setup_64.c revision 476792839467c08ddeedd8b44a7423d415b68259 
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/ide.h>
24#include <linux/seq_file.h>
25#include <linux/ioport.h>
26#include <linux/console.h>
27#include <linux/utsname.h>
28#include <linux/tty.h>
29#include <linux/root_dev.h>
30#include <linux/notifier.h>
31#include <linux/cpu.h>
32#include <linux/unistd.h>
33#include <linux/serial.h>
34#include <linux/serial_8250.h>
35#include <linux/bootmem.h>
36#include <asm/io.h>
37#include <asm/kdump.h>
38#include <asm/prom.h>
39#include <asm/processor.h>
40#include <asm/pgtable.h>
41#include <asm/smp.h>
42#include <asm/elf.h>
43#include <asm/machdep.h>
44#include <asm/paca.h>
45#include <asm/time.h>
46#include <asm/cputable.h>
47#include <asm/sections.h>
48#include <asm/btext.h>
49#include <asm/nvram.h>
50#include <asm/setup.h>
51#include <asm/system.h>
52#include <asm/rtas.h>
53#include <asm/iommu.h>
54#include <asm/serial.h>
55#include <asm/cache.h>
56#include <asm/page.h>
57#include <asm/mmu.h>
58#include <asm/lmb.h>
59#include <asm/firmware.h>
60#include <asm/xmon.h>
61#include <asm/udbg.h>
62#include <asm/kexec.h>
63
64#include "setup.h"
65
66#ifdef DEBUG
67#define DBG(fmt...) udbg_printf(fmt)
68#else
69#define DBG(fmt...)
70#endif
71
72int have_of = 1;
73int boot_cpuid = 0;
74dev_t boot_dev;
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 = 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	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
174	setup_paca(0);
175
176	/* Enable early debugging if any specified (see udbg.h) */
177	udbg_early_init();
178
179 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
180
181	/*
182	 * Do early initializations using the flattened device
183	 * tree, like retreiving the physical memory map or
184	 * calculating/retreiving the hash table size
185	 */
186	early_init_devtree(__va(dt_ptr));
187
188	/* Now we know the logical id of our boot cpu, setup the paca. */
189	setup_paca(boot_cpuid);
190
191	/* Fix up paca fields required for the boot cpu */
192	get_paca()->cpu_start = 1;
193	get_paca()->stab_real = __pa((u64)&initial_stab);
194	get_paca()->stab_addr = (u64)&initial_stab;
195
196	/* Probe the machine type */
197	probe_machine();
198
199	setup_kdump_trampoline();
200
201	DBG("Found, Initializing memory management...\n");
202
203	/*
204	 * Initialize the MMU Hash table and create the linear mapping
205	 * of memory. Has to be done before stab/slb initialization as
206	 * this is currently where the page size encoding is obtained
207	 */
208	htab_initialize();
209
210	/*
211	 * Initialize stab / SLB management except on iSeries
212	 */
213	if (cpu_has_feature(CPU_FTR_SLB))
214		slb_initialize();
215	else if (!firmware_has_feature(FW_FEATURE_ISERIES))
216		stab_initialize(get_paca()->stab_real);
217
218	DBG(" <- early_setup()\n");
219}
220
221#ifdef CONFIG_SMP
222void early_setup_secondary(void)
223{
224	struct paca_struct *lpaca = get_paca();
225
226	/* Mark enabled in PACA */
227	lpaca->proc_enabled = 0;
228
229	/* Initialize hash table for that CPU */
230	htab_initialize_secondary();
231
232	/* Initialize STAB/SLB. We use a virtual address as it works
233	 * in real mode on pSeries and we want a virutal address on
234	 * iSeries anyway
235	 */
236	if (cpu_has_feature(CPU_FTR_SLB))
237		slb_initialize();
238	else
239		stab_initialize(lpaca->stab_addr);
240}
241
242#endif /* CONFIG_SMP */
243
244#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
245void smp_release_cpus(void)
246{
247	extern unsigned long __secondary_hold_spinloop;
248	unsigned long *ptr;
249
250	DBG(" -> smp_release_cpus()\n");
251
252	/* All secondary cpus are spinning on a common spinloop, release them
253	 * all now so they can start to spin on their individual paca
254	 * spinloops. For non SMP kernels, the secondary cpus never get out
255	 * of the common spinloop.
