xref: /openbmc/linux/arch/powerpc/kernel/prom.c (revision 2babf5c2)
1 /*
2  * Procedures for creating, accessing and interpreting the device tree.
3  *
4  * Paul Mackerras	August 1996.
5  * Copyright (C) 1996-2005 Paul Mackerras.
6  *
7  *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
8  *    {engebret|bergner}@us.ibm.com
9  *
10  *      This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  */
15 
16 #undef DEBUG
17 
18 #include <stdarg.h>
19 #include <linux/config.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/init.h>
23 #include <linux/threads.h>
24 #include <linux/spinlock.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/stringify.h>
28 #include <linux/delay.h>
29 #include <linux/initrd.h>
30 #include <linux/bitops.h>
31 #include <linux/module.h>
32 #include <linux/kexec.h>
33 
34 #include <asm/prom.h>
35 #include <asm/rtas.h>
36 #include <asm/lmb.h>
37 #include <asm/page.h>
38 #include <asm/processor.h>
39 #include <asm/irq.h>
40 #include <asm/io.h>
41 #include <asm/kdump.h>
42 #include <asm/smp.h>
43 #include <asm/system.h>
44 #include <asm/mmu.h>
45 #include <asm/pgtable.h>
46 #include <asm/pci.h>
47 #include <asm/iommu.h>
48 #include <asm/btext.h>
49 #include <asm/sections.h>
50 #include <asm/machdep.h>
51 #include <asm/pSeries_reconfig.h>
52 #include <asm/pci-bridge.h>
53 #include <asm/kexec.h>
54 
55 #ifdef DEBUG
56 #define DBG(fmt...) printk(KERN_ERR fmt)
57 #else
58 #define DBG(fmt...)
59 #endif
60 
61 
62 static int __initdata dt_root_addr_cells;
63 static int __initdata dt_root_size_cells;
64 
65 #ifdef CONFIG_PPC64
66 int __initdata iommu_is_off;
67 int __initdata iommu_force_on;
68 unsigned long tce_alloc_start, tce_alloc_end;
69 #endif
70 
71 typedef u32 cell_t;
72 
73 #if 0
74 static struct boot_param_header *initial_boot_params __initdata;
75 #else
76 struct boot_param_header *initial_boot_params;
77 #endif
78 
79 static struct device_node *allnodes = NULL;
80 
81 /* use when traversing tree through the allnext, child, sibling,
82  * or parent members of struct device_node.
83  */
84 static DEFINE_RWLOCK(devtree_lock);
85 
86 /* export that to outside world */
87 struct device_node *of_chosen;
88 
89 struct device_node *dflt_interrupt_controller;
90 int num_interrupt_controllers;
91 
92 /*
93  * Wrapper for allocating memory for various data that needs to be
94  * attached to device nodes as they are processed at boot or when
95  * added to the device tree later (e.g. DLPAR).  At boot there is
96  * already a region reserved so we just increment *mem_start by size;
97  * otherwise we call kmalloc.
98  */
99 static void * prom_alloc(unsigned long size, unsigned long *mem_start)
100 {
101 	unsigned long tmp;
102 
103 	if (!mem_start)
104 		return kmalloc(size, GFP_KERNEL);
105 
106 	tmp = *mem_start;
107 	*mem_start += size;
108 	return (void *)tmp;
109 }
110 
111 /*
112  * Find the device_node with a given phandle.
113  */
114 static struct device_node * find_phandle(phandle ph)
115 {
116 	struct device_node *np;
117 
118 	for (np = allnodes; np != 0; np = np->allnext)
119 		if (np->linux_phandle == ph)
120 			return np;
121 	return NULL;
122 }
123 
124 /*
125  * Find the interrupt parent of a node.
126  */
127 static struct device_node * __devinit intr_parent(struct device_node *p)
128 {
129 	phandle *parp;
130 
131 	parp = (phandle *) get_property(p, "interrupt-parent", NULL);
132 	if (parp == NULL)
133 		return p->parent;
134 	p = find_phandle(*parp);
135 	if (p != NULL)
136 		return p;
137 	/*
138 	 * On a powermac booted with BootX, we don't get to know the
139 	 * phandles for any nodes, so find_phandle will return NULL.
140 	 * Fortunately these machines only have one interrupt controller
141 	 * so there isn't in fact any ambiguity.  -- paulus
142 	 */
143 	if (num_interrupt_controllers == 1)
144 		p = dflt_interrupt_controller;
145 	return p;
146 }
147 
148 /*
149  * Find out the size of each entry of the interrupts property
150  * for a node.
151  */
152 int __devinit prom_n_intr_cells(struct device_node *np)
153 {
154 	struct device_node *p;
155 	unsigned int *icp;
156 
157 	for (p = np; (p = intr_parent(p)) != NULL; ) {
158 		icp = (unsigned int *)
159 			get_property(p, "#interrupt-cells", NULL);
160 		if (icp != NULL)
161 			return *icp;
162 		if (get_property(p, "interrupt-controller", NULL) != NULL
163 		    || get_property(p, "interrupt-map", NULL) != NULL) {
164 			printk("oops, node %s doesn't have #interrupt-cells\n",
165 			       p->full_name);
166 			return 1;
167 		}
168 	}
169 #ifdef DEBUG_IRQ
170 	printk("prom_n_intr_cells failed for %s\n", np->full_name);
171 #endif
172 	return 1;
173 }
174 
175 /*
176  * Map an interrupt from a device up to the platform interrupt
177  * descriptor.
178  */
179 static int __devinit map_interrupt(unsigned int **irq, struct device_node **ictrler,
180 				   struct device_node *np, unsigned int *ints,
181 				   int nintrc)
182 {
183 	struct device_node *p, *ipar;
184 	unsigned int *imap, *imask, *ip;
185 	int i, imaplen, match;
186 	int newintrc = 0, newaddrc = 0;
187 	unsigned int *reg;
188 	int naddrc;
189 
190 	reg = (unsigned int *) get_property(np, "reg", NULL);
191 	naddrc = prom_n_addr_cells(np);
192 	p = intr_parent(np);
193 	while (p != NULL) {
194 		if (get_property(p, "interrupt-controller", NULL) != NULL)
195 			/* this node is an interrupt controller, stop here */
196 			break;
197 		imap = (unsigned int *)
198 			get_property(p, "interrupt-map", &imaplen);
199 		if (imap == NULL) {
200 			p = intr_parent(p);
201 			continue;
202 		}
203 		imask = (unsigned int *)
204 			get_property(p, "interrupt-map-mask", NULL);
205 		if (imask == NULL) {
206 			printk("oops, %s has interrupt-map but no mask\n",
207 			       p->full_name);
208 			return 0;
209 		}
210 		imaplen /= sizeof(unsigned int);
211 		match = 0;
212 		ipar = NULL;
213 		while (imaplen > 0 && !match) {
214 			/* check the child-interrupt field */
215 			match = 1;
216 			for (i = 0; i < naddrc && match; ++i)
217 				match = ((reg[i] ^ imap[i]) & imask[i]) == 0;
218 			for (; i < naddrc + nintrc && match; ++i)
219 				match = ((ints[i-naddrc] ^ imap[i]) & imask[i]) == 0;
220 			imap += naddrc + nintrc;
221 			imaplen -= naddrc + nintrc;
222 			/* grab the interrupt parent */
223 			ipar = find_phandle((phandle) *imap++);
224 			--imaplen;
225 			if (ipar == NULL && num_interrupt_controllers == 1)
226 				/* cope with BootX not giving us phandles */
227 				ipar = dflt_interrupt_controller;
228 			if (ipar == NULL) {
229 				printk("oops, no int parent %x in map of %s\n",
230 				       imap[-1], p->full_name);
231 				return 0;
232 			}
233 			/* find the parent's # addr and intr cells */
234 			ip = (unsigned int *)
235 				get_property(ipar, "#interrupt-cells", NULL);
236 			if (ip == NULL) {
237 				printk("oops, no #interrupt-cells on %s\n",
238 				       ipar->full_name);
239 				return 0;
240 			}
241 			newintrc = *ip;
242 			ip = (unsigned int *)
243 				get_property(ipar, "#address-cells", NULL);
244 			newaddrc = (ip == NULL)? 0: *ip;
245 			imap += newaddrc + newintrc;
246 			imaplen -= newaddrc + newintrc;
247 		}
248 		if (imaplen < 0) {
249 			printk("oops, error decoding int-map on %s, len=%d\n",
250 			       p->full_name, imaplen);
251 			return 0;
252 		}
253 		if (!match) {
254 #ifdef DEBUG_IRQ
255 			printk("oops, no match in %s int-map for %s\n",
256 			       p->full_name, np->full_name);
257 #endif
258 			return 0;
259 		}
260 		p = ipar;
261 		naddrc = newaddrc;
262 		nintrc = newintrc;
263 		ints = imap - nintrc;
264 		reg = ints - naddrc;
265 	}
266 	if (p == NULL) {
267 #ifdef DEBUG_IRQ
268 		printk("hmmm, int tree for %s doesn't have ctrler\n",
269 		       np->full_name);
270 #endif
271 		return 0;
272 	}
273 	*irq = ints;
274 	*ictrler = p;
275 	return nintrc;
276 }
277 
278 static unsigned char map_isa_senses[4] = {
279 	IRQ_SENSE_LEVEL | IRQ_POLARITY_NEGATIVE,
280 	IRQ_SENSE_LEVEL | IRQ_POLARITY_POSITIVE,
281 	IRQ_SENSE_EDGE  | IRQ_POLARITY_NEGATIVE,
282 	IRQ_SENSE_EDGE  | IRQ_POLARITY_POSITIVE
283 };
284 
285 static unsigned char map_mpic_senses[4] = {
286 	IRQ_SENSE_EDGE  | IRQ_POLARITY_POSITIVE,
287 	IRQ_SENSE_LEVEL | IRQ_POLARITY_NEGATIVE,
288 	/* 2 seems to be used for the 8259 cascade... */
289 	IRQ_SENSE_LEVEL | IRQ_POLARITY_POSITIVE,
290 	IRQ_SENSE_EDGE  | IRQ_POLARITY_NEGATIVE,
291 };
292 
293 static int __devinit finish_node_interrupts(struct device_node *np,
294 					    unsigned long *mem_start,
295 					    int measure_only)
296 {
297 	unsigned int *ints;
298 	int intlen, intrcells, intrcount;
299 	int i, j, n, sense;
300 	unsigned int *irq, virq;
301 	struct device_node *ic;
302 	int trace = 0;
303 
304 	//#define TRACE(fmt...) do { if (trace) { printk(fmt); mdelay(1000); } } while(0)
305 #define TRACE(fmt...)
