setup_64.c revision 945feb174b14e7098cc7ecf0cf4768d35bc52f9c 
1/*
2 *
3 * Common boot and setup code.
4 *
5 * Copyright (C) 2001 PPC64 Team, IBM Corp
6 *
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.
11 */
12
13#undef DEBUG
14
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>
38#include <asm/io.h>
39#include <asm/kdump.h>
40#include <asm/prom.h>
41#include <asm/processor.h>
42#include <asm/pgtable.h>
43#include <asm/smp.h>
44#include <asm/elf.h>
45#include <asm/machdep.h>
46#include <asm/paca.h>
47#include <asm/time.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>
54#include <asm/rtas.h>
55#include <asm/iommu.h>
56#include <asm/serial.h>
57#include <asm/cache.h>
58#include <asm/page.h>
59#include <asm/mmu.h>
60#include <asm/firmware.h>
61#include <asm/xmon.h>
62#include <asm/udbg.h>
63#include <asm/kexec.h>
64
65#include "setup.h"
66
67#ifdef DEBUG
68#define DBG(fmt...) udbg_printf(fmt)
69#else
70#define DBG(fmt...)
71#endif
72
73int have_of = 1;
74int boot_cpuid = 0;
75u64 ppc64_pft_size;
76
77/* Pick defaults since we might want to patch instructions
78 * before we've read this from the device tree.
79 */
80struct ppc64_caches ppc64_caches = {
81	.dline_size = 0x40,
82	.log_dline_size = 6,
83	.iline_size = 0x40,
84	.log_iline_size = 6
85};
86EXPORT_SYMBOL_GPL(ppc64_caches);
87
88/*
89 * These are used in binfmt_elf.c to put aux entries on the stack
90 * for each elf executable being started.
91 */
92int dcache_bsize;
93int icache_bsize;
94int ucache_bsize;
95
96#ifdef CONFIG_SMP
97
98static int smt_enabled_cmdline;
99
100/* Look for ibm,smt-enabled OF option */
101static void check_smt_enabled(void)
102{
103	struct device_node *dn;
104	const char *smt_option;
105
106	/* Allow the command line to overrule the OF option */
107	if (smt_enabled_cmdline)
108		return;
109
110	dn = of_find_node_by_path("/options");
111
112	if (dn) {
113		smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
114
115                if (smt_option) {
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;
120                }
121        }
122}
123
124/* Look for smt-enabled= cmdline option */
125static int __init early_smt_enabled(char *p)
126{
127	smt_enabled_cmdline = 1;
128
129	if (!p)
130		return 0;
131
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;
136
137	return 0;
138}
139early_param("smt-enabled", early_smt_enabled);
140
141#else
142#define check_smt_enabled()
143#endif /* CONFIG_SMP */
144
145/* Put the paca pointer into r13 and SPRG3 */
146void __init setup_paca(int cpu)
147{
148	local_paca = &paca[cpu];
149	mtspr(SPRN_SPRG3, local_paca);
150}
151
152/*
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
161 * enabled.
162 *
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.
169 */
170
171void __init early_setup(unsigned long dt_ptr)
172{
173	/* Identify CPU type */
174	identify_cpu(0, mfspr(SPRN_PVR));
175
176	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
177	setup_paca(0);
178
179	/* Enable early debugging if any specified (see udbg.h) */
180	udbg_early_init();
181
182	/* Initialize lockdep early or else spinlocks will blow */
183	lockdep_init();
184
185 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
186
187	/*
188	 * Do early initialization using the flattened device
189	 * tree, such as retrieving the physical memory map or
190	 * calculating/retrieving the hash table size.
