xref: /openbmc/linux/arch/powerpc/kernel/setup_64.c (revision 81d67439)
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/memblock.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 #include <asm/mmu_context.h>
65 #include <asm/code-patching.h>
66 #include <asm/kvm_ppc.h>
67 
68 #include "setup.h"
69 
70 #ifdef DEBUG
71 #define DBG(fmt...) udbg_printf(fmt)
72 #else
73 #define DBG(fmt...)
74 #endif
75 
76 int boot_cpuid = 0;
77 int __initdata boot_cpu_count;
78 u64 ppc64_pft_size;
79 
80 /* Pick defaults since we might want to patch instructions
81  * before we've read this from the device tree.
82  */
83 struct ppc64_caches ppc64_caches = {
84 	.dline_size = 0x40,
85 	.log_dline_size = 6,
86 	.iline_size = 0x40,
87 	.log_iline_size = 6
88 };
89 EXPORT_SYMBOL_GPL(ppc64_caches);
90 
91 /*
92  * These are used in binfmt_elf.c to put aux entries on the stack
93  * for each elf executable being started.
94  */
95 int dcache_bsize;
96 int icache_bsize;
97 int ucache_bsize;
98 
99 #ifdef CONFIG_SMP
100 
101 static char *smt_enabled_cmdline;
102 
103 /* Look for ibm,smt-enabled OF option */
104 static void check_smt_enabled(void)
105 {
106 	struct device_node *dn;
107 	const char *smt_option;
108 
109 	/* Default to enabling all threads */
110 	smt_enabled_at_boot = threads_per_core;
111 
112 	/* Allow the command line to overrule the OF option */
113 	if (smt_enabled_cmdline) {
114 		if (!strcmp(smt_enabled_cmdline, "on"))
115 			smt_enabled_at_boot = threads_per_core;
116 		else if (!strcmp(smt_enabled_cmdline, "off"))
117 			smt_enabled_at_boot = 0;
118 		else {
119 			long smt;
120 			int rc;
121 
122 			rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
123 			if (!rc)
124 				smt_enabled_at_boot =
125 					min(threads_per_core, (int)smt);
126 		}
127 	} else {
128 		dn = of_find_node_by_path("/options");
129 		if (dn) {
130 			smt_option = of_get_property(dn, "ibm,smt-enabled",
131 						     NULL);
132 
133 			if (smt_option) {
134 				if (!strcmp(smt_option, "on"))
135 					smt_enabled_at_boot = threads_per_core;
136 				else if (!strcmp(smt_option, "off"))
137 					smt_enabled_at_boot = 0;
138 			}
139 
140 			of_node_put(dn);
141 		}
142 	}
143 }
144 
145 /* Look for smt-enabled= cmdline option */
146 static int __init early_smt_enabled(char *p)
147 {
148 	smt_enabled_cmdline = p;
149 	return 0;
150 }
151 early_param("smt-enabled", early_smt_enabled);
152 
153 #else
154 #define check_smt_enabled()
155 #endif /* CONFIG_SMP */
156 
157 /*
158  * Early initialization entry point. This is called by head.S
159  * with MMU translation disabled. We rely on the "feature" of
160  * the CPU that ignores the top 2 bits of the address in real
161  * mode so we can access kernel globals normally provided we
162  * only toy with things in the RMO region. From here, we do
163  * some early parsing of the device-tree to setup out MEMBLOCK
164  * data structures, and allocate & initialize the hash table
165  * and segment tables so we can start running with translation
166  * enabled.
167  *
168  * It is this function which will call the probe() callback of
169  * the various platform types and copy the matching one to the
170  * global ppc_md structure. Your platform can eventually do
171  * some very early initializations from the probe() routine, but
172  * this is not recommended, be very careful as, for example, the
173  * device-tree is not accessible via normal means at this point.
