CPU type */
179 identify_cpu(0, mfspr(SPRN_PVR));
181 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
182 initialise_paca(&boot_paca, 0);
183 setup_paca(&boot_paca);
185 /* Initialize lockdep early or else spinlocks will blow */
188 /* -------- printk is now safe to use ------- */
190 /* Enable early debugging if any specified (see udbg.h) */
193 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
196 * Do early initialization using the flattened device
197 * tree, such as retrieving the physical memory map or
198 * calculating/retrieving the hash table size.
200 early_init_devtree(__va(dt_ptr));
202 /* Now we know the logical id of our boot cpu, setup the paca. */
203 setup_paca(&paca[boot_cpuid]);
205 /* Fix up paca fields required for the boot cpu */
206 get_paca()->cpu_start = 1;
208 /* Probe the machine type */
211 setup_kdump_trampoline();
213 DBG("Found, Initializing memory management...\n");
215 /* Initialize the hash table or TLB handling */
218 DBG(" <- early_setup()\n");
222 void early_setup_secondary(void)
224 /* Mark interrupts enabled in PACA */
225 get_paca()->soft_enabled = 0;
227 /* Initialize the hash table or TLB handling */
228 early_init_mmu_secondary();
231 #endif /* CONFIG_SMP */
233 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
234 void smp_release_cpus(void)
238 DBG(" -> smp_release_cpus()\n");
240 /* All secondary cpus are spinning on a common spinloop, release them
241 * all now so they can start to spin on their individual paca
242 * spinloops. For non SMP kernels, the secondary cpus never get out
243 * of the common spinloop.
246 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
248 *ptr = __pa(generic_secondary_smp_init);
251 DBG(" <- smp_release_cpus()\n");
253 #endif /* CONFIG_SMP || CONFIG_KEXEC */
256 * Initialize some remaining members of the ppc64_caches and systemcfg
258 * (at least until we get rid of them completely). This is mostly some
259 * cache informations about the CPU that will be used by cache flush
260 * routines and/or provided to userland
262 static void __init initialize_cache_info(void)
264 struct device_node *np;
265 unsigned long num_cpus = 0;
267 DBG(" -> initialize_cache_info()\n");
269 for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
272 /* We're assuming *all* of the CPUs have the same
273 * d-cache and i-cache sizes... -Peter
276 if ( num_cpus == 1 ) {
277 const u32 *sizep, *lsizep;
281 lsize = cur_cpu_spec->dcache_bsize;
282 sizep = of_get_property(np, "d-cache-size", NULL);
285 lsizep = of_get_property(np, "d-cache-block-size", NULL);
286 /* fallback if block size missing */
288 lsizep = of_get_property(np, "d-cache-line-size", NULL);
291 if (sizep == 0 || lsizep == 0)
292 DBG("Argh, can't find dcache properties ! "
293 "sizep: %p, lsizep: %p\n", sizep, lsizep);
295 ppc64_caches.dsize = size;
296 ppc64_caches.dline_size = lsize;
297 ppc64_caches.log_dline_size = __ilog2(lsize);
298 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
301 lsize = cur_cpu_spec->icache_bsize;
302 sizep = of_get_property(np, "i-cache-size", NULL);
305 lsizep = of_get_property(np, "i-cache-block-size", NULL);
307 lsizep = of_get_property(np, "i-cache-line-size", NULL);
310 if (sizep == 0 || lsizep == 0)
311 DBG("Argh, can't find icache properties ! "
312 "sizep: %p, lsizep: %p\n", sizep, lsizep);
314 ppc64_caches.isize = size;
315 ppc64_caches.iline_size = lsize;
316 ppc64_caches.log_iline_size = __ilog2(lsize);
317 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
321 DBG(" <- initialize_cache_info()\n");
326 * Do some initial setup of the system. The parameters are those which
327 * were passed in from the bootloader.