191	 */
192	early_init_devtree(__va(dt_ptr));
193
194	/* Now we know the logical id of our boot cpu, setup the paca. */
195	setup_paca(boot_cpuid);
196
197	/* Fix up paca fields required for the boot cpu */
198	get_paca()->cpu_start = 1;
199	get_paca()->stab_real = __pa((u64)&initial_stab);
200	get_paca()->stab_addr = (u64)&initial_stab;
201
202	/* Probe the machine type */
203	probe_machine();
204
205	setup_kdump_trampoline();
206
207	DBG("Found, Initializing memory management...\n");
208
209	/*
210	 * Initialize the MMU Hash table and create the linear mapping
211	 * of memory. Has to be done before stab/slb initialization as
212	 * this is currently where the page size encoding is obtained
213	 */
214	htab_initialize();
215
216	/*
217	 * Initialize stab / SLB management except on iSeries
218	 */
219	if (cpu_has_feature(CPU_FTR_SLB))
220		slb_initialize();
221	else if (!firmware_has_feature(FW_FEATURE_ISERIES))
222		stab_initialize(get_paca()->stab_real);
223
224	DBG(" <- early_setup()\n");
225}
226
227#ifdef CONFIG_SMP
228void early_setup_secondary(void)
229{
230	struct paca_struct *lpaca = get_paca();
231
232	/* Mark interrupts enabled in PACA */
233	lpaca->soft_enabled = 0;
234
235	/* Initialize hash table for that CPU */
236	htab_initialize_secondary();
237
238	/* Initialize STAB/SLB. We use a virtual address as it works
239	 * in real mode on pSeries and we want a virutal address on
240	 * iSeries anyway
241	 */
242	if (cpu_has_feature(CPU_FTR_SLB))
243		slb_initialize();
244	else
245		stab_initialize(lpaca->stab_addr);
246}
247
248#endif /* CONFIG_SMP */
249
250#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
251void smp_release_cpus(void)
252{
253	extern unsigned long __secondary_hold_spinloop;
254	unsigned long *ptr;
255
256	DBG(" -> smp_release_cpus()\n");
257
258	/* All secondary cpus are spinning on a common spinloop, release them
259	 * all now so they can start to spin on their individual paca
260	 * spinloops. For non SMP kernels, the secondary cpus never get out
261	 * of the common spinloop.
262	 * This is useless but harmless on iSeries, secondaries are already
263	 * waiting on their paca spinloops. */
264
265	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
266			- PHYSICAL_START);
267	*ptr = 1;
268	mb();
269
270	DBG(" <- smp_release_cpus()\n");
271}
272#endif /* CONFIG_SMP || CONFIG_KEXEC */
273
274/*
275 * Initialize some remaining members of the ppc64_caches and systemcfg
276 * structures
277 * (at least until we get rid of them completely). This is mostly some
278 * cache informations about the CPU that will be used by cache flush
279 * routines and/or provided to userland
280 */
281static void __init initialize_cache_info(void)
282{
283	struct device_node *np;
284	unsigned long num_cpus = 0;
285
286	DBG(" -> initialize_cache_info()\n");
287
288	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
289		num_cpus += 1;
290
291		/* We're assuming *all* of the CPUs have the same
292		 * d-cache and i-cache sizes... -Peter
293		 */
294
295		if ( num_cpus == 1 ) {
296			const u32 *sizep, *lsizep;
297			u32 size, lsize;
298
299			size = 0;
300			lsize = cur_cpu_spec->dcache_bsize;
301			sizep = of_get_property(np, "d-cache-size", NULL);
302			if (sizep != NULL)
303				size = *sizep;
304			lsizep = of_get_property(np, "d-cache-block-size", NULL);
305			/* fallback if block size missing */
306			if (lsizep == NULL)
307				lsizep = of_get_property(np, "d-cache-line-size", NULL);
308			if (lsizep != NULL)
309				lsize = *lsizep;
310			if (sizep == 0 || lsizep == 0)
311				DBG("Argh, can't find dcache properties ! "
312				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
313
314			ppc64_caches.dsize = size;
315			ppc64_caches.dline_size = lsize;
316			ppc64_caches.log_dline_size = __ilog2(lsize);
317			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
318
319			size = 0;
320			lsize = cur_cpu_spec->icache_bsize;
321			sizep = of_get_property(np, "i-cache-size", NULL);
322			if (sizep != NULL)
323				size = *sizep;
324			lsizep = of_get_property(np, "i-cache-block-size", NULL);
325			if (lsizep == NULL)
326				lsizep = of_get_property(np, "i-cache-line-size", NULL);
327			if (lsizep != NULL)
328				lsize = *lsizep;
329			if (sizep == 0 || lsizep == 0)
330				DBG("Argh, can't find icache properties ! "
331				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
332
333			ppc64_caches.isize = size;
334			ppc64_caches.iline_size = lsize;
335			ppc64_caches.log_iline_size = __ilog2(lsize);
336			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
337		}
338	}
339
340	DBG(" <- initialize_cache_info()\n");
341}
342
343
344/*
345 * Do some initial setup of the system.  The parameters are those which
346 * were passed in from the bootloader.