306 
307 	if (!strcmp(np->name, "smu-doorbell"))
308 		trace = 1;
309 
310 	TRACE("Finishing SMU doorbell ! num_interrupt_controllers = %d\n",
311 	      num_interrupt_controllers);
312 
313 	if (num_interrupt_controllers == 0) {
314 		/*
315 		 * Old machines just have a list of interrupt numbers
316 		 * and no interrupt-controller nodes.
317 		 */
318 		ints = (unsigned int *) get_property(np, "AAPL,interrupts",
319 						     &intlen);
320 		/* XXX old interpret_pci_props looked in parent too */
321 		/* XXX old interpret_macio_props looked for interrupts
322 		   before AAPL,interrupts */
323 		if (ints == NULL)
324 			ints = (unsigned int *) get_property(np, "interrupts",
325 							     &intlen);
326 		if (ints == NULL)
327 			return 0;
328 
329 		np->n_intrs = intlen / sizeof(unsigned int);
330 		np->intrs = prom_alloc(np->n_intrs * sizeof(np->intrs[0]),
331 				       mem_start);
332 		if (!np->intrs)
333 			return -ENOMEM;
334 		if (measure_only)
335 			return 0;
336 
337 		for (i = 0; i < np->n_intrs; ++i) {
338 			np->intrs[i].line = *ints++;
339 			np->intrs[i].sense = IRQ_SENSE_LEVEL
340 				| IRQ_POLARITY_NEGATIVE;
341 		}
342 		return 0;
343 	}
344 
345 	ints = (unsigned int *) get_property(np, "interrupts", &intlen);
346 	TRACE("ints=%p, intlen=%d\n", ints, intlen);
347 	if (ints == NULL)
348 		return 0;
349 	intrcells = prom_n_intr_cells(np);
350 	intlen /= intrcells * sizeof(unsigned int);
351 	TRACE("intrcells=%d, new intlen=%d\n", intrcells, intlen);
352 	np->intrs = prom_alloc(intlen * sizeof(*(np->intrs)), mem_start);
353 	if (!np->intrs)
354 		return -ENOMEM;
355 
356 	if (measure_only)
357 		return 0;
358 
359 	intrcount = 0;
360 	for (i = 0; i < intlen; ++i, ints += intrcells) {
361 		n = map_interrupt(&irq, &ic, np, ints, intrcells);
362 		TRACE("map, irq=%d, ic=%p, n=%d\n", irq, ic, n);
363 		if (n <= 0)
364 			continue;
365 
366 		/* don't map IRQ numbers under a cascaded 8259 controller */
367 		if (ic && device_is_compatible(ic, "chrp,iic")) {
368 			np->intrs[intrcount].line = irq[0];
369 			sense = (n > 1)? (irq[1] & 3): 3;
370 			np->intrs[intrcount].sense = map_isa_senses[sense];
371 		} else {
372 			virq = virt_irq_create_mapping(irq[0]);
373 			TRACE("virq=%d\n", virq);
374 #ifdef CONFIG_PPC64
375 			if (virq == NO_IRQ) {
376 				printk(KERN_CRIT "Could not allocate interrupt"
377 				       " number for %s\n", np->full_name);
378 				continue;
379 			}
380 #endif
381 			np->intrs[intrcount].line = irq_offset_up(virq);
382 			sense = (n > 1)? (irq[1] & 3): 1;
383 
384 			/* Apple uses bits in there in a different way, let's
385 			 * only keep the real sense bit on macs
386 			 */
387 			if (machine_is(powermac))
388 				sense &= 0x1;
389 			np->intrs[intrcount].sense = map_mpic_senses[sense];
390 		}
391 
392 #ifdef CONFIG_PPC64
393 		/* We offset irq numbers for the u3 MPIC by 128 in PowerMac */
394 		if (machine_is(powermac) && ic && ic->parent) {
395 			char *name = get_property(ic->parent, "name", NULL);
396 			if (name && !strcmp(name, "u3"))
397 				np->intrs[intrcount].line += 128;
398 			else if (!(name && (!strcmp(name, "mac-io") ||
399 					    !strcmp(name, "u4"))))
400 				/* ignore other cascaded controllers, such as
401 				   the k2-sata-root */
402 				break;
403 		}
404 #endif /* CONFIG_PPC64 */
405 		if (n > 2) {
406 			printk("hmmm, got %d intr cells for %s:", n,
407 			       np->full_name);
408 			for (j = 0; j < n; ++j)
409 				printk(" %d", irq[j]);
410 			printk("\n");
411 		}
412 		++intrcount;
413 	}
414 	np->n_intrs = intrcount;
415 
416 	return 0;
417 }
418 
419 static int __devinit finish_node(struct device_node *np,
420 				 unsigned long *mem_start,
421 				 int measure_only)
422 {
423 	struct device_node *child;
424 	int rc = 0;
425 
426 	rc = finish_node_interrupts(np, mem_start, measure_only);
427 	if (rc)
428 		goto out;
429 
430 	for (child = np->child; child != NULL; child = child->sibling) {
431 		rc = finish_node(child, mem_start, measure_only);
432 		if (rc)
433 			goto out;
434 	}
435 out:
436 	return rc;
437 }
438 
439 static void __init scan_interrupt_controllers(void)
440 {
441 	struct device_node *np;
442 	int n = 0;
443 	char *name, *ic;
444 	int iclen;
445 
446 	for (np = allnodes; np != NULL; np = np->allnext) {
447 		ic = get_property(np, "interrupt-controller", &iclen);
448 		name = get_property(np, "name", NULL);
449 		/* checking iclen makes sure we don't get a false
450 		   match on /chosen.interrupt_controller */
451 		if ((name != NULL
452 		     && strcmp(name, "interrupt-controller") == 0)
453 		    || (ic != NULL && iclen == 0
454 			&& strcmp(name, "AppleKiwi"))) {
455 			if (n == 0)
456 				dflt_interrupt_controller = np;
457 			++n;
458 		}
459 	}
460 	num_interrupt_controllers = n;
461 }
462 
463 /**
464  * finish_device_tree is called once things are running normally
465  * (i.e. with text and data mapped to the address they were linked at).
466  * It traverses the device tree and fills in some of the additional,
467  * fields in each node like {n_}addrs and {n_}intrs, the virt interrupt
468  * mapping is also initialized at this point.
469  */
470 void __init finish_device_tree(void)
471 {
472 	unsigned long start, end, size = 0;
473 
474 	DBG(" -> finish_device_tree\n");
475 
476 #ifdef CONFIG_PPC64
477 	/* Initialize virtual IRQ map */
478 	virt_irq_init();
479 #endif
480 	scan_interrupt_controllers();
481 
482 	/*
483 	 * Finish device-tree (pre-parsing some properties etc...)
