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/lmb.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>
68 #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 SPRG3 */
146 void __init setup_paca(int cpu)
148 local_paca = &paca[cpu];
149 mtspr(SPRN_SPRG3, local_paca);
153 * Early initialization entry point. This is called by head.S
154 * with MMU translation disabled. We rely on the "feature" of
155 * the CPU that ignores the top 2 bits of the address in real
156 * mode so we can access kernel globals normally provided we
157 * only toy with things in the RMO region. From here, we do
158 * some early parsing of the device-tree to setup out LMB
159 * data structures, and allocate & initialize the hash table
160 * and segment tables so we can start running with translation
163 * It is this function which will call the probe() callback of
164 * the various platform types and copy the matching one to the
165 * global ppc_md structure. Your platform can eventually do
166 * some very early initializations from the probe() routine, but
167 * this is not recommended, be very careful as, for example, the
168 * device-tree is not accessible via normal means at this point.
171 void __init early_setup(unsigned long dt_ptr)
173 /* -------- printk is _NOT_ safe to use here ! ------- */
175 /* Fill in any unititialised pacas */
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! */
184 /* Initialize lockdep early or else spinlocks will blow */
187 /* -------- printk is now safe to use ------- */
189 /* Enable early debugging if any specified (see udbg.h) */
192 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
195 * Do early initialization using the flattened device
196 * tree, such as retrieving the physical memory map or
197 * calculating/retrieving the hash table size.
199 early_init_devtree(__va(dt_ptr));
201 /* Now we know the logical id of our boot cpu, setup the paca. */
202 setup_paca(boot_cpuid);
204 /* Fix up paca fields required for the boot cpu */
205 get_paca()->cpu_start = 1;
206 get_paca()->stab_real = __pa((u64)&initial_stab);
207 get_paca()->stab_addr = (u64)&initial_stab;
209 /* Probe the machine type */
212 setup_kdump_trampoline();
214 DBG("Found, Initializing memory management...\n");
217 * Initialize the MMU Hash table and create the linear mapping
218 * of memory. Has to be done before stab/slb initialization as
219 * this is currently where the page size encoding is obtained
224 * Initialize stab / SLB management except on iSeries
226 if (cpu_has_feature(CPU_FTR_SLB))
228 else if (!firmware_has_feature(FW_FEATURE_ISERIES))
229 stab_initialize(get_paca()->stab_real);
231 DBG(" <- early_setup()\n");
235 void early_setup_secondary(void)
237 struct paca_struct *lpaca = get_paca();
239 /* Mark interrupts enabled in PACA */
240 lpaca->soft_enabled = 0;
242 /* Initialize hash table for that CPU */
243 htab_initialize_secondary();
245 /* Initialize STAB/SLB. We use a virtual address as it works
246 * in real mode on pSeries and we want a virutal address on
249 if (cpu_has_feature(CPU_FTR_SLB))
252 stab_initialize(lpaca->stab_addr);
255 #endif /* CONFIG_SMP */
257 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
258 void smp_release_cpus(void)
260 extern unsigned long __secondary_hold_spinloop;
263 DBG(" -> smp_release_cpus()\n");
265 /* All secondary cpus are spinning on a common spinloop, release them
266 * all now so they can start to spin on their individual paca
267 * spinloops. For non SMP kernels, the secondary cpus never get out
268 * of the common spinloop.