174  */
175 
176 void __init early_setup(unsigned long dt_ptr)
177 {
178 	/* -------- printk is _NOT_ safe to use here ! ------- */
179 
180 	/* Identify CPU type */
181 	identify_cpu(0, mfspr(SPRN_PVR));
182 
183 	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
184 	initialise_paca(&boot_paca, 0);
185 	setup_paca(&boot_paca);
186 
187 	/* Initialize lockdep early or else spinlocks will blow */
188 	lockdep_init();
189 
190 	/* -------- printk is now safe to use ------- */
191 
192 	/* Enable early debugging if any specified (see udbg.h) */
193 	udbg_early_init();
194 
195  	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
196 
197 	/*
198 	 * Do early initialization using the flattened device
199 	 * tree, such as retrieving the physical memory map or
200 	 * calculating/retrieving the hash table size.
201 	 */
202 	early_init_devtree(__va(dt_ptr));
203 
204 	/* Now we know the logical id of our boot cpu, setup the paca. */
205 	setup_paca(&paca[boot_cpuid]);
206 
207 	/* Fix up paca fields required for the boot cpu */
208 	get_paca()->cpu_start = 1;
209 
210 	/* Probe the machine type */
211 	probe_machine();
212 
213 	setup_kdump_trampoline();
214 
215 	DBG("Found, Initializing memory management...\n");
216 
217 	/* Initialize the hash table or TLB handling */
218 	early_init_mmu();
219 
220 	DBG(" <- early_setup()\n");
221 }
222 
223 #ifdef CONFIG_SMP
224 void early_setup_secondary(void)
225 {
226 	/* Mark interrupts enabled in PACA */
227 	get_paca()->soft_enabled = 0;
228 
229 	/* Initialize the hash table or TLB handling */
230 	early_init_mmu_secondary();
231 }
232 
233 #endif /* CONFIG_SMP */
234 
235 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
236 void smp_release_cpus(void)
237 {
238 	unsigned long *ptr;
239 	int i;
240 
241 	DBG(" -> smp_release_cpus()\n");
242 
243 	/* All secondary cpus are spinning on a common spinloop, release them
244 	 * all now so they can start to spin on their individual paca
245 	 * spinloops. For non SMP kernels, the secondary cpus never get out
246 	 * of the common spinloop.
247 	 */
248 
249 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
250 			- PHYSICAL_START);
251 	*ptr = __pa(generic_secondary_smp_init);
252 
253 	/* And wait a bit for them to catch up */
254 	for (i = 0; i < 100000; i++) {
255 		mb();
256 		HMT_low();
257 		if (boot_cpu_count == 0)
258 			break;
259 		udelay(1);
260 	}
261 	DBG("boot_cpu_count = %d\n", boot_cpu_count);
262 
263 	DBG(" <- smp_release_cpus()\n");
264 }
265 #endif /* CONFIG_SMP || CONFIG_KEXEC */
266 
267 /*
268  * Initialize some remaining members of the ppc64_caches and systemcfg
269  * structures
270  * (at least until we get rid of them completely). This is mostly some
271  * cache informations about the CPU that will be used by cache flush
272  * routines and/or provided to userland
273  */
274 static void __init initialize_cache_info(void)
275 {
276 	struct device_node *np;
277 	unsigned long num_cpus = 0;
278 
279 	DBG(" -> initialize_cache_info()\n");
280 
281 	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
282 		num_cpus += 1;
283 
284 		/* We're assuming *all* of the CPUs have the same
285 		 * d-cache and i-cache sizes... -Peter
286 		 */
287 
288 		if ( num_cpus == 1 ) {
289 			const u32 *sizep, *lsizep;
290 			u32 size, lsize;
291 
292 			size = 0;
293 			lsize = cur_cpu_spec->dcache_bsize;
294 			sizep = of_get_property(np, "d-cache-size", NULL);
295 			if (sizep != NULL)
296 				size = *sizep;
297 			lsizep = of_get_property(np, "d-cache-block-size", NULL);
298 			/* fallback if block size missing */
299 			if (lsizep == NULL)
300 				lsizep = of_get_property(np, "d-cache-line-size", NULL);
301 			if (lsizep != NULL)
302 				lsize = *lsizep;
303 			if (sizep == 0 || lsizep == 0)
304 				DBG("Argh, can't find dcache properties ! "
305 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
306 
307 			ppc64_caches.dsize = size;
308 			ppc64_caches.dline_size = lsize;
309 			ppc64_caches.log_dline_size = __ilog2(lsize);
310 			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
311 
312 			size = 0;
313 			lsize = cur_cpu_spec->icache_bsize;
314 			sizep = of_get_property(np, "i-cache-size", NULL);
315 			if (sizep != NULL)
316 				size = *sizep;
317 			lsizep = of_get_property(np, "i-cache-block-size", NULL);
318 			if (lsizep == NULL)
319 				lsizep = of_get_property(np, "i-cache-line-size", NULL);
320 			if (lsizep != NULL)
321 				lsize = *lsizep;
322 			if (sizep == 0 || lsizep == 0)
323 				DBG("Argh, can't find icache properties ! "
324 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
325 
326 			ppc64_caches.isize = size;
327 			ppc64_caches.iline_size = lsize;
328 			ppc64_caches.log_iline_size = __ilog2(lsize);
329 			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
330 		}
331 	}
332 
333 	DBG(" <- initialize_cache_info()\n");
334 }
335 
336 
337 /*
338  * Do some initial setup of the system.  The parameters are those which
339  * were passed in from the bootloader.