484 	 * We do this in 2 passes. One with "measure_only" set, which
485 	 * will only measure the amount of memory needed, then we can
486 	 * allocate that memory, and call finish_node again. However,
487 	 * we must be careful as most routines will fail nowadays when
488 	 * prom_alloc() returns 0, so we must make sure our first pass
489 	 * doesn't start at 0. We pre-initialize size to 16 for that
490 	 * reason and then remove those additional 16 bytes
491 	 */
492 	size = 16;
493 	finish_node(allnodes, &size, 1);
494 	size -= 16;
495 
496 	if (0 == size)
497 		end = start = 0;
498 	else
499 		end = start = (unsigned long)__va(lmb_alloc(size, 128));
500 
501 	finish_node(allnodes, &end, 0);
502 	BUG_ON(end != start + size);
503 
504 	DBG(" <- finish_device_tree\n");
505 }
506 
507 static inline char *find_flat_dt_string(u32 offset)
508 {
509 	return ((char *)initial_boot_params) +
510 		initial_boot_params->off_dt_strings + offset;
511 }
512 
513 /**
514  * This function is used to scan the flattened device-tree, it is
515  * used to extract the memory informations at boot before we can
516  * unflatten the tree
517  */
518 int __init of_scan_flat_dt(int (*it)(unsigned long node,
519 				     const char *uname, int depth,
520 				     void *data),
521 			   void *data)
522 {
523 	unsigned long p = ((unsigned long)initial_boot_params) +
524 		initial_boot_params->off_dt_struct;
525 	int rc = 0;
526 	int depth = -1;
527 
528 	do {
529 		u32 tag = *((u32 *)p);
530 		char *pathp;
531 
532 		p += 4;
533 		if (tag == OF_DT_END_NODE) {
534 			depth --;
535 			continue;
536 		}
537 		if (tag == OF_DT_NOP)
538 			continue;
539 		if (tag == OF_DT_END)
540 			break;
541 		if (tag == OF_DT_PROP) {
542 			u32 sz = *((u32 *)p);
543 			p += 8;
544 			if (initial_boot_params->version < 0x10)
545 				p = _ALIGN(p, sz >= 8 ? 8 : 4);
546 			p += sz;
547 			p = _ALIGN(p, 4);
548 			continue;
549 		}
550 		if (tag != OF_DT_BEGIN_NODE) {
551 			printk(KERN_WARNING "Invalid tag %x scanning flattened"
552 			       " device tree !\n", tag);
553 			return -EINVAL;
554 		}
555 		depth++;
556 		pathp = (char *)p;
557 		p = _ALIGN(p + strlen(pathp) + 1, 4);
558 		if ((*pathp) == '/') {
559 			char *lp, *np;
560 			for (lp = NULL, np = pathp; *np; np++)
561 				if ((*np) == '/')
562 					lp = np+1;
563 			if (lp != NULL)
564 				pathp = lp;
565 		}
566 		rc = it(p, pathp, depth, data);
567 		if (rc != 0)
568 			break;
569 	} while(1);
570 
571 	return rc;
572 }
573 
574 unsigned long __init of_get_flat_dt_root(void)
575 {
576 	unsigned long p = ((unsigned long)initial_boot_params) +
577 		initial_boot_params->off_dt_struct;
578 
579 	while(*((u32 *)p) == OF_DT_NOP)
580 		p += 4;
581 	BUG_ON (*((u32 *)p) != OF_DT_BEGIN_NODE);
582 	p += 4;
583 	return _ALIGN(p + strlen((char *)p) + 1, 4);
584 }
585 
586 /**
587  * This  function can be used within scan_flattened_dt callback to get
588  * access to properties
589  */
590 void* __init of_get_flat_dt_prop(unsigned long node, const char *name,
591 				 unsigned long *size)
592 {
593 	unsigned long p = node;
594 
595 	do {
596 		u32 tag = *((u32 *)p);
597 		u32 sz, noff;
598 		const char *nstr;
599 
600 		p += 4;
601 		if (tag == OF_DT_NOP)
602 			continue;
603 		if (tag != OF_DT_PROP)
604 			return NULL;
605 
606 		sz = *((u32 *)p);
607 		noff = *((u32 *)(p + 4));
608 		p += 8;
609 		if (initial_boot_params->version < 0x10)
610 			p = _ALIGN(p, sz >= 8 ? 8 : 4);
611 
612 		nstr = find_flat_dt_string(noff);
613 		if (nstr == NULL) {
614 			printk(KERN_WARNING "Can't find property index"
615 			       " name !\n");
616 			return NULL;
617 		}
618 		if (strcmp(name, nstr) == 0) {
619 			if (size)
620 				*size = sz;
621 			return (void *)p;
622 		}
623 		p += sz;
624 		p = _ALIGN(p, 4);
625 	} while(1);
626 }
627 
628 int __init of_flat_dt_is_compatible(unsigned long node, const char *compat)
629 {
630 	const char* cp;
631 	unsigned long cplen, l;
632 
633 	cp = of_get_flat_dt_prop(node, "compatible", &cplen);
634 	if (cp == NULL)
635 		return 0;
636 	while (cplen > 0) {
637 		if (strncasecmp(cp, compat, strlen(compat)) == 0)
638 			return 1;
639 		l = strlen(cp) + 1;
640 		cp += l;
641 		cplen -= l;
642 	}
643 
644 	return 0;
645 }
646 
647 static void *__init unflatten_dt_alloc(unsigned long *mem, unsigned long size,
648 				       unsigned long align)
649 {
650 	void *res;
651 
652 	*mem = _ALIGN(*mem, align);
653 	res = (void *)*mem;
654 	*mem += size;
655 
656 	return res;
657 }
658 
659 static unsigned long __init unflatten_dt_node(unsigned long mem,
660 					      unsigned long *p,
661 					      struct device_node *dad,
662 					      struct device_node ***allnextpp,
663 					      unsigned long fpsize)
664 {
665 	struct device_node *np;
666 	struct property *pp, **prev_pp = NULL;
667 	char *pathp;
668 	u32 tag;
669 	unsigned int l, allocl;
670 	int has_name = 0;
671 	int new_format = 0;
672 
673 	tag = *((u32 *)(*p));
674 	if (tag != OF_DT_BEGIN_NODE) {
675 		printk("Weird tag at start of node: %x\n", tag);
676 		return mem;
677 	}
678 	*p += 4;
679 	pathp = (char *)*p;
680 	l = allocl = strlen(pathp) + 1;
681 	*p = _ALIGN(*p + l, 4);
682 
683 	/* version 0x10 has a more compact unit name here instead of the full
684 	 * path. we accumulate the full path size using "fpsize", we'll rebuild
685 	 * it later. We detect this because the first character of the name is
686 	 * not '/'.
687 	 */
688 	if ((*pathp) != '/') {
689 		new_format = 1;
690 		if (fpsize == 0) {
691 			/* root node: special case. fpsize accounts for path
692 			 * plus terminating zero. root node only has '/', so
693 			 * fpsize should be 2, but we want to avoid the first
694 			 * level nodes to have two '/' so we use fpsize 1 here
695 			 */
696 			fpsize = 1;
697 			allocl = 2;
698 		} else {
699 			/* account for '/' and path size minus terminal 0
700 			 * already in 'l'
701 			 */
702 			fpsize += l;
703 			allocl = fpsize;
704 		}
705 	}
706 
707 
708 	np = unflatten_dt_alloc(&mem, sizeof(struct device_node) + allocl,
709 				__alignof__(struct device_node));
710 	if (allnextpp) {
711 		memset(np, 0, sizeof(*np));
712 		np->full_name = ((char*)np) + sizeof(struct device_node);
713 		if (new_format) {
714 			char *p = np->full_name;
715 			/* rebuild full path for new format */
716 			if (dad && dad->parent) {
717 				strcpy(p, dad->full_name);
718 #ifdef DEBUG
719 				if ((strlen(p) + l + 1) != allocl) {
720 					DBG("%s: p: %d, l: %d, a: %d\n",
721 					    pathp, (int)strlen(p), l, allocl);
722 				}
723 #endif
724 				p += strlen(p);
725 			}
726 			*(p++) = '/';
727 			memcpy(p, pathp, l);
728 		} else
729 			memcpy(np->full_name, pathp, l);
730 		prev_pp = &np->properties;
731 		**allnextpp = np;
732 		*allnextpp = &np->allnext;
733 		if (dad != NULL) {
734 			np->parent = dad;
735 			/* we temporarily use the next field as `last_child'*/
736 			if (dad->next == 0)
737 				dad->child = np;
738 			else
739 				dad->next->sibling = np;
740 			dad->next = np;
741 		}
742 		kref_init(&np->kref);
743 	}
744 	while(1) {
745 		u32 sz, noff;
746 		char *pname;
747 
748 		tag = *((u32 *)(*p));
749 		if (tag == OF_DT_NOP) {
750 			*p += 4;
751 			continue;
752 		}
753 		if (tag != OF_DT_PROP)
754 			break;
755 		*p += 4;
756 		sz = *((u32 *)(*p));
757 		noff = *((u32 *)((*p) + 4));
758 		*p += 8;
759 		if (initial_boot_params->version < 0x10)
760 			*p = _ALIGN(*p, sz >= 8 ? 8 : 4);
761 
762 		pname = find_flat_dt_string(noff);
763 		if (pname == NULL) {
764 			printk("Can't find property name in list !\n");
765 			break;
766 		}
767 		if (strcmp(pname, "name") == 0)
768 			has_name = 1;
769 		l = strlen(pname) + 1;
770 		pp = unflatten_dt_alloc(&mem, sizeof(struct property),
771 					__alignof__(struct property));
772 		if (allnextpp) {
773 			if (strcmp(pname, "linux,phandle") == 0) {
774 				np->node = *((u32 *)*p);
775 				if (np->linux_phandle == 0)
776 					np->linux_phandle = np->node;
777 			}
778 			if (strcmp(pname, "ibm,phandle") == 0)
779 				np->linux_phandle = *((u32 *)*p);
780 			pp->name = pname;
781 			pp->length = sz;
782 			pp->value = (void *)*p;
783 			*prev_pp = pp;
784 			prev_pp = &pp->next;
785 		}
786 		*p = _ALIGN((*p) + sz, 4);
787 	}
788 	/* with version 0x10 we may not have the name property, recreate
789 	 * it here from the unit name if absent
790 	 */
791 	if (!has_name) {
792 		char *p = pathp, *ps = pathp, *pa = NULL;
793 		int sz;
794 
795 		while (*p) {
796 			if ((*p) == '@')
797 				pa = p;
798 			if ((*p) == '/')
799 				ps = p + 1;
800 			p++;
801 		}
802 		if (pa < ps)
803 			pa = p;
804 		sz = (pa - ps) + 1;
805 		pp = unflatten_dt_alloc(&mem, sizeof(struct property) + sz,
806 					__alignof__(struct property));
807 		if (allnextpp) {
808 			pp->name = "name";
809 			pp->length = sz;
810 			pp->value = (unsigned char *)(pp + 1);
811 			*prev_pp = pp;
812 			prev_pp = &pp->next;
813 			memcpy(pp->value, ps, sz - 1);
814 			((char *)pp->value)[sz - 1] = 0;
815 			DBG("fixed up name for %s -> %s\n", pathp, pp->value);
816 		}
817 	}
818 	if (allnextpp) {
819 		*prev_pp = NULL;
820 		np->name = get_property(np, "name", NULL);
821 		np->type = get_property(np, "device_type", NULL);
822 
823 		if (!np->name)
824 			np->name = "<NULL>";
825 		if (!np->type)
826 			np->type = "<NULL>";
827 	}
828 	while (tag == OF_DT_BEGIN_NODE) {
829 		mem = unflatten_dt_node(mem, p, np, allnextpp, fpsize);
830 		tag = *((u32 *)(*p));
831 	}
832 	if (tag != OF_DT_END_NODE) {
833 		printk("Weird tag at end of node: %x\n", tag);
834 		return mem;
835 	}
836 	*p += 4;
837 	return mem;
838 }
839 
840 static int __init early_parse_mem(char *p)
841 {
842 	if (!p)
843 		return 1;
844 
845 	memory_limit = PAGE_ALIGN(memparse(p, &p));
846 	DBG("memory limit = 0x%lx\n", memory_limit);
847 
848 	return 0;
849 }
850 early_param("mem", early_parse_mem);
851 
852 /*
853  * The device tree may be allocated below our memory limit, or inside the
854  * crash kernel region for kdump. If so, move it out now.