256	 * This is useless but harmless on iSeries, secondaries are already
257	 * waiting on their paca spinloops. */
258
259	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
260			- PHYSICAL_START);
261	*ptr = 1;
262	mb();
263
264	DBG(" <- smp_release_cpus()\n");
265}
266#endif /* CONFIG_SMP || CONFIG_KEXEC */
267
268/*
269 * Initialize some remaining members of the ppc64_caches and systemcfg
270 * structures
271 * (at least until we get rid of them completely). This is mostly some
272 * cache informations about the CPU that will be used by cache flush
273 * routines and/or provided to userland
274 */
275static void __init initialize_cache_info(void)
276{
277	struct device_node *np;
278	unsigned long num_cpus = 0;
279
280	DBG(" -> initialize_cache_info()\n");
281
282	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
283		num_cpus += 1;
284
285		/* We're assuming *all* of the CPUs have the same
286		 * d-cache and i-cache sizes... -Peter
287		 */
288
289		if ( num_cpus == 1 ) {
290			const u32 *sizep, *lsizep;
291			u32 size, lsize;
292			const char *dc, *ic;
293
294			/* Then read cache informations */
295			if (machine_is(powermac)) {
296				dc = "d-cache-block-size";
297				ic = "i-cache-block-size";
298			} else {
299				dc = "d-cache-line-size";
300				ic = "i-cache-line-size";
301			}
302
303			size = 0;
304			lsize = cur_cpu_spec->dcache_bsize;
305			sizep = get_property(np, "d-cache-size", NULL);
306			if (sizep != NULL)
307				size = *sizep;
308			lsizep = get_property(np, dc, NULL);
309			if (lsizep != NULL)
310				lsize = *lsizep;
311			if (sizep == 0 || lsizep == 0)
312				DBG("Argh, can't find dcache properties ! "
313				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
314
315			ppc64_caches.dsize = size;
316			ppc64_caches.dline_size = lsize;
317			ppc64_caches.log_dline_size = __ilog2(lsize);
318			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
319
320			size = 0;
321			lsize = cur_cpu_spec->icache_bsize;
322			sizep = get_property(np, "i-cache-size", NULL);
323			if (sizep != NULL)
324				size = *sizep;
325			lsizep = get_property(np, ic, NULL);
326			if (lsizep != NULL)
327				lsize = *lsizep;
328			if (sizep == 0 || lsizep == 0)
329				DBG("Argh, can't find icache properties ! "
330				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
331
332			ppc64_caches.isize = size;
333			ppc64_caches.iline_size = lsize;
334			ppc64_caches.log_iline_size = __ilog2(lsize);
335			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
336		}
337	}
338
339	DBG(" <- initialize_cache_info()\n");
340}
341
342
343/*
344 * Do some initial setup of the system.  The parameters are those which
345 * were passed in from the bootloader.
346 */
347void __init setup_system(void)
348{
349	DBG(" -> setup_system()\n");
350
351	/*
352	 * Unflatten the device-tree passed by prom_init or kexec
353	 */
354	unflatten_device_tree();
355
356	/*
357	 * Fill the ppc64_caches & systemcfg structures with informations
358 	 * retrieved from the device-tree.
359	 */
360	initialize_cache_info();
361
362	/*
363	 * Initialize irq remapping subsystem
364	 */
365	irq_early_init();
366
367#ifdef CONFIG_PPC_RTAS
368	/*
369	 * Initialize RTAS if available
370	 */
371	rtas_initialize();
372#endif /* CONFIG_PPC_RTAS */
373
374	/*
375	 * Check if we have an initrd provided via the device-tree
376	 */
377	check_for_initrd();
378
379	/*
380	 * Do some platform specific early initializations, that includes
381	 * setting up the hash table pointers. It also sets up some interrupt-mapping
382	 * related options that will be used by finish_device_tree()
383	 */
384	ppc_md.init_early();
385
386 	/*
387	 * We can discover serial ports now since the above did setup the
388	 * hash table management for us, thus ioremap works. We do that early
389	 * so that further code can be debugged
390	 */
391	find_legacy_serial_ports();
392
393	/*
394	 * Register early console
395	 */
396	register_early_udbg_console();
397
398	/*
399	 * Initialize xmon
400	 */
401	xmon_setup();
402
403	check_smt_enabled();
404	smp_setup_cpu_maps();
405
406#ifdef CONFIG_SMP
407	/* Release secondary cpus out of their spinloops at 0x60 now that
408	 * we can map physical -> logical CPU ids
409	 */
410	smp_release_cpus();
411#endif
412
413	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
414
415	printk("-----------------------------------------------------\n");
416	printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
417	printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
418	printk("ppc64_caches.dcache_line_size = 0x%x\n",
419	       ppc64_caches.dline_size);
420	printk("ppc64_caches.icache_line_size = 0x%x\n",
421	       ppc64_caches.iline_size);
422	printk("htab_address                  = 0x%p\n", htab_address);
423	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
424#if PHYSICAL_START > 0
425	printk("physical_start                = 0x%x\n", PHYSICAL_START);
426#endif
427	printk("-----------------------------------------------------\n");
428
429	DBG(" <- setup_system()\n");
430}
431
432#ifdef CONFIG_IRQSTACKS
433static void __init irqstack_early_init(void)
434{
435	unsigned int i;
436
437	/*
438	 * interrupt stacks must be under 256MB, we cannot afford to take
439	 * SLB misses on them.