340  */
341 void __init setup_system(void)
342 {
343 	DBG(" -> setup_system()\n");
344 
345 	/* Apply the CPUs-specific and firmware specific fixups to kernel
346 	 * text (nop out sections not relevant to this CPU or this firmware)
347 	 */
348 	do_feature_fixups(cur_cpu_spec->cpu_features,
349 			  &__start___ftr_fixup, &__stop___ftr_fixup);
350 	do_feature_fixups(cur_cpu_spec->mmu_features,
351 			  &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
352 	do_feature_fixups(powerpc_firmware_features,
353 			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
354 	do_lwsync_fixups(cur_cpu_spec->cpu_features,
355 			 &__start___lwsync_fixup, &__stop___lwsync_fixup);
356 
357 	/*
358 	 * Unflatten the device-tree passed by prom_init or kexec
359 	 */
360 	unflatten_device_tree();
361 
362 	/*
363 	 * Fill the ppc64_caches & systemcfg structures with informations
364  	 * retrieved from the device-tree.
365 	 */
366 	initialize_cache_info();
367 
368 #ifdef CONFIG_PPC_RTAS
369 	/*
370 	 * Initialize RTAS if available
371 	 */
372 	rtas_initialize();
373 #endif /* CONFIG_PPC_RTAS */
374 
375 	/*
376 	 * Check if we have an initrd provided via the device-tree
377 	 */
378 	check_for_initrd();
379 
380 	/*
381 	 * Do some platform specific early initializations, that includes
382 	 * setting up the hash table pointers. It also sets up some interrupt-mapping
383 	 * related options that will be used by finish_device_tree()
384 	 */
385 	if (ppc_md.init_early)
386 		ppc_md.init_early();
387 
388  	/*
389 	 * We can discover serial ports now since the above did setup the
390 	 * hash table management for us, thus ioremap works. We do that early
391 	 * so that further code can be debugged
392 	 */
393 	find_legacy_serial_ports();
394 
395 	/*
396 	 * Register early console
397 	 */
398 	register_early_udbg_console();
399 
400 	/*
401 	 * Initialize xmon
402 	 */
403 	xmon_setup();
404 
405 	smp_setup_cpu_maps();
406 	check_smt_enabled();
407 
408 #ifdef CONFIG_SMP
409 	/* Release secondary cpus out of their spinloops at 0x60 now that
410 	 * we can map physical -> logical CPU ids
411 	 */
412 	smp_release_cpus();
413 #endif
414 
415 	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
416 
417 	printk("-----------------------------------------------------\n");
418 	printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
419 	printk("physicalMemorySize            = 0x%llx\n", memblock_phys_mem_size());
420 	if (ppc64_caches.dline_size != 0x80)
421 		printk("ppc64_caches.dcache_line_size = 0x%x\n",
422 		       ppc64_caches.dline_size);
423 	if (ppc64_caches.iline_size != 0x80)
424 		printk("ppc64_caches.icache_line_size = 0x%x\n",
425 		       ppc64_caches.iline_size);
426 #ifdef CONFIG_PPC_STD_MMU_64
427 	if (htab_address)
428 		printk("htab_address                  = 0x%p\n", htab_address);
429 	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
430 #endif /* CONFIG_PPC_STD_MMU_64 */
431 	if (PHYSICAL_START > 0)
432 		printk("physical_start                = 0x%llx\n",
433 		       (unsigned long long)PHYSICAL_START);
434 	printk("-----------------------------------------------------\n");
435 
436 	DBG(" <- setup_system()\n");
437 }
438 
439 /* This returns the limit below which memory accesses to the linear
440  * mapping are guarnateed not to cause a TLB or SLB miss. This is
441  * used to allocate interrupt or emergency stacks for which our
442  * exception entry path doesn't deal with being interrupted.