347 */
348void __init setup_system(void)
349{
350	DBG(" -> setup_system()\n");
351
352	/* Apply the CPUs-specific and firmware specific fixups to kernel
353	 * text (nop out sections not relevant to this CPU or this firmware)
354	 */
355	do_feature_fixups(cur_cpu_spec->cpu_features,
356			  &__start___ftr_fixup, &__stop___ftr_fixup);
357	do_feature_fixups(powerpc_firmware_features,
358			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
359
360	/*
361	 * Unflatten the device-tree passed by prom_init or kexec
362	 */
363	unflatten_device_tree();
364
365	/*
366	 * Fill the ppc64_caches & systemcfg structures with informations
367 	 * retrieved from the device-tree.
368	 */
369	initialize_cache_info();
370
371	/*
372	 * Initialize irq remapping subsystem
373	 */
374	irq_early_init();
375
376#ifdef CONFIG_PPC_RTAS
377	/*
378	 * Initialize RTAS if available
379	 */
380	rtas_initialize();
381#endif /* CONFIG_PPC_RTAS */
382
383	/*
384	 * Check if we have an initrd provided via the device-tree
385	 */
386	check_for_initrd();
387
388	/*
389	 * Do some platform specific early initializations, that includes
390	 * setting up the hash table pointers. It also sets up some interrupt-mapping
391	 * related options that will be used by finish_device_tree()
392	 */
393	if (ppc_md.init_early)
394		ppc_md.init_early();
395
396 	/*
397	 * We can discover serial ports now since the above did setup the
398	 * hash table management for us, thus ioremap works. We do that early
399	 * so that further code can be debugged
400	 */
401	find_legacy_serial_ports();
402
403	/*
404	 * Register early console
405	 */
406	register_early_udbg_console();
407
408	/*
409	 * Initialize xmon
410	 */
411	xmon_setup();
412
413	check_smt_enabled();
414	smp_setup_cpu_maps();
415
416#ifdef CONFIG_SMP
417	/* Release secondary cpus out of their spinloops at 0x60 now that
418	 * we can map physical -> logical CPU ids
419	 */
420	smp_release_cpus();
421#endif
422
423	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
424
425	printk("-----------------------------------------------------\n");
426	printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
427	printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
428	if (ppc64_caches.dline_size != 0x80)
429		printk("ppc64_caches.dcache_line_size = 0x%x\n",
430		       ppc64_caches.dline_size);
431	if (ppc64_caches.iline_size != 0x80)
432		printk("ppc64_caches.icache_line_size = 0x%x\n",
433		       ppc64_caches.iline_size);
434	if (htab_address)
435		printk("htab_address                  = 0x%p\n", htab_address);
436	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
437#if PHYSICAL_START > 0
438	printk("physical_start                = 0x%x\n", PHYSICAL_START);
439#endif
440	printk("-----------------------------------------------------\n");
441
442	DBG(" <- setup_system()\n");
443}
444
445#ifdef CONFIG_IRQSTACKS
446static void __init irqstack_early_init(void)
447{
448	unsigned int i;
449
450	/*
451	 * interrupt stacks must be under 256MB, we cannot afford to take
452	 * SLB misses on them.
453	 */
454	for_each_possible_cpu(i) {
455		softirq_ctx[i] = (struct thread_info *)
456			__va(lmb_alloc_base(THREAD_SIZE,
457					    THREAD_SIZE, 0x10000000));
458		hardirq_ctx[i] = (struct thread_info *)
459			__va(lmb_alloc_base(THREAD_SIZE,
460					    THREAD_SIZE, 0x10000000));
461	}
462}
463#else
464#define irqstack_early_init()
465#endif
466
467/*
468 * Stack space used when we detect a bad kernel stack pointer, and
469 * early in SMP boots before relocation is enabled.
470 */
471static void __init emergency_stack_init(void)
472{
473	unsigned long limit;
474	unsigned int i;
475
476	/*
477	 * Emergency stacks must be under 256MB, we cannot afford to take
478	 * SLB misses on them. The ABI also requires them to be 128-byte
479	 * aligned.