855  */
856 static void move_device_tree(void)
857 {
858 	unsigned long start, size;
859 	void *p;
860 
861 	DBG("-> move_device_tree\n");
862 
863 	start = __pa(initial_boot_params);
864 	size = initial_boot_params->totalsize;
865 
866 	if ((memory_limit && (start + size) > memory_limit) ||
867 			overlaps_crashkernel(start, size)) {
868 		p = __va(lmb_alloc_base(size, PAGE_SIZE, lmb.rmo_size));
869 		memcpy(p, initial_boot_params, size);
870 		initial_boot_params = (struct boot_param_header *)p;
871 		DBG("Moved device tree to 0x%p\n", p);
872 	}
873 
874 	DBG("<- move_device_tree\n");
875 }
876 
877 /**
878  * unflattens the device-tree passed by the firmware, creating the
879  * tree of struct device_node. It also fills the "name" and "type"
880  * pointers of the nodes so the normal device-tree walking functions
881  * can be used (this used to be done by finish_device_tree)
882  */
883 void __init unflatten_device_tree(void)
884 {
885 	unsigned long start, mem, size;
886 	struct device_node **allnextp = &allnodes;
887 
888 	DBG(" -> unflatten_device_tree()\n");
889 
890 	/* First pass, scan for size */
891 	start = ((unsigned long)initial_boot_params) +
892 		initial_boot_params->off_dt_struct;
893 	size = unflatten_dt_node(0, &start, NULL, NULL, 0);
894 	size = (size | 3) + 1;
895 
896 	DBG("  size is %lx, allocating...\n", size);
897 
898 	/* Allocate memory for the expanded device tree */
899 	mem = lmb_alloc(size + 4, __alignof__(struct device_node));
900 	mem = (unsigned long) __va(mem);
901 
902 	((u32 *)mem)[size / 4] = 0xdeadbeef;
903 
904 	DBG("  unflattening %lx...\n", mem);
905 
906 	/* Second pass, do actual unflattening */
907 	start = ((unsigned long)initial_boot_params) +
908 		initial_boot_params->off_dt_struct;
909 	unflatten_dt_node(mem, &start, NULL, &allnextp, 0);
910 	if (*((u32 *)start) != OF_DT_END)
911 		printk(KERN_WARNING "Weird tag at end of tree: %08x\n", *((u32 *)start));
912 	if (((u32 *)mem)[size / 4] != 0xdeadbeef)
913 		printk(KERN_WARNING "End of tree marker overwritten: %08x\n",
914 		       ((u32 *)mem)[size / 4] );
915 	*allnextp = NULL;
916 
917 	/* Get pointer to OF "/chosen" node for use everywhere */
918 	of_chosen = of_find_node_by_path("/chosen");
919 	if (of_chosen == NULL)
920 		of_chosen = of_find_node_by_path("/chosen@0");
921 
922 	DBG(" <- unflatten_device_tree()\n");
923 }
924 
925 /*
926  * ibm,pa-features is a per-cpu property that contains a string of
927  * attribute descriptors, each of which has a 2 byte header plus up
928  * to 254 bytes worth of processor attribute bits.  First header
929  * byte specifies the number of bytes following the header.
930  * Second header byte is an "attribute-specifier" type, of which
931  * zero is the only currently-defined value.
932  * Implementation:  Pass in the byte and bit offset for the feature
933  * that we are interested in.  The function will return -1 if the
934  * pa-features property is missing, or a 1/0 to indicate if the feature
935  * is supported/not supported.  Note that the bit numbers are
936  * big-endian to match the definition in PAPR.
937  */
938 static struct ibm_pa_feature {
939 	unsigned long	cpu_features;	/* CPU_FTR_xxx bit */
940 	unsigned int	cpu_user_ftrs;	/* PPC_FEATURE_xxx bit */
941 	unsigned char	pabyte;		/* byte number in ibm,pa-features */
942 	unsigned char	pabit;		/* bit number (big-endian) */
943 	unsigned char	invert;		/* if 1, pa bit set => clear feature */
944 } ibm_pa_features[] __initdata = {
945 	{0, PPC_FEATURE_HAS_MMU,	0, 0, 0},
946 	{0, PPC_FEATURE_HAS_FPU,	0, 1, 0},
947 	{CPU_FTR_SLB, 0,		0, 2, 0},
948 	{CPU_FTR_CTRL, 0,		0, 3, 0},
949 	{CPU_FTR_NOEXECUTE, 0,		0, 6, 0},
950 	{CPU_FTR_NODSISRALIGN, 0,	1, 1, 1},
951 	{CPU_FTR_CI_LARGE_PAGE, 0,	1, 2, 0},
952 };
953 
954 static void __init check_cpu_pa_features(unsigned long node)
955 {
956 	unsigned char *pa_ftrs;
957 	unsigned long len, tablelen, i, bit;
958 
959 	pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
960 	if (pa_ftrs == NULL)
961 		return;
962 
963 	/* find descriptor with type == 0 */
964 	for (;;) {
965 		if (tablelen < 3)
966 			return;
967 		len = 2 + pa_ftrs[0];
968 		if (tablelen < len)
969 			return;		/* descriptor 0 not found */
970 		if (pa_ftrs[1] == 0)
971 			break;
972 		tablelen -= len;
973 		pa_ftrs += len;
974 	}
975 
976 	/* loop over bits we know about */
977 	for (i = 0; i < ARRAY_SIZE(ibm_pa_features); ++i) {
978 		struct ibm_pa_feature *fp = &ibm_pa_features[i];
979 
980 		if (fp->pabyte >= pa_ftrs[0])
981 			continue;
982 		bit = (pa_ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
983 		if (bit ^ fp->invert) {
984 			cur_cpu_spec->cpu_features |= fp->cpu_features;
985 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
986 		} else {
987 			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
988 			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
989 		}
990 	}
991 }
992 
993 static int __init early_init_dt_scan_cpus(unsigned long node,
994 					  const char *uname, int depth,
995 					  void *data)
996 {
997 	static int logical_cpuid = 0;
998 	char *type = of_get_flat_dt_prop(node, "device_type", NULL);
999 #ifdef CONFIG_ALTIVEC
1000 	u32 *prop;
1001 #endif
1002 	u32 *intserv;
1003 	int i, nthreads;
1004 	unsigned long len;
1005 	int found = 0;
1006 
1007 	/* We are scanning "cpu" nodes only */
1008 	if (type == NULL || strcmp(type, "cpu") != 0)
1009 		return 0;
1010 
1011 	/* Get physical cpuid */
1012 	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
1013 	if (intserv) {
1014 		nthreads = len / sizeof(int);
1015 	} else {
1016 		intserv = of_get_flat_dt_prop(node, "reg", NULL);
1017 		nthreads = 1;
1018 	}
1019 
1020 	/*
1021 	 * Now see if any of these threads match our boot cpu.
