xref: /openbmc/linux/arch/powerpc/kernel/setup_64.c (revision 384740dc)
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 
73 int have_of = 1;
74 int boot_cpuid = 0;
75 u64 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  */
80 struct ppc64_caches ppc64_caches = {
81 	.dline_size = 0x40,
82 	.log_dline_size = 6,
83 	.iline_size = 0x40,
84 	.log_iline_size = 6
85 };
86 EXPORT_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  */
92 int dcache_bsize;
93 int icache_bsize;
94 int ucache_bsize;
95 
96 #ifdef CONFIG_SMP
97 
98 static int smt_enabled_cmdline;
99 
100 /* Look for ibm,smt-enabled OF option */
101 static 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 */
125 static 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 }
139 early_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 */
146 void __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 
171 void __init early_setup(unsigned long dt_ptr)
172 {
173 	/* -------- printk is _NOT_ safe to use here ! ------- */
174 
175 	/* Fill in any unititialised pacas */
176 	initialise_pacas();
177 
178 	/* Identify CPU type */
179 	identify_cpu(0, mfspr(SPRN_PVR));
180 
181 	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
182 	setup_paca(0);
183 
184 	/* Initialize lockdep early or else spinlocks will blow */
185 	lockdep_init();
186 
187 	/* -------- printk is now safe to use ------- */
188 
189 	/* Enable early debugging if any specified (see udbg.h) */
190 	udbg_early_init();
191 
192  	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
193 
194 	/*
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.
198 	 */
199 	early_init_devtree(__va(dt_ptr));
200 
201 	/* Now we know the logical id of our boot cpu, setup the paca. */
202 	setup_paca(boot_cpuid);
203 
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;
208 
209 	/* Probe the machine type */
210 	probe_machine();
211 
212 	setup_kdump_trampoline();
213 
214 	DBG("Found, Initializing memory management...\n");
215 
216 	/*
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
220 	 */
221 	htab_initialize();
222 
223 	/*
224 	 * Initialize stab / SLB management except on iSeries
225 	 */
226 	if (cpu_has_feature(CPU_FTR_SLB))
227 		slb_initialize();
228 	else if (!firmware_has_feature(FW_FEATURE_ISERIES))
229 		stab_initialize(get_paca()->stab_real);
230 
231 	DBG(" <- early_setup()\n");
232 }
233 
234 #ifdef CONFIG_SMP
235 void early_setup_secondary(void)
236 {
237 	struct paca_struct *lpaca = get_paca();
238 
239 	/* Mark interrupts enabled in PACA */
240 	lpaca->soft_enabled = 0;
241 
242 	/* Initialize hash table for that CPU */
243 	htab_initialize_secondary();
244 
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
247 	 * iSeries anyway
248 	 */
249 	if (cpu_has_feature(CPU_FTR_SLB))
250 		slb_initialize();
251 	else
252 		stab_initialize(lpaca->stab_addr);
253 }
254 
255 #endif /* CONFIG_SMP */
256 
257 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
258 void smp_release_cpus(void)
259 {
260 	extern unsigned long __secondary_hold_spinloop;
261 	unsigned long *ptr;
262 
263 	DBG(" -> smp_release_cpus()\n");
264 
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. */
271 
272 	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
273 			- PHYSICAL_START);
274 	*ptr = 1;
275 	mb();
276 
277 	DBG(" <- smp_release_cpus()\n");
278 }
279 #endif /* CONFIG_SMP || CONFIG_KEXEC */
280 
281 /*
282  * Initialize some remaining members of the ppc64_caches and systemcfg
283  * structures
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
287  */
288 static void __init initialize_cache_info(void)
289 {
290 	struct device_node *np;
291 	unsigned long num_cpus = 0;
292 
293 	DBG(" -> initialize_cache_info()\n");
294 
295 	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
296 		num_cpus += 1;
297 
298 		/* We're assuming *all* of the CPUs have the same
299 		 * d-cache and i-cache sizes... -Peter
300 		 */
301 
302 		if ( num_cpus == 1 ) {
303 			const u32 *sizep, *lsizep;
304 			u32 size, lsize;
305 
306 			size = 0;
307 			lsize = cur_cpu_spec->dcache_bsize;
308 			sizep = of_get_property(np, "d-cache-size", NULL);
309 			if (sizep != NULL)
310 				size = *sizep;
311 			lsizep = of_get_property(np, "d-cache-block-size", NULL);
312 			/* fallback if block size missing */
313 			if (lsizep == NULL)
314 				lsizep = of_get_property(np, "d-cache-line-size", NULL);
315 			if (lsizep != NULL)
316 				lsize = *lsizep;
317 			if (sizep == 0 || lsizep == 0)
318 				DBG("Argh, can't find dcache properties ! "
319 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
320 
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;
325 
326 			size = 0;
327 			lsize = cur_cpu_spec->icache_bsize;
328 			sizep = of_get_property(np, "i-cache-size", NULL);
329 			if (sizep != NULL)
330 				size = *sizep;
331 			lsizep = of_get_property(np, "i-cache-block-size", NULL);
332 			if (lsizep == NULL)
333 				lsizep = of_get_property(np, "i-cache-line-size", NULL);
334 			if (lsizep != NULL)
335 				lsize = *lsizep;
336 			if (sizep == 0 || lsizep == 0)
337 				DBG("Argh, can't find icache properties ! "
338 				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
339 
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;
344 		}
345 	}
346 
347 	DBG(" <- initialize_cache_info()\n");
348 }
349 
350 
351 /*
352  * Do some initial setup of the system.  The parameters are those which
353  * were passed in from the bootloader.