329 void __init setup_system(void)
331 DBG(" -> setup_system()\n");
333 /* Apply the CPUs-specific and firmware specific fixups to kernel
334 * text (nop out sections not relevant to this CPU or this firmware)
336 do_feature_fixups(cur_cpu_spec->cpu_features,
337 &__start___ftr_fixup, &__stop___ftr_fixup);
338 do_feature_fixups(cur_cpu_spec->mmu_features,
339 &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
340 do_feature_fixups(powerpc_firmware_features,
341 &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
342 do_lwsync_fixups(cur_cpu_spec->cpu_features,
343 &__start___lwsync_fixup, &__stop___lwsync_fixup);
346 * Unflatten the device-tree passed by prom_init or kexec
348 unflatten_device_tree();
351 * Fill the ppc64_caches & systemcfg structures with informations
352 * retrieved from the device-tree.
354 initialize_cache_info();
356 #ifdef CONFIG_PPC_RTAS
358 * Initialize RTAS if available
361 #endif /* CONFIG_PPC_RTAS */
364 * Check if we have an initrd provided via the device-tree
369 * Do some platform specific early initializations, that includes
370 * setting up the hash table pointers. It also sets up some interrupt-mapping
371 * related options that will be used by finish_device_tree()
373 if (ppc_md.init_early)
377 * We can discover serial ports now since the above did setup the
378 * hash table management for us, thus ioremap works. We do that early
379 * so that further code can be debugged
381 find_legacy_serial_ports();
384 * Register early console
386 register_early_udbg_console();
394 smp_setup_cpu_maps();
397 /* Release secondary cpus out of their spinloops at 0x60 now that
398 * we can map physical -> logical CPU ids
403 printk("Starting Linux PPC64 %s\n", init_utsname()->version);
405 printk("-----------------------------------------------------\n");
406 printk("ppc64_pft_size = 0x%llx\n", ppc64_pft_size);
407 printk("physicalMemorySize = 0x%llx\n", memblock_phys_mem_size());
408 if (ppc64_caches.dline_size != 0x80)
409 printk("ppc64_caches.dcache_line_size = 0x%x\n",
410 ppc64_caches.dline_size);
411 if (ppc64_caches.iline_size != 0x80)
412 printk("ppc64_caches.icache_line_size = 0x%x\n",
413 ppc64_caches.iline_size);
414 #ifdef CONFIG_PPC_STD_MMU_64
416 printk("htab_address = 0x%p\n", htab_address);
417 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask);
418 #endif /* CONFIG_PPC_STD_MMU_64 */
419 if (PHYSICAL_START > 0)
420 printk("physical_start = 0x%llx\n",
421 (unsigned long long)PHYSICAL_START);
422 printk("-----------------------------------------------------\n");
424 DBG(" <- setup_system()\n");
427 static u64 slb0_limit(void)
429 if (cpu_has_feature(CPU_FTR_1T_SEGMENT)) {
430 return 1UL << SID_SHIFT_1T;
432 return 1UL << SID_SHIFT;
435 static void __init irqstack_early_init(void)
437 u64 limit = slb0_limit();
441 * interrupt stacks must be under 256MB, we cannot afford to take
442 * SLB misses on them.
444 for_each_possible_cpu(i) {
445 softirq_ctx[i] = (struct thread_info *)
446 __va(memblock_alloc_base(THREAD_SIZE,
447 THREAD_SIZE, limit));
448 hardirq_ctx[i] = (struct thread_info *)
449 __va(memblock_alloc_base(THREAD_SIZE,
450 THREAD_SIZE, limit));
454 #ifdef CONFIG_PPC_BOOK3E
455 static void __init exc_lvl_early_init(void)
459 for_each_possible_cpu(i) {
460 critirq_ctx[i] = (struct thread_info *)
461 __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
462 dbgirq_ctx[i] = (struct thread_info *)
463 __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
464 mcheckirq_ctx[i] = (struct thread_info *)
465 __va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
469 #define exc_lvl_early_init()
473 * Stack space used when we detect a bad kernel stack pointer, and
474 * early in SMP boots before relocation is enabled.