1022 	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
1023 	 */
1024 	for (i = 0; i < nthreads; i++) {
1025 		/*
1026 		 * version 2 of the kexec param format adds the phys cpuid of
1027 		 * booted proc.
1028 		 */
1029 		if (initial_boot_params && initial_boot_params->version >= 2) {
1030 			if (intserv[i] ==
1031 					initial_boot_params->boot_cpuid_phys) {
1032 				found = 1;
1033 				break;
1034 			}
1035 		} else {
1036 			/*
1037 			 * Check if it's the boot-cpu, set it's hw index now,
1038 			 * unfortunately this format did not support booting
1039 			 * off secondary threads.
1040 			 */
1041 			if (of_get_flat_dt_prop(node,
1042 					"linux,boot-cpu", NULL) != NULL) {
1043 				found = 1;
1044 				break;
1045 			}
1046 		}
1047 
1048 #ifdef CONFIG_SMP
1049 		/* logical cpu id is always 0 on UP kernels */
1050 		logical_cpuid++;
1051 #endif
1052 	}
1053 
1054 	if (found) {
1055 		DBG("boot cpu: logical %d physical %d\n", logical_cpuid,
1056 			intserv[i]);
1057 		boot_cpuid = logical_cpuid;
1058 		set_hard_smp_processor_id(boot_cpuid, intserv[i]);
1059 	}
1060 
1061 #ifdef CONFIG_ALTIVEC
1062 	/* Check if we have a VMX and eventually update CPU features */
1063 	prop = (u32 *)of_get_flat_dt_prop(node, "ibm,vmx", NULL);
1064 	if (prop && (*prop) > 0) {
1065 		cur_cpu_spec->cpu_features |= CPU_FTR_ALTIVEC;
1066 		cur_cpu_spec->cpu_user_features |= PPC_FEATURE_HAS_ALTIVEC;
1067 	}
1068 
1069 	/* Same goes for Apple's "altivec" property */
1070 	prop = (u32 *)of_get_flat_dt_prop(node, "altivec", NULL);
1071 	if (prop) {
1072 		cur_cpu_spec->cpu_features |= CPU_FTR_ALTIVEC;
1073 		cur_cpu_spec->cpu_user_features |= PPC_FEATURE_HAS_ALTIVEC;
1074 	}
1075 #endif /* CONFIG_ALTIVEC */
1076 
1077 	check_cpu_pa_features(node);
1078 
1079 #ifdef CONFIG_PPC_PSERIES
1080 	if (nthreads > 1)
1081 		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
1082 	else
1083 		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
1084 #endif
1085 
1086 	return 0;
1087 }
1088 
1089 static int __init early_init_dt_scan_chosen(unsigned long node,
1090 					    const char *uname, int depth, void *data)
1091 {
1092 	unsigned long *lprop;
1093 	unsigned long l;
1094 	char *p;
1095 
1096 	DBG("search \"chosen\", depth: %d, uname: %s\n", depth, uname);
1097 
1098 	if (depth != 1 ||
1099 	    (strcmp(uname, "chosen") != 0 && strcmp(uname, "chosen@0") != 0))
1100 		return 0;
1101 
1102 #ifdef CONFIG_PPC64
1103 	/* check if iommu is forced on or off */
1104 	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
1105 		iommu_is_off = 1;
1106 	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
1107 		iommu_force_on = 1;
1108 #endif
1109 
1110 	/* mem=x on the command line is the preferred mechanism */
1111  	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
1112  	if (lprop)
1113  		memory_limit = *lprop;
1114 
1115 #ifdef CONFIG_PPC64
1116  	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
1117  	if (lprop)
1118  		tce_alloc_start = *lprop;
1119  	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
1120  	if (lprop)
1121  		tce_alloc_end = *lprop;
1122 #endif
1123 
1124 #ifdef CONFIG_PPC_RTAS
1125 	/* To help early debugging via the front panel, we retrieve a minimal
1126 	 * set of RTAS infos now if available
1127 	 */
1128 	{
1129 		u64 *basep, *entryp, *sizep;
1130 
1131 		basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
1132 		entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
1133 		sizep = of_get_flat_dt_prop(node, "linux,rtas-size", NULL);
1134 		if (basep && entryp && sizep) {
1135 			rtas.base = *basep;
1136 			rtas.entry = *entryp;
1137 			rtas.size = *sizep;
1138 		}
1139 	}
1140 #endif /* CONFIG_PPC_RTAS */
1141 
1142 #ifdef CONFIG_KEXEC
1143        lprop = (u64*)of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
1144        if (lprop)
1145                crashk_res.start = *lprop;
1146 
1147        lprop = (u64*)of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
1148        if (lprop)
1149                crashk_res.end = crashk_res.start + *lprop - 1;
1150 #endif
1151 
1152 	/* Retreive command line */
1153  	p = of_get_flat_dt_prop(node, "bootargs", &l);
1154 	if (p != NULL && l > 0)
1155 		strlcpy(cmd_line, p, min((int)l, COMMAND_LINE_SIZE));
1156 
1157 #ifdef CONFIG_CMDLINE
1158 	if (l == 0 || (l == 1 && (*p) == 0))
1159 		strlcpy(cmd_line, CONFIG_CMDLINE, COMMAND_LINE_SIZE);
1160 #endif /* CONFIG_CMDLINE */
1161 
1162 	DBG("Command line is: %s\n", cmd_line);
1163 
1164 	/* break now */
1165 	return 1;
1166 }
1167 
1168 static int __init early_init_dt_scan_root(unsigned long node,
1169 					  const char *uname, int depth, void *data)
1170 {
1171 	u32 *prop;
1172 
1173 	if (depth != 0)
1174 		return 0;
1175 
1176 	prop = of_get_flat_dt_prop(node, "#size-cells", NULL);
1177 	dt_root_size_cells = (prop == NULL) ? 1 : *prop;
1178 	DBG("dt_root_size_cells = %x\n", dt_root_size_cells);
1179 
1180 	prop = of_get_flat_dt_prop(node, "#address-cells", NULL);
1181 	dt_root_addr_cells = (prop == NULL) ? 2 : *prop;
1182 	DBG("dt_root_addr_cells = %x\n", dt_root_addr_cells);
1183 
1184 	/* break now */
1185 	return 1;
1186 }
1187 
1188 static unsigned long __init dt_mem_next_cell(int s, cell_t **cellp)
1189 {
1190 	cell_t *p = *cellp;
1191 	unsigned long r;
1192 
1193 	/* Ignore more than 2 cells */
1194 	while (s > sizeof(unsigned long) / 4) {
1195 		p++;
1196 		s--;
1197 	}
1198 	r = *p++;
1199 #ifdef CONFIG_PPC64
1200 	if (s > 1) {
1201 		r <<= 32;
1202 		r |= *(p++);
1203 		s--;
1204 	}
1205 #endif
1206 
1207 	*cellp = p;
1208 	return r;
1209 }
1210 
1211 
1212 static int __init early_init_dt_scan_memory(unsigned long node,
1213 					    const char *uname, int depth, void *data)
1214 {
1215 	char *type = of_get_flat_dt_prop(node, "device_type", NULL);
1216 	cell_t *reg, *endp;
1217 	unsigned long l;
1218 
1219 	/* We are scanning "memory" nodes only */
1220 	if (type == NULL) {
1221 		/*
1222 		 * The longtrail doesn't have a device_type on the
1223 		 * /memory node, so look for the node called /memory@0.
1224 		 */
1225 		if (depth != 1 || strcmp(uname, "memory@0") != 0)
1226 			return 0;
1227 	} else if (strcmp(type, "memory") != 0)
1228 		return 0;
1229 
1230 	reg = (cell_t *)of_get_flat_dt_prop(node, "linux,usable-memory", &l);
1231 	if (reg == NULL)
1232 		reg = (cell_t *)of_get_flat_dt_prop(node, "reg", &l);
1233 	if (reg == NULL)
1234 		return 0;
1235 
1236 	endp = reg + (l / sizeof(cell_t));
1237 
1238 	DBG("memory scan node %s, reg size %ld, data: %x %x %x %x,\n",
1239 	    uname, l, reg[0], reg[1], reg[2], reg[3]);
1240 
1241 	while ((endp - reg) >= (dt_root_addr_cells + dt_root_size_cells)) {
1242 		unsigned long base, size;
1243 
1244 		base = dt_mem_next_cell(dt_root_addr_cells, &reg);
1245 		size = dt_mem_next_cell(dt_root_size_cells, &reg);
1246 
1247 		if (size == 0)
1248 			continue;
1249 		DBG(" - %lx ,  %lx\n", base, size);
1250 #ifdef CONFIG_PPC64
1251 		if (iommu_is_off) {
1252 			if (base >= 0x80000000ul)
1253 				continue;
1254 			if ((base + size) > 0x80000000ul)
1255 				size = 0x80000000ul - base;
1256 		}
1257 #endif
1258 		lmb_add(base, size);
1259 	}
1260 	return 0;
1261 }
1262 
1263 static void __init early_reserve_mem(void)
1264 {
1265 	u64 base, size;
1266 	u64 *reserve_map;
1267 
1268 	reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
1269 					initial_boot_params->off_mem_rsvmap);
1270 #ifdef CONFIG_PPC32
1271 	/*
1272 	 * Handle the case where we might be booting from an old kexec
1273 	 * image that setup the mem_rsvmap as pairs of 32-bit values
1274 	 */
1275 	if (*reserve_map > 0xffffffffull) {
1276 		u32 base_32, size_32;
1277 		u32 *reserve_map_32 = (u32 *)reserve_map;
1278 
1279 		while (1) {
1280 			base_32 = *(reserve_map_32++);
1281 			size_32 = *(reserve_map_32++);
1282 			if (size_32 == 0)
1283 				break;
1284 			DBG("reserving: %x -> %x\n", base_32, size_32);
1285 			lmb_reserve(base_32, size_32);
1286 		}
1287 		return;
1288 	}
1289 #endif
1290 	while (1) {
1291 		base = *(reserve_map++);
1292 		size = *(reserve_map++);
1293 		if (size == 0)
1294 			break;
1295 		DBG("reserving: %llx -> %llx\n", base, size);
1296 		lmb_reserve(base, size);
1297 	}
1298 
1299 #if 0
1300 	DBG("memory reserved, lmbs :\n");
1301       	lmb_dump_all();
1302 #endif
1303 }
1304 
1305 void __init early_init_devtree(void *params)
1306 {
1307 	DBG(" -> early_init_devtree()\n");
1308 
1309 	/* Setup flat device-tree pointer */
1310 	initial_boot_params = params;
1311 
1312 	/* Retrieve various informations from the /chosen node of the
1313 	 * device-tree, including the platform type, initrd location and
1314 	 * size, TCE reserve, and more ...