440	 */
441	for_each_possible_cpu(i) {
442		softirq_ctx[i] = (struct thread_info *)
443			__va(lmb_alloc_base(THREAD_SIZE,
444					    THREAD_SIZE, 0x10000000));
445		hardirq_ctx[i] = (struct thread_info *)
446			__va(lmb_alloc_base(THREAD_SIZE,
447					    THREAD_SIZE, 0x10000000));
448	}
449}
450#else
451#define irqstack_early_init()
452#endif
453
454/*
455 * Stack space used when we detect a bad kernel stack pointer, and
456 * early in SMP boots before relocation is enabled.
457 */
458static void __init emergency_stack_init(void)
459{
460	unsigned long limit;
461	unsigned int i;
462
463	/*
464	 * Emergency stacks must be under 256MB, we cannot afford to take
465	 * SLB misses on them. The ABI also requires them to be 128-byte
466	 * aligned.
467	 *
468	 * Since we use these as temporary stacks during secondary CPU
469	 * bringup, we need to get at them in real mode. This means they
470	 * must also be within the RMO region.
471	 */
472	limit = min(0x10000000UL, lmb.rmo_size);
473
474	for_each_possible_cpu(i)
475		paca[i].emergency_sp =
476		__va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
477}
478
479/*
480 * Called into from start_kernel, after lock_kernel has been called.
481 * Initializes bootmem, which is unsed to manage page allocation until
482 * mem_init is called.
483 */
484void __init setup_arch(char **cmdline_p)
485{
486	ppc64_boot_msg(0x12, "Setup Arch");
487
488	*cmdline_p = cmd_line;
489
490	/*
491	 * Set cache line size based on type of cpu as a default.
492	 * Systems with OF can look in the properties on the cpu node(s)
493	 * for a possibly more accurate value.
494	 */
495	dcache_bsize = ppc64_caches.dline_size;
496	icache_bsize = ppc64_caches.iline_size;
497
498	/* reboot on panic */
499	panic_timeout = 180;
500
501	if (ppc_md.panic)
502		setup_panic();
503
504	init_mm.start_code = PAGE_OFFSET;
505	init_mm.end_code = (unsigned long) _etext;
506	init_mm.end_data = (unsigned long) _edata;
507	init_mm.brk = klimit;
508
509	irqstack_early_init();
510	emergency_stack_init();
511
512	stabs_alloc();
513
514	/* set up the bootmem stuff with available memory */
515	do_init_bootmem();
516	sparse_init();
517
518#ifdef CONFIG_DUMMY_CONSOLE
519	conswitchp = &dummy_con;
520#endif
521
522	ppc_md.setup_arch();
523
524	paging_init();
525	ppc64_boot_msg(0x15, "Setup Done");
526}
527
528
529/* ToDo: do something useful if ppc_md is not yet setup. */
530#define PPC64_LINUX_FUNCTION 0x0f000000
531#define PPC64_IPL_MESSAGE 0xc0000000
532#define PPC64_TERM_MESSAGE 0xb0000000
533
534static void ppc64_do_msg(unsigned int src, const char *msg)
535{
536	if (ppc_md.progress) {
537		char buf[128];
538
539		sprintf(buf, "%08X\n", src);
540		ppc_md.progress(buf, 0);
541		snprintf(buf, 128, "%s", msg);
542		ppc_md.progress(buf, 0);
543	}
544}
545
546/* Print a boot progress message. */
547void ppc64_boot_msg(unsigned int src, const char *msg)
548{
549	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
550	printk("[boot]%04x %s\n", src, msg);
551}
552
553/* Print a termination message (print only -- does not stop the kernel) */
554void ppc64_terminate_msg(unsigned int src, const char *msg)
555{
556	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
557	printk("[terminate]%04x %s\n", src, msg);
558}
559
560void cpu_die(void)
561{
562	if (ppc_md.cpu_die)
563		ppc_md.cpu_die();
564}
565
566#ifdef CONFIG_SMP
567void __init setup_per_cpu_areas(void)
568{
569	int i;
570	unsigned long size;
571	char *ptr;
572
573	/* Copy section for each CPU (we discard the original) */
574	size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
575#ifdef CONFIG_MODULES
576	if (size < PERCPU_ENOUGH_ROOM)
577		size = PERCPU_ENOUGH_ROOM;
578#endif
579
580	for_each_possible_cpu(i) {
581		ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
582		if (!ptr)
583			panic("Cannot allocate cpu data for CPU %d\n", i);
584
585		paca[i].data_offset = ptr - __per_cpu_start;
586		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
587	}
588}
589#endif
590