480	 *
481	 * Since we use these as temporary stacks during secondary CPU
482	 * bringup, we need to get at them in real mode. This means they
483	 * must also be within the RMO region.
484	 */
485	limit = min(0x10000000UL, lmb.rmo_size);
486
487	for_each_possible_cpu(i)
488		paca[i].emergency_sp =
489		__va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
490}
491
492/*
493 * Called into from start_kernel, after lock_kernel has been called.
494 * Initializes bootmem, which is unsed to manage page allocation until
495 * mem_init is called.
496 */
497void __init setup_arch(char **cmdline_p)
498{
499	ppc64_boot_msg(0x12, "Setup Arch");
500
501	*cmdline_p = cmd_line;
502
503	/*
504	 * Set cache line size based on type of cpu as a default.
505	 * Systems with OF can look in the properties on the cpu node(s)
506	 * for a possibly more accurate value.
507	 */
508	dcache_bsize = ppc64_caches.dline_size;
509	icache_bsize = ppc64_caches.iline_size;
510
511	/* reboot on panic */
512	panic_timeout = 180;
513
514	if (ppc_md.panic)
515		setup_panic();
516
517	init_mm.start_code = (unsigned long)_stext;
518	init_mm.end_code = (unsigned long) _etext;
519	init_mm.end_data = (unsigned long) _edata;
520	init_mm.brk = klimit;
521
522	irqstack_early_init();
523	emergency_stack_init();
524
525	stabs_alloc();
526
527	/* set up the bootmem stuff with available memory */
528	do_init_bootmem();
529	sparse_init();
530
531#ifdef CONFIG_DUMMY_CONSOLE
532	conswitchp = &dummy_con;
533#endif
534
535	if (ppc_md.setup_arch)
536		ppc_md.setup_arch();
537
538	paging_init();
539	ppc64_boot_msg(0x15, "Setup Done");
540}
541
542
543/* ToDo: do something useful if ppc_md is not yet setup. */
544#define PPC64_LINUX_FUNCTION 0x0f000000
545#define PPC64_IPL_MESSAGE 0xc0000000
546#define PPC64_TERM_MESSAGE 0xb0000000
547
548static void ppc64_do_msg(unsigned int src, const char *msg)
549{
550	if (ppc_md.progress) {
551		char buf[128];
552
553		sprintf(buf, "%08X\n", src);
554		ppc_md.progress(buf, 0);
555		snprintf(buf, 128, "%s", msg);
556		ppc_md.progress(buf, 0);
557	}
558}
559
560/* Print a boot progress message. */
561void ppc64_boot_msg(unsigned int src, const char *msg)
562{
563	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
564	printk("[boot]%04x %s\n", src, msg);
565}
566
567/* Print a termination message (print only -- does not stop the kernel) */
568void ppc64_terminate_msg(unsigned int src, const char *msg)
569{
570	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
571	printk("[terminate]%04x %s\n", src, msg);
572}
573
574void cpu_die(void)
575{
576	if (ppc_md.cpu_die)
577		ppc_md.cpu_die();
578}
579
580#ifdef CONFIG_SMP
581void __init setup_per_cpu_areas(void)
582{
583	int i;
584	unsigned long size;
585	char *ptr;
586
587	/* Copy section for each CPU (we discard the original) */
588	size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
589#ifdef CONFIG_MODULES
590	if (size < PERCPU_ENOUGH_ROOM)
591		size = PERCPU_ENOUGH_ROOM;
592#endif
593
594	for_each_possible_cpu(i) {
595		ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
596		if (!ptr)
597			panic("Cannot allocate cpu data for CPU %d\n", i);
598
599		paca[i].data_offset = ptr - __per_cpu_start;
600		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
601	}
602
603	/* Now that per_cpu is setup, initialize cpu_sibling_map */
604	smp_setup_cpu_sibling_map();
605}
606#endif
607
608
609#ifdef CONFIG_PPC_INDIRECT_IO
610struct ppc_pci_io ppc_pci_io;
611EXPORT_SYMBOL(ppc_pci_io);
612#endif /* CONFIG_PPC_INDIRECT_IO */
613
614