1315 	 */
1316 	of_scan_flat_dt(early_init_dt_scan_chosen, NULL);
1317 
1318 	/* Scan memory nodes and rebuild LMBs */
1319 	lmb_init();
1320 	of_scan_flat_dt(early_init_dt_scan_root, NULL);
1321 	of_scan_flat_dt(early_init_dt_scan_memory, NULL);
1322 
1323 	/* Save command line for /proc/cmdline and then parse parameters */
1324 	strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
1325 	parse_early_param();
1326 
1327 	/* Reserve LMB regions used by kernel, initrd, dt, etc... */
1328 	lmb_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
1329 #ifdef CONFIG_CRASH_DUMP
1330 	lmb_reserve(0, KDUMP_RESERVE_LIMIT);
1331 #endif
1332 	early_reserve_mem();
1333 
1334 	lmb_enforce_memory_limit(memory_limit);
1335 	lmb_analyze();
1336 
1337 	DBG("Phys. mem: %lx\n", lmb_phys_mem_size());
1338 
1339 	/* We may need to relocate the flat tree, do it now.
1340 	 * FIXME .. and the initrd too? */
1341 	move_device_tree();
1342 
1343 	DBG("Scanning CPUs ...\n");
1344 
1345 	/* Retreive CPU related informations from the flat tree
1346 	 * (altivec support, boot CPU ID, ...)
1347 	 */
1348 	of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
1349 
1350 	DBG(" <- early_init_devtree()\n");
1351 }
1352 
1353 #undef printk
1354 
1355 int
1356 prom_n_addr_cells(struct device_node* np)
1357 {
1358 	int* ip;
1359 	do {
1360 		if (np->parent)
1361 			np = np->parent;
1362 		ip = (int *) get_property(np, "#address-cells", NULL);
1363 		if (ip != NULL)
1364 			return *ip;
1365 	} while (np->parent);
1366 	/* No #address-cells property for the root node, default to 1 */
1367 	return 1;
1368 }
1369 EXPORT_SYMBOL(prom_n_addr_cells);
1370 
1371 int
1372 prom_n_size_cells(struct device_node* np)
1373 {
1374 	int* ip;
1375 	do {
1376 		if (np->parent)
1377 			np = np->parent;
1378 		ip = (int *) get_property(np, "#size-cells", NULL);
1379 		if (ip != NULL)
1380 			return *ip;
1381 	} while (np->parent);
1382 	/* No #size-cells property for the root node, default to 1 */
1383 	return 1;
1384 }
1385 EXPORT_SYMBOL(prom_n_size_cells);
1386 
1387 /**
1388  * Work out the sense (active-low level / active-high edge)
1389  * of each interrupt from the device tree.
1390  */
1391 void __init prom_get_irq_senses(unsigned char *senses, int off, int max)
1392 {
1393 	struct device_node *np;
1394 	int i, j;
1395 
1396 	/* default to level-triggered */
1397 	memset(senses, IRQ_SENSE_LEVEL | IRQ_POLARITY_NEGATIVE, max - off);
1398 
1399 	for (np = allnodes; np != 0; np = np->allnext) {
1400 		for (j = 0; j < np->n_intrs; j++) {
1401 			i = np->intrs[j].line;
1402 			if (i >= off && i < max)
1403 				senses[i-off] = np->intrs[j].sense;
1404 		}
1405 	}
1406 }
1407 
1408 /**
1409  * Construct and return a list of the device_nodes with a given name.
1410  */
1411 struct device_node *find_devices(const char *name)
1412 {
1413 	struct device_node *head, **prevp, *np;
1414 
1415 	prevp = &head;
1416 	for (np = allnodes; np != 0; np = np->allnext) {
1417 		if (np->name != 0 && strcasecmp(np->name, name) == 0) {
1418 			*prevp = np;
1419 			prevp = &np->next;
1420 		}
1421 	}
1422 	*prevp = NULL;
1423 	return head;
1424 }
1425 EXPORT_SYMBOL(find_devices);
1426 
1427 /**
1428  * Construct and return a list of the device_nodes with a given type.
1429  */
1430 struct device_node *find_type_devices(const char *type)
1431 {
1432 	struct device_node *head, **prevp, *np;
1433 
1434 	prevp = &head;
1435 	for (np = allnodes; np != 0; np = np->allnext) {
1436 		if (np->type != 0 && strcasecmp(np->type, type) == 0) {
1437 			*prevp = np;
1438 			prevp = &np->next;
1439 		}
1440 	}
1441 	*prevp = NULL;
1442 	return head;
1443 }
1444 EXPORT_SYMBOL(find_type_devices);
1445 
1446 /**
1447  * Returns all nodes linked together
1448  */
1449 struct device_node *find_all_nodes(void)
1450 {
1451 	struct device_node *head, **prevp, *np;
1452 
1453 	prevp = &head;
1454 	for (np = allnodes; np != 0; np = np->allnext) {
1455 		*prevp = np;
1456 		prevp = &np->next;
1457 	}
1458 	*prevp = NULL;
1459 	return head;
1460 }
1461 EXPORT_SYMBOL(find_all_nodes);
1462 
1463 /** Checks if the given "compat" string matches one of the strings in
1464  * the device's "compatible" property
1465  */
1466 int device_is_compatible(struct device_node *device, const char *compat)
1467 {
1468 	const char* cp;
1469 	int cplen, l;
1470 
1471 	cp = (char *) get_property(device, "compatible", &cplen);
1472 	if (cp == NULL)
1473 		return 0;
1474 	while (cplen > 0) {
1475 		if (strncasecmp(cp, compat, strlen(compat)) == 0)
1476 			return 1;
1477 		l = strlen(cp) + 1;
1478 		cp += l;
1479 		cplen -= l;
1480 	}
1481 
1482 	return 0;
1483 }
1484 EXPORT_SYMBOL(device_is_compatible);
1485 
1486 
1487 /**
1488  * Indicates whether the root node has a given value in its
1489  * compatible property.
1490  */
1491 int machine_is_compatible(const char *compat)
1492 {
1493 	struct device_node *root;
1494 	int rc = 0;
1495 
1496 	root = of_find_node_by_path("/");
1497 	if (root) {
1498 		rc = device_is_compatible(root, compat);
1499 		of_node_put(root);
1500 	}
1501 	return rc;
1502 }
1503 EXPORT_SYMBOL(machine_is_compatible);
1504 
1505 /**
1506  * Construct and return a list of the device_nodes with a given type
1507  * and compatible property.
1508  */
1509 struct device_node *find_compatible_devices(const char *type,
1510 					    const char *compat)
1511 {
1512 	struct device_node *head, **prevp, *np;
1513 
1514 	prevp = &head;
1515 	for (np = allnodes; np != 0; np = np->allnext) {
1516 		if (type != NULL
1517 		    && !(np->type != 0 && strcasecmp(np->type, type) == 0))
1518 			continue;
1519 		if (device_is_compatible(np, compat)) {
1520 			*prevp = np;
1521 			prevp = &np->next;
1522 		}
1523 	}
1524 	*prevp = NULL;
1525 	return head;
1526 }
1527 EXPORT_SYMBOL(find_compatible_devices);
1528 
1529 /**
1530  * Find the device_node with a given full_name.
1531  */
1532 struct device_node *find_path_device(const char *path)
1533 {
1534 	struct device_node *np;
1535 
1536 	for (np = allnodes; np != 0; np = np->allnext)
1537 		if (np->full_name != 0 && strcasecmp(np->full_name, path) == 0)
1538 			return np;
1539 	return NULL;
1540 }
1541 EXPORT_SYMBOL(find_path_device);
1542 
1543 /*******
1544  *
1545  * New implementation of the OF "find" APIs, return a refcounted
1546  * object, call of_node_put() when done.  The device tree and list
1547  * are protected by a rw_lock.