476 static void __init emergency_stack_init(void)
482 * Emergency stacks must be under 256MB, we cannot afford to take
483 * SLB misses on them. The ABI also requires them to be 128-byte
486 * Since we use these as temporary stacks during secondary CPU
487 * bringup, we need to get at them in real mode. This means they
488 * must also be within the RMO region.
490 limit = min(slb0_limit(), memblock.rmo_size);
492 for_each_possible_cpu(i) {
494 sp = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
496 paca[i].emergency_sp = __va(sp);
501 * Called into from start_kernel, after lock_kernel has been called.
502 * Initializes bootmem, which is unsed to manage page allocation until
503 * mem_init is called.
505 void __init setup_arch(char **cmdline_p)
507 ppc64_boot_msg(0x12, "Setup Arch");
509 *cmdline_p = cmd_line;
512 * Set cache line size based on type of cpu as a default.
513 * Systems with OF can look in the properties on the cpu node(s)
514 * for a possibly more accurate value.
516 dcache_bsize = ppc64_caches.dline_size;
517 icache_bsize = ppc64_caches.iline_size;
519 /* reboot on panic */
525 init_mm.start_code = (unsigned long)_stext;
526 init_mm.end_code = (unsigned long) _etext;
527 init_mm.end_data = (unsigned long) _edata;
528 init_mm.brk = klimit;
530 irqstack_early_init();
531 exc_lvl_early_init();
532 emergency_stack_init();
534 #ifdef CONFIG_PPC_STD_MMU_64
537 /* set up the bootmem stuff with available memory */
541 #ifdef CONFIG_DUMMY_CONSOLE
542 conswitchp = &dummy_con;
545 if (ppc_md.setup_arch)
550 /* Initialize the MMU context management stuff */
553 ppc64_boot_msg(0x15, "Setup Done");
557 /* ToDo: do something useful if ppc_md is not yet setup. */
558 #define PPC64_LINUX_FUNCTION 0x0f000000
559 #define PPC64_IPL_MESSAGE 0xc0000000
560 #define PPC64_TERM_MESSAGE 0xb0000000
562 static void ppc64_do_msg(unsigned int src, const char *msg)
564 if (ppc_md.progress) {
567 sprintf(buf, "%08X\n", src);
568 ppc_md.progress(buf, 0);
569 snprintf(buf, 128, "%s", msg);
570 ppc_md.progress(buf, 0);
574 /* Print a boot progress message. */
575 void ppc64_boot_msg(unsigned int src, const char *msg)
577 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
578 printk("[boot]%04x %s\n", src, msg);
582 #define PCPU_DYN_SIZE ()
584 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
586 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
587 __pa(MAX_DMA_ADDRESS));
590 static void __init pcpu_fc_free(void *ptr, size_t size)
592 free_bootmem(__pa(ptr), size);
595 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
597 if (cpu_to_node(from) == cpu_to_node(to))
598 return LOCAL_DISTANCE;
600 return REMOTE_DISTANCE;
603 void __init setup_per_cpu_areas(void)
605 const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
612 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
613 * to group units. For larger mappings, use 1M atom which
614 * should be large enough to contain a number of units.
616 if (mmu_linear_psize == MMU_PAGE_4K)
617 atom_size = PAGE_SIZE;
621 rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
622 pcpu_fc_alloc, pcpu_fc_free);
624 panic("cannot initialize percpu area (err=%d)", rc);
626 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
627 for_each_possible_cpu(cpu)
628 paca[cpu].data_offset = delta + pcpu_unit_offsets[cpu];
633 #ifdef CONFIG_PPC_INDIRECT_IO
634 struct ppc_pci_io ppc_pci_io;
635 EXPORT_SYMBOL(ppc_pci_io);
636 #endif /* CONFIG_PPC_INDIRECT_IO */