443  */
444 static u64 safe_stack_limit(void)
445 {
446 #ifdef CONFIG_PPC_BOOK3E
447 	/* Freescale BookE bolts the entire linear mapping */
448 	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
449 		return linear_map_top;
450 	/* Other BookE, we assume the first GB is bolted */
451 	return 1ul << 30;
452 #else
453 	/* BookS, the first segment is bolted */
454 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
455 		return 1UL << SID_SHIFT_1T;
456 	return 1UL << SID_SHIFT;
457 #endif
458 }
459 
460 static void __init irqstack_early_init(void)
461 {
462 	u64 limit = safe_stack_limit();
463 	unsigned int i;
464 
465 	/*
466 	 * Interrupt stacks must be in the first segment since we
467 	 * cannot afford to take SLB misses on them.
468 	 */
469 	for_each_possible_cpu(i) {
470 		softirq_ctx[i] = (struct thread_info *)
471 			__va(memblock_alloc_base(THREAD_SIZE,
472 					    THREAD_SIZE, limit));
473 		hardirq_ctx[i] = (struct thread_info *)
474 			__va(memblock_alloc_base(THREAD_SIZE,
475 					    THREAD_SIZE, limit));
476 	}
477 }
478 
479 #ifdef CONFIG_PPC_BOOK3E
480 static void __init exc_lvl_early_init(void)
481 {
482 	extern unsigned int interrupt_base_book3e;
483 	extern unsigned int exc_debug_debug_book3e;
484 
485 	unsigned int i;
486 
487 	for_each_possible_cpu(i) {
488 		critirq_ctx[i] = (struct thread_info *)
489 			__va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
490 		dbgirq_ctx[i] = (struct thread_info *)
491 			__va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
492 		mcheckirq_ctx[i] = (struct thread_info *)
493 			__va(memblock_alloc(THREAD_SIZE, THREAD_SIZE));
494 	}
495 
496 	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
497 		patch_branch(&interrupt_base_book3e + (0x040 / 4) + 1,
498 			     (unsigned long)&exc_debug_debug_book3e, 0);
499 }
500 #else
501 #define exc_lvl_early_init()
502 #endif
503 
504 /*
505  * Stack space used when we detect a bad kernel stack pointer, and
506  * early in SMP boots before relocation is enabled.
507  */
508 static void __init emergency_stack_init(void)
509 {
510 	u64 limit;
511 	unsigned int i;
512 
513 	/*
514 	 * Emergency stacks must be under 256MB, we cannot afford to take
515 	 * SLB misses on them. The ABI also requires them to be 128-byte
516 	 * aligned.
517 	 *
518 	 * Since we use these as temporary stacks during secondary CPU
519 	 * bringup, we need to get at them in real mode. This means they
520 	 * must also be within the RMO region.
521 	 */
522 	limit = min(safe_stack_limit(), ppc64_rma_size);
523 
524 	for_each_possible_cpu(i) {
525 		unsigned long sp;
526 		sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
527 		sp += THREAD_SIZE;
528 		paca[i].emergency_sp = __va(sp);
529 	}
530 }
531 
532 /*
533  * Called into from start_kernel this initializes bootmem, which is used
534  * to manage page allocation until mem_init is called.
535  */
536 void __init setup_arch(char **cmdline_p)
537 {
538 	ppc64_boot_msg(0x12, "Setup Arch");
539 
540 	*cmdline_p = cmd_line;
541 
542 	/*
543 	 * Set cache line size based on type of cpu as a default.
544 	 * Systems with OF can look in the properties on the cpu node(s)
545 	 * for a possibly more accurate value.