1548  *
1549  * Note that property management will need some locking as well,
1550  * this isn't dealt with yet.
1551  *
1552  *******/
1553 
1554 /**
1555  *	of_find_node_by_name - Find a node by its "name" property
1556  *	@from:	The node to start searching from or NULL, the node
1557  *		you pass will not be searched, only the next one
1558  *		will; typically, you pass what the previous call
1559  *		returned. of_node_put() will be called on it
1560  *	@name:	The name string to match against
1561  *
1562  *	Returns a node pointer with refcount incremented, use
1563  *	of_node_put() on it when done.
1564  */
1565 struct device_node *of_find_node_by_name(struct device_node *from,
1566 	const char *name)
1567 {
1568 	struct device_node *np;
1569 
1570 	read_lock(&devtree_lock);
1571 	np = from ? from->allnext : allnodes;
1572 	for (; np != NULL; np = np->allnext)
1573 		if (np->name != NULL && strcasecmp(np->name, name) == 0
1574 		    && of_node_get(np))
1575 			break;
1576 	if (from)
1577 		of_node_put(from);
1578 	read_unlock(&devtree_lock);
1579 	return np;
1580 }
1581 EXPORT_SYMBOL(of_find_node_by_name);
1582 
1583 /**
1584  *	of_find_node_by_type - Find a node by its "device_type" property
1585  *	@from:	The node to start searching from or NULL, the node
1586  *		you pass will not be searched, only the next one
1587  *		will; typically, you pass what the previous call
1588  *		returned. of_node_put() will be called on it
1589  *	@name:	The type string to match against
1590  *
1591  *	Returns a node pointer with refcount incremented, use
1592  *	of_node_put() on it when done.
1593  */
1594 struct device_node *of_find_node_by_type(struct device_node *from,
1595 	const char *type)
1596 {
1597 	struct device_node *np;
1598 
1599 	read_lock(&devtree_lock);
1600 	np = from ? from->allnext : allnodes;
1601 	for (; np != 0; np = np->allnext)
1602 		if (np->type != 0 && strcasecmp(np->type, type) == 0
1603 		    && of_node_get(np))
1604 			break;
1605 	if (from)
1606 		of_node_put(from);
1607 	read_unlock(&devtree_lock);
1608 	return np;
1609 }
1610 EXPORT_SYMBOL(of_find_node_by_type);
1611 
1612 /**
1613  *	of_find_compatible_node - Find a node based on type and one of the
1614  *                                tokens in its "compatible" property
1615  *	@from:		The node to start searching from or NULL, the node
1616  *			you pass will not be searched, only the next one
1617  *			will; typically, you pass what the previous call
1618  *			returned. of_node_put() will be called on it
1619  *	@type:		The type string to match "device_type" or NULL to ignore
1620  *	@compatible:	The string to match to one of the tokens in the device
1621  *			"compatible" list.
1622  *
1623  *	Returns a node pointer with refcount incremented, use
1624  *	of_node_put() on it when done.
1625  */
1626 struct device_node *of_find_compatible_node(struct device_node *from,
1627 	const char *type, const char *compatible)
1628 {
1629 	struct device_node *np;
1630 
1631 	read_lock(&devtree_lock);
1632 	np = from ? from->allnext : allnodes;
1633 	for (; np != 0; np = np->allnext) {
1634 		if (type != NULL
1635 		    && !(np->type != 0 && strcasecmp(np->type, type) == 0))
1636 			continue;
1637 		if (device_is_compatible(np, compatible) && of_node_get(np))
1638 			break;
1639 	}
1640 	if (from)
1641 		of_node_put(from);
1642 	read_unlock(&devtree_lock);
1643 	return np;
1644 }
1645 EXPORT_SYMBOL(of_find_compatible_node);
1646 
1647 /**
1648  *	of_find_node_by_path - Find a node matching a full OF path
1649  *	@path:	The full path to match
1650  *
1651  *	Returns a node pointer with refcount incremented, use
1652  *	of_node_put() on it when done.
1653  */
1654 struct device_node *of_find_node_by_path(const char *path)
1655 {
1656 	struct device_node *np = allnodes;
1657 
1658 	read_lock(&devtree_lock);
1659 	for (; np != 0; np = np->allnext) {
1660 		if (np->full_name != 0 && strcasecmp(np->full_name, path) == 0
1661 		    && of_node_get(np))
1662 			break;
1663 	}
1664 	read_unlock(&devtree_lock);
1665 	return np;
1666 }
1667 EXPORT_SYMBOL(of_find_node_by_path);
1668 
1669 /**
1670  *	of_find_node_by_phandle - Find a node given a phandle
1671  *	@handle:	phandle of the node to find
1672  *
1673  *	Returns a node pointer with refcount incremented, use
1674  *	of_node_put() on it when done.
1675  */
1676 struct device_node *of_find_node_by_phandle(phandle handle)
1677 {
1678 	struct device_node *np;
1679 
1680 	read_lock(&devtree_lock);
1681 	for (np = allnodes; np != 0; np = np->allnext)
1682 		if (np->linux_phandle == handle)
1683 			break;
1684 	if (np)
1685 		of_node_get(np);
1686 	read_unlock(&devtree_lock);
1687 	return np;
1688 }
1689 EXPORT_SYMBOL(of_find_node_by_phandle);
1690 
1691 /**
1692  *	of_find_all_nodes - Get next node in global list
1693  *	@prev:	Previous node or NULL to start iteration
1694  *		of_node_put() will be called on it
1695  *
1696  *	Returns a node pointer with refcount incremented, use
1697  *	of_node_put() on it when done.
1698  */
1699 struct device_node *of_find_all_nodes(struct device_node *prev)
1700 {
1701 	struct device_node *np;
1702 
1703 	read_lock(&devtree_lock);
1704 	np = prev ? prev->allnext : allnodes;
1705 	for (; np != 0; np = np->allnext)
1706 		if (of_node_get(np))
1707 			break;
1708 	if (prev)
1709 		of_node_put(prev);
1710 	read_unlock(&devtree_lock);
1711 	return np;
1712 }
1713 EXPORT_SYMBOL(of_find_all_nodes);
1714 
1715 /**
1716  *	of_get_parent - Get a node's parent if any
1717  *	@node:	Node to get parent
1718  *
1719  *	Returns a node pointer with refcount incremented, use
1720  *	of_node_put() on it when done.
1721  */
1722 struct device_node *of_get_parent(const struct device_node *node)
1723 {
1724 	struct device_node *np;
1725 
1726 	if (!node)
1727 		return NULL;
1728 
1729 	read_lock(&devtree_lock);
1730 	np = of_node_get(node->parent);
1731 	read_unlock(&devtree_lock);
1732 	return np;
1733 }
1734 EXPORT_SYMBOL(of_get_parent);
1735 
1736 /**
1737  *	of_get_next_child - Iterate a node childs
1738  *	@node:	parent node
1739  *	@prev:	previous child of the parent node, or NULL to get first
1740  *
1741  *	Returns a node pointer with refcount incremented, use
1742  *	of_node_put() on it when done.
1743  */
1744 struct device_node *of_get_next_child(const struct device_node *node,
1745 	struct device_node *prev)
1746 {
1747 	struct device_node *next;
1748 
1749 	read_lock(&devtree_lock);
1750 	next = prev ? prev->sibling : node->child;
1751 	for (; next != 0; next = next->sibling)
1752 		if (of_node_get(next))
1753 			break;
1754 	if (prev)
1755 		of_node_put(prev);
1756 	read_unlock(&devtree_lock);
1757 	return next;
1758 }
1759 EXPORT_SYMBOL(of_get_next_child);
1760 
1761 /**
1762  *	of_node_get - Increment refcount of a node
1763  *	@node:	Node to inc refcount, NULL is supported to
1764  *		simplify writing of callers
1765  *
1766  *	Returns node.
1767  */
1768 struct device_node *of_node_get(struct device_node *node)
1769 {
1770 	if (node)
1771 		kref_get(&node->kref);
1772 	return node;
1773 }
1774 EXPORT_SYMBOL(of_node_get);
1775 
1776 static inline struct device_node * kref_to_device_node(struct kref *kref)
1777 {
1778 	return container_of(kref, struct device_node, kref);
1779 }
1780 
1781 /**
1782  *	of_node_release - release a dynamically allocated node
1783  *	@kref:  kref element of the node to be released
1784  *
1785  *	In of_node_put() this function is passed to kref_put()
1786  *	as the destructor.