269 * This is useless but harmless on iSeries, secondaries are already
270 * waiting on their paca spinloops. */
272 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
277 DBG(" <- smp_release_cpus()\n");
279 #endif /* CONFIG_SMP || CONFIG_KEXEC */
282 * Initialize some remaining members of the ppc64_caches and systemcfg
284 * (at least until we get rid of them completely). This is mostly some
285 * cache informations about the CPU that will be used by cache flush
286 * routines and/or provided to userland
288 static void __init initialize_cache_info(void)
290 struct device_node *np;
291 unsigned long num_cpus = 0;
293 DBG(" -> initialize_cache_info()\n");
295 for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
298 /* We're assuming *all* of the CPUs have the same
299 * d-cache and i-cache sizes... -Peter
302 if ( num_cpus == 1 ) {
303 const u32 *sizep, *lsizep;
307 lsize = cur_cpu_spec->dcache_bsize;
308 sizep = of_get_property(np, "d-cache-size", NULL);
311 lsizep = of_get_property(np, "d-cache-block-size", NULL);
312 /* fallback if block size missing */
314 lsizep = of_get_property(np, "d-cache-line-size", NULL);
317 if (sizep == 0 || lsizep == 0)
318 DBG("Argh, can't find dcache properties ! "
319 "sizep: %p, lsizep: %p\n", sizep, lsizep);
321 ppc64_caches.dsize = size;
322 ppc64_caches.dline_size = lsize;
323 ppc64_caches.log_dline_size = __ilog2(lsize);
324 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
327 lsize = cur_cpu_spec->icache_bsize;
328 sizep = of_get_property(np, "i-cache-size", NULL);
331 lsizep = of_get_property(np, "i-cache-block-size", NULL);
333 lsizep = of_get_property(np, "i-cache-line-size", NULL);
336 if (sizep == 0 || lsizep == 0)
337 DBG("Argh, can't find icache properties ! "
338 "sizep: %p, lsizep: %p\n", sizep, lsizep);
340 ppc64_caches.isize = size;
341 ppc64_caches.iline_size = lsize;
342 ppc64_caches.log_iline_size = __ilog2(lsize);
343 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
347 DBG(" <- initialize_cache_info()\n");
352 * Do some initial setup of the system. The parameters are those which
353 * were passed in from the bootloader.
355 void __init setup_system(void)
357 DBG(" -> setup_system()\n");
359 /* Apply the CPUs-specific and firmware specific fixups to kernel
360 * text (nop out sections not relevant to this CPU or this firmware)
362 do_feature_fixups(cur_cpu_spec->cpu_features,
363 &__start___ftr_fixup, &__stop___ftr_fixup);
364 do_feature_fixups(powerpc_firmware_features,
365 &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
366 do_lwsync_fixups(cur_cpu_spec->cpu_features,
367 &__start___lwsync_fixup, &__stop___lwsync_fixup);
370 * Unflatten the device-tree passed by prom_init or kexec
372 unflatten_device_tree();
375 * Fill the ppc64_caches & systemcfg structures with informations
376 * retrieved from the device-tree.
378 initialize_cache_info();
381 * Initialize irq remapping subsystem
385 #ifdef CONFIG_PPC_RTAS
387 * Initialize RTAS if available
390 #endif /* CONFIG_PPC_RTAS */
393 * Check if we have an initrd provided via the device-tree
398 * Do some platform specific early initializations, that includes
399 * setting up the hash table pointers. It also sets up some interrupt-mapping
400 * related options that will be used by finish_device_tree()
402 if (ppc_md.init_early)
406 * We can discover serial ports now since the above did setup the
407 * hash table management for us, thus ioremap works. We do that early
408 * so that further code can be debugged
410 find_legacy_serial_ports();
413 * Register early console
415 register_early_udbg_console();
423 smp_setup_cpu_maps();
426 /* Release secondary cpus out of their spinloops at 0x60 now that
427 * we can map physical -> logical CPU ids
432 printk("Starting Linux PPC64 %s\n", init_utsname()->version);
434 printk("-----------------------------------------------------\n");
435 printk("ppc64_pft_size = 0x%lx\n", ppc64_pft_size);
436 printk("physicalMemorySize = 0x%lx\n", lmb_phys_mem_size());
437 if (ppc64_caches.dline_size != 0x80)
438 printk("ppc64_caches.dcache_line_size = 0x%x\n",
439 ppc64_caches.dline_size);
440 if (ppc64_caches.iline_size != 0x80)
441 printk("ppc64_caches.icache_line_size = 0x%x\n",
442 ppc64_caches.iline_size);
444 printk("htab_address = 0x%p\n", htab_address);
445 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask);
446 #if PHYSICAL_START > 0
447 printk("physical_start = 0x%lx\n", PHYSICAL_START);
449 printk("-----------------------------------------------------\n");
451 DBG(" <- setup_system()\n");
454 #ifdef CONFIG_IRQSTACKS
455 static void __init irqstack_early_init(void)
460 * interrupt stacks must be under 256MB, we cannot afford to take
461 * SLB misses on them.