354  */
355 void __init setup_system(void)
356 {
357 	DBG(" -> setup_system()\n");
358 
359 	/* Apply the CPUs-specific and firmware specific fixups to kernel
360 	 * text (nop out sections not relevant to this CPU or this firmware)
361 	 */
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);
368 
369 	/*
370 	 * Unflatten the device-tree passed by prom_init or kexec
371 	 */
372 	unflatten_device_tree();
373 
374 	/*
375 	 * Fill the ppc64_caches & systemcfg structures with informations
376  	 * retrieved from the device-tree.
377 	 */
378 	initialize_cache_info();
379 
380 	/*
381 	 * Initialize irq remapping subsystem
382 	 */
383 	irq_early_init();
384 
385 #ifdef CONFIG_PPC_RTAS
386 	/*
387 	 * Initialize RTAS if available
388 	 */
389 	rtas_initialize();
390 #endif /* CONFIG_PPC_RTAS */
391 
392 	/*
393 	 * Check if we have an initrd provided via the device-tree
394 	 */
395 	check_for_initrd();
396 
397 	/*
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()
401 	 */
402 	if (ppc_md.init_early)
403 		ppc_md.init_early();
404 
405  	/*
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
409 	 */
410 	find_legacy_serial_ports();
411 
412 	/*
413 	 * Register early console
414 	 */
415 	register_early_udbg_console();
416 
417 	/*
418 	 * Initialize xmon
419 	 */
420 	xmon_setup();
421 
422 	check_smt_enabled();
423 	smp_setup_cpu_maps();
424 
425 #ifdef CONFIG_SMP
426 	/* Release secondary cpus out of their spinloops at 0x60 now that
427 	 * we can map physical -> logical CPU ids
428 	 */
429 	smp_release_cpus();
430 #endif
431 
432 	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
433 
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);
443 	if (htab_address)
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);
448 #endif
449 	printk("-----------------------------------------------------\n");
450 
451 	DBG(" <- setup_system()\n");
452 }
453 
454 #ifdef CONFIG_IRQSTACKS
455 static void __init irqstack_early_init(void)
456 {
457 	unsigned int i;
458 
459 	/*
460 	 * interrupt stacks must be under 256MB, we cannot afford to take
461 	 * SLB misses on them.
462 	 */
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));
470 	}
471 }
472 #else
473 #define irqstack_early_init()
474 #endif
475 
476 /*
477  * Stack space used when we detect a bad kernel stack pointer, and
478  * early in SMP boots before relocation is enabled.
479  */
480 static void __init emergency_stack_init(void)
481 {
482 	unsigned long limit;
483 	unsigned int i;
484 
485 	/*
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
488 	 * aligned.
489 	 *
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.
493 	 */
494 	limit = min(0x10000000UL, lmb.rmo_size);
495 
496 	for_each_possible_cpu(i) {
497 		unsigned long sp;
498 		sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
499 		sp += THREAD_SIZE;
500 		paca[i].emergency_sp = __va(sp);
501 	}
502 }
503 
504 /*
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.
508  */
509 void __init setup_arch(char **cmdline_p)
510 {
511 	ppc64_boot_msg(0x12, "Setup Arch");
512 
513 	*cmdline_p = cmd_line;
514 
515 	/*
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.
519 	 */
520 	dcache_bsize = ppc64_caches.dline_size;
521 	icache_bsize = ppc64_caches.iline_size;
522 
523 	/* reboot on panic */
524 	panic_timeout = 180;
525 
526 	if (ppc_md.panic)
527 		setup_panic();
528 
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;
533 
534 	irqstack_early_init();
535 	emergency_stack_init();
536 
537 	stabs_alloc();
538 
539 	/* set up the bootmem stuff with available memory */
540 	do_init_bootmem();
541 	sparse_init();
542 
543 #ifdef CONFIG_DUMMY_CONSOLE
544 	conswitchp = &dummy_con;
545 #endif
546 
547 	if (ppc_md.setup_arch)
548 		ppc_md.setup_arch();
549 
550 	paging_init();
551 	ppc64_boot_msg(0x15, "Setup Done");
552 }
553 
554 
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
559 
560 static void ppc64_do_msg(unsigned int src, const char *msg)
561 {
562 	if (ppc_md.progress) {
563 		char buf[128];
564 
565 		sprintf(buf, "%08X\n", src);
566 		ppc_md.progress(buf, 0);
567 		snprintf(buf, 128, "%s", msg);
568 		ppc_md.progress(buf, 0);
569 	}
570 }
571 
572 /* Print a boot progress message. */
573 void ppc64_boot_msg(unsigned int src, const char *msg)
574 {
575 	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
576 	printk("[boot]%04x %s\n", src, msg);
577 }
578 
579 /* Print a termination message (print only -- does not stop the kernel) */
580 void ppc64_terminate_msg(unsigned int src, const char *msg)
581 {
582 	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
583 	printk("[terminate]%04x %s\n", src, msg);
584 }
585 
586 void cpu_die(void)
587 {
588 	if (ppc_md.cpu_die)
589 		ppc_md.cpu_die();
590 }
591 
592 #ifdef CONFIG_SMP
593 void __init setup_per_cpu_areas(void)
594 {
595 	int i;
596 	unsigned long size;
597 	char *ptr;
598 
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;
604 #endif
605 
606 	for_each_possible_cpu(i) {
607 		ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
608 		if (!ptr)
609 			panic("Cannot allocate cpu data for CPU %d\n", i);
610 
611 		paca[i].data_offset = ptr - __per_cpu_start;
612 		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
613 	}
614 }
615 #endif
616 
617 
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 */
622 
623