1787  */
1788 static void of_node_release(struct kref *kref)
1789 {
1790 	struct device_node *node = kref_to_device_node(kref);
1791 	struct property *prop = node->properties;
1792 
1793 	if (!OF_IS_DYNAMIC(node))
1794 		return;
1795 	while (prop) {
1796 		struct property *next = prop->next;
1797 		kfree(prop->name);
1798 		kfree(prop->value);
1799 		kfree(prop);
1800 		prop = next;
1801 
1802 		if (!prop) {
1803 			prop = node->deadprops;
1804 			node->deadprops = NULL;
1805 		}
1806 	}
1807 	kfree(node->intrs);
1808 	kfree(node->full_name);
1809 	kfree(node->data);
1810 	kfree(node);
1811 }
1812 
1813 /**
1814  *	of_node_put - Decrement refcount of a node
1815  *	@node:	Node to dec refcount, NULL is supported to
1816  *		simplify writing of callers
1817  *
1818  */
1819 void of_node_put(struct device_node *node)
1820 {
1821 	if (node)
1822 		kref_put(&node->kref, of_node_release);
1823 }
1824 EXPORT_SYMBOL(of_node_put);
1825 
1826 /*
1827  * Plug a device node into the tree and global list.
1828  */
1829 void of_attach_node(struct device_node *np)
1830 {
1831 	write_lock(&devtree_lock);
1832 	np->sibling = np->parent->child;
1833 	np->allnext = allnodes;
1834 	np->parent->child = np;
1835 	allnodes = np;
1836 	write_unlock(&devtree_lock);
1837 }
1838 
1839 /*
1840  * "Unplug" a node from the device tree.  The caller must hold
1841  * a reference to the node.  The memory associated with the node
1842  * is not freed until its refcount goes to zero.
1843  */
1844 void of_detach_node(const struct device_node *np)
1845 {
1846 	struct device_node *parent;
1847 
1848 	write_lock(&devtree_lock);
1849 
1850 	parent = np->parent;
1851 
1852 	if (allnodes == np)
1853 		allnodes = np->allnext;
1854 	else {
1855 		struct device_node *prev;
1856 		for (prev = allnodes;
1857 		     prev->allnext != np;
1858 		     prev = prev->allnext)
1859 			;
1860 		prev->allnext = np->allnext;
1861 	}
1862 
1863 	if (parent->child == np)
1864 		parent->child = np->sibling;
1865 	else {
1866 		struct device_node *prevsib;
1867 		for (prevsib = np->parent->child;
1868 		     prevsib->sibling != np;
1869 		     prevsib = prevsib->sibling)
1870 			;
1871 		prevsib->sibling = np->sibling;
1872 	}
1873 
1874 	write_unlock(&devtree_lock);
1875 }
1876 
1877 #ifdef CONFIG_PPC_PSERIES
1878 /*
1879  * Fix up the uninitialized fields in a new device node:
1880  * name, type, n_addrs, addrs, n_intrs, intrs, and pci-specific fields
1881  *
1882  * A lot of boot-time code is duplicated here, because functions such
1883  * as finish_node_interrupts, interpret_pci_props, etc. cannot use the
1884  * slab allocator.
1885  *
1886  * This should probably be split up into smaller chunks.
1887  */
1888 
1889 static int of_finish_dynamic_node(struct device_node *node)
1890 {
1891 	struct device_node *parent = of_get_parent(node);
1892 	int err = 0;
1893 	phandle *ibm_phandle;
1894 
1895 	node->name = get_property(node, "name", NULL);
1896 	node->type = get_property(node, "device_type", NULL);
1897 
1898 	if (!parent) {
1899 		err = -ENODEV;
1900 		goto out;
1901 	}
1902 
1903 	/* We don't support that function on PowerMac, at least
1904 	 * not yet
1905 	 */
1906 	if (machine_is(powermac))
1907 		return -ENODEV;
1908 
1909 	/* fix up new node's linux_phandle field */
1910 	if ((ibm_phandle = (unsigned int *)get_property(node,
1911 							"ibm,phandle", NULL)))
1912 		node->linux_phandle = *ibm_phandle;
1913 
1914 out:
1915 	of_node_put(parent);
1916 	return err;
1917 }
1918 
1919 static int prom_reconfig_notifier(struct notifier_block *nb,
1920 				  unsigned long action, void *node)
1921 {
1922 	int err;
1923 
1924 	switch (action) {
1925 	case PSERIES_RECONFIG_ADD:
1926 		err = of_finish_dynamic_node(node);
1927 		if (!err)
1928 			finish_node(node, NULL, 0);
1929 		if (err < 0) {
1930 			printk(KERN_ERR "finish_node returned %d\n", err);
1931 			err = NOTIFY_BAD;
1932 		}
1933 		break;
1934 	default:
1935 		err = NOTIFY_DONE;
1936 		break;
1937 	}
1938 	return err;
1939 }
1940 
1941 static struct notifier_block prom_reconfig_nb = {
1942 	.notifier_call = prom_reconfig_notifier,
1943 	.priority = 10, /* This one needs to run first */
1944 };
1945 
1946 static int __init prom_reconfig_setup(void)
1947 {
1948 	return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
1949 }
1950 __initcall(prom_reconfig_setup);
1951 #endif
1952 
1953 struct property *of_find_property(struct device_node *np, const char *name,
1954 				  int *lenp)
1955 {
1956 	struct property *pp;
1957 
1958 	read_lock(&devtree_lock);
1959 	for (pp = np->properties; pp != 0; pp = pp->next)
1960 		if (strcmp(pp->name, name) == 0) {
1961 			if (lenp != 0)
1962 				*lenp = pp->length;
1963 			break;
1964 		}
1965 	read_unlock(&devtree_lock);
1966 
1967 	return pp;
1968 }
1969 
1970 /*
1971  * Find a property with a given name for a given node
1972  * and return the value.
1973  */
1974 unsigned char *get_property(struct device_node *np, const char *name,
1975 			    int *lenp)
1976 {
1977 	struct property *pp = of_find_property(np,name,lenp);
1978 	return pp ? pp->value : NULL;
1979 }
1980 EXPORT_SYMBOL(get_property);
1981 
1982 /*
1983  * Add a property to a node
1984  */
1985 int prom_add_property(struct device_node* np, struct property* prop)
1986 {
1987 	struct property **next;
1988 
1989 	prop->next = NULL;
1990 	write_lock(&devtree_lock);
1991 	next = &np->properties;
1992 	while (*next) {
1993 		if (strcmp(prop->name, (*next)->name) == 0) {
1994 			/* duplicate ! don't insert it */
1995 			write_unlock(&devtree_lock);
1996 			return -1;
1997 		}
1998 		next = &(*next)->next;
1999 	}
2000 	*next = prop;
2001 	write_unlock(&devtree_lock);
2002 
2003 #ifdef CONFIG_PROC_DEVICETREE
2004 	/* try to add to proc as well if it was initialized */
2005 	if (np->pde)
2006 		proc_device_tree_add_prop(np->pde, prop);
2007 #endif /* CONFIG_PROC_DEVICETREE */
2008 
2009 	return 0;
2010 }
2011 
2012 /*
2013  * Remove a property from a node.  Note that we don't actually
2014  * remove it, since we have given out who-knows-how-many pointers
2015  * to the data using get-property.  Instead we just move the property
2016  * to the "dead properties" list, so it won't be found any more.
2017  */
2018 int prom_remove_property(struct device_node *np, struct property *prop)
2019 {
2020 	struct property **next;
2021 	int found = 0;
2022 
2023 	write_lock(&devtree_lock);
2024 	next = &np->properties;
2025 	while (*next) {
2026 		if (*next == prop) {
2027 			/* found the node */
2028 			*next = prop->next;
2029 			prop->next = np->deadprops;
2030 			np->deadprops = prop;
2031 			found = 1;
2032 			break;
2033 		}
2034 		next = &(*next)->next;
2035 	}
2036 	write_unlock(&devtree_lock);
2037 
2038 	if (!found)
2039 		return -ENODEV;
2040 
2041 #ifdef CONFIG_PROC_DEVICETREE
2042 	/* try to remove the proc node as well */
2043 	if (np->pde)
2044 		proc_device_tree_remove_prop(np->pde, prop);
2045 #endif /* CONFIG_PROC_DEVICETREE */
2046 
2047 	return 0;
2048 }
2049 
2050 /*
2051  * Update a property in a node.  Note that we don't actually
2052  * remove it, since we have given out who-knows-how-many pointers
2053  * to the data using get-property.  Instead we just move the property
2054  * to the "dead properties" list, and add the new property to the
2055  * property list
2056  */
2057 int prom_update_property(struct device_node *np,
2058 			 struct property *newprop,
2059 			 struct property *oldprop)
2060 {
2061 	struct property **next;
2062 	int found = 0;
2063 
2064 	write_lock(&devtree_lock);
2065 	next = &np->properties;
2066 	while (*next) {
2067 		if (*next == oldprop) {
2068 			/* found the node */
2069 			newprop->next = oldprop->next;
2070 			*next = newprop;
2071 			oldprop->next = np->deadprops;
2072 			np->deadprops = oldprop;
2073 			found = 1;
2074 			break;
2075 		}
2076 		next = &(*next)->next;
2077 	}
2078 	write_unlock(&devtree_lock);
2079 
2080 	if (!found)
2081 		return -ENODEV;
2082 
2083 #ifdef CONFIG_PROC_DEVICETREE
2084 	/* try to add to proc as well if it was initialized */
2085 	if (np->pde)
2086 		proc_device_tree_update_prop(np->pde, newprop, oldprop);
2087 #endif /* CONFIG_PROC_DEVICETREE */
2088 
2089 	return 0;
2090 }
2091 
2092