463 for_each_possible_cpu(i) {
464 softirq_ctx[i] = (struct thread_info *)
465 __va(lmb_alloc_base(THREAD_SIZE,
466 THREAD_SIZE, 0x10000000));
467 hardirq_ctx[i] = (struct thread_info *)
468 __va(lmb_alloc_base(THREAD_SIZE,
469 THREAD_SIZE, 0x10000000));
473 #define irqstack_early_init()
477 * Stack space used when we detect a bad kernel stack pointer, and
478 * early in SMP boots before relocation is enabled.
480 static void __init emergency_stack_init(void)
486 * Emergency stacks must be under 256MB, we cannot afford to take
487 * SLB misses on them. The ABI also requires them to be 128-byte
490 * Since we use these as temporary stacks during secondary CPU
491 * bringup, we need to get at them in real mode. This means they
492 * must also be within the RMO region.
494 limit = min(0x10000000UL, lmb.rmo_size);
496 for_each_possible_cpu(i) {
498 sp = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
500 paca[i].emergency_sp = __va(sp);
505 * Called into from start_kernel, after lock_kernel has been called.
506 * Initializes bootmem, which is unsed to manage page allocation until
507 * mem_init is called.
509 void __init setup_arch(char **cmdline_p)
511 ppc64_boot_msg(0x12, "Setup Arch");
513 *cmdline_p = cmd_line;
516 * Set cache line size based on type of cpu as a default.
517 * Systems with OF can look in the properties on the cpu node(s)
518 * for a possibly more accurate value.
520 dcache_bsize = ppc64_caches.dline_size;
521 icache_bsize = ppc64_caches.iline_size;
523 /* reboot on panic */
529 init_mm.start_code = (unsigned long)_stext;
530 init_mm.end_code = (unsigned long) _etext;
531 init_mm.end_data = (unsigned long) _edata;
532 init_mm.brk = klimit;
534 irqstack_early_init();
535 emergency_stack_init();
539 /* set up the bootmem stuff with available memory */
543 #ifdef CONFIG_DUMMY_CONSOLE
544 conswitchp = &dummy_con;
547 if (ppc_md.setup_arch)
551 ppc64_boot_msg(0x15, "Setup Done");
555 /* ToDo: do something useful if ppc_md is not yet setup. */
556 #define PPC64_LINUX_FUNCTION 0x0f000000
557 #define PPC64_IPL_MESSAGE 0xc0000000
558 #define PPC64_TERM_MESSAGE 0xb0000000
560 static void ppc64_do_msg(unsigned int src, const char *msg)
562 if (ppc_md.progress) {
565 sprintf(buf, "%08X\n", src);
566 ppc_md.progress(buf, 0);
567 snprintf(buf, 128, "%s", msg);
568 ppc_md.progress(buf, 0);
572 /* Print a boot progress message. */
573 void ppc64_boot_msg(unsigned int src, const char *msg)
575 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
576 printk("[boot]%04x %s\n", src, msg);
579 /* Print a termination message (print only -- does not stop the kernel) */
580 void ppc64_terminate_msg(unsigned int src, const char *msg)
582 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
583 printk("[terminate]%04x %s\n", src, msg);
593 void __init setup_per_cpu_areas(void)
599 /* Copy section for each CPU (we discard the original) */
600 size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
601 #ifdef CONFIG_MODULES
602 if (size < PERCPU_ENOUGH_ROOM)
603 size = PERCPU_ENOUGH_ROOM;
606 for_each_possible_cpu(i) {
607 ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
609 panic("Cannot allocate cpu data for CPU %d\n", i);
611 paca[i].data_offset = ptr - __per_cpu_start;
612 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
618 #ifdef CONFIG_PPC_INDIRECT_IO
619 struct ppc_pci_io ppc_pci_io;
620 EXPORT_SYMBOL(ppc_pci_io);
621 #endif /* CONFIG_PPC_INDIRECT_IO */