546 	 */
547 	dcache_bsize = ppc64_caches.dline_size;
548 	icache_bsize = ppc64_caches.iline_size;
549 
550 	/* reboot on panic */
551 	panic_timeout = 180;
552 
553 	if (ppc_md.panic)
554 		setup_panic();
555 
556 	init_mm.start_code = (unsigned long)_stext;
557 	init_mm.end_code = (unsigned long) _etext;
558 	init_mm.end_data = (unsigned long) _edata;
559 	init_mm.brk = klimit;
560 
561 	irqstack_early_init();
562 	exc_lvl_early_init();
563 	emergency_stack_init();
564 
565 #ifdef CONFIG_PPC_STD_MMU_64
566 	stabs_alloc();
567 #endif
568 	/* set up the bootmem stuff with available memory */
569 	do_init_bootmem();
570 	sparse_init();
571 
572 #ifdef CONFIG_DUMMY_CONSOLE
573 	conswitchp = &dummy_con;
574 #endif
575 
576 	if (ppc_md.setup_arch)
577 		ppc_md.setup_arch();
578 
579 	paging_init();
580 
581 	/* Initialize the MMU context management stuff */
582 	mmu_context_init();
583 
584 	kvm_rma_init();
585 
586 	ppc64_boot_msg(0x15, "Setup Done");
587 }
588 
589 
590 /* ToDo: do something useful if ppc_md is not yet setup. */
591 #define PPC64_LINUX_FUNCTION 0x0f000000
592 #define PPC64_IPL_MESSAGE 0xc0000000
593 #define PPC64_TERM_MESSAGE 0xb0000000
594 
595 static void ppc64_do_msg(unsigned int src, const char *msg)
596 {
597 	if (ppc_md.progress) {
598 		char buf[128];
599 
600 		sprintf(buf, "%08X\n", src);
601 		ppc_md.progress(buf, 0);
602 		snprintf(buf, 128, "%s", msg);
603 		ppc_md.progress(buf, 0);
604 	}
605 }
606 
607 /* Print a boot progress message. */
608 void ppc64_boot_msg(unsigned int src, const char *msg)
609 {
610 	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
611 	printk("[boot]%04x %s\n", src, msg);
612 }
613 
614 #ifdef CONFIG_SMP
615 #define PCPU_DYN_SIZE		()
616 
617 static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
618 {
619 	return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
620 				    __pa(MAX_DMA_ADDRESS));
621 }
622 
623 static void __init pcpu_fc_free(void *ptr, size_t size)
624 {
625 	free_bootmem(__pa(ptr), size);
626 }
627 
628 static int pcpu_cpu_distance(unsigned int from, unsigned int to)
629 {
630 	if (cpu_to_node(from) == cpu_to_node(to))
631 		return LOCAL_DISTANCE;
632 	else
633 		return REMOTE_DISTANCE;
634 }
635 
636 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
637 EXPORT_SYMBOL(__per_cpu_offset);
638 
639 void __init setup_per_cpu_areas(void)
640 {
641 	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
642 	size_t atom_size;
643 	unsigned long delta;
644 	unsigned int cpu;
645 	int rc;
646 
647 	/*
648 	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
649 	 * to group units.  For larger mappings, use 1M atom which
650 	 * should be large enough to contain a number of units.
651 	 */
652 	if (mmu_linear_psize == MMU_PAGE_4K)
653 		atom_size = PAGE_SIZE;
654 	else
655 		atom_size = 1 << 20;
656 
657 	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
658 				    pcpu_fc_alloc, pcpu_fc_free);
659 	if (rc < 0)
660 		panic("cannot initialize percpu area (err=%d)", rc);
661 
662 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
663 	for_each_possible_cpu(cpu) {
664                 __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
665 		paca[cpu].data_offset = __per_cpu_offset[cpu];
666 	}
667 }
668 #endif
669 
670 
671 #ifdef CONFIG_PPC_INDIRECT_IO
672 struct ppc_pci_io ppc_pci_io;
673 EXPORT_SYMBOL(ppc_pci_io);
674 #endif /* CONFIG_PPC_INDIRECT_IO */
675 
676