Merge branch 'linus' into core/iommu
[linux-2.6] / arch / powerpc / kernel / prom.c
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/kernel.h>
20 #include <linux/string.h>
21 #include <linux/init.h>
22 #include <linux/threads.h>
23 #include <linux/spinlock.h>
24 #include <linux/types.h>
25 #include <linux/pci.h>
26 #include <linux/stringify.h>
27 #include <linux/delay.h>
28 #include <linux/initrd.h>
29 #include <linux/bitops.h>
30 #include <linux/module.h>
31 #include <linux/kexec.h>
32 #include <linux/debugfs.h>
33 #include <linux/irq.h>
34 #include <linux/lmb.h>
35
36 #include <asm/prom.h>
37 #include <asm/rtas.h>
38 #include <asm/page.h>
39 #include <asm/processor.h>
40 #include <asm/irq.h>
41 #include <asm/io.h>
42 #include <asm/kdump.h>
43 #include <asm/smp.h>
44 #include <asm/system.h>
45 #include <asm/mmu.h>
46 #include <asm/pgtable.h>
47 #include <asm/pci.h>
48 #include <asm/iommu.h>
49 #include <asm/btext.h>
50 #include <asm/sections.h>
51 #include <asm/machdep.h>
52 #include <asm/pSeries_reconfig.h>
53 #include <asm/pci-bridge.h>
54 #include <asm/phyp_dump.h>
55 #include <asm/kexec.h>
56 #include <mm/mmu_decl.h>
57
58 #ifdef DEBUG
59 #define DBG(fmt...) printk(KERN_ERR fmt)
60 #else
61 #define DBG(fmt...)
62 #endif
63
64
65 static int __initdata dt_root_addr_cells;
66 static int __initdata dt_root_size_cells;
67
68 #ifdef CONFIG_PPC64
69 int __initdata iommu_is_off;
70 int __initdata iommu_force_on;
71 unsigned long tce_alloc_start, tce_alloc_end;
72 #endif
73
74 typedef u32 cell_t;
75
76 #if 0
77 static struct boot_param_header *initial_boot_params __initdata;
78 #else
79 struct boot_param_header *initial_boot_params;
80 #endif
81
82 extern struct device_node *allnodes;    /* temporary while merging */
83
84 extern rwlock_t devtree_lock;   /* temporary while merging */
85
86 /* export that to outside world */
87 struct device_node *of_chosen;
88
89 static inline char *find_flat_dt_string(u32 offset)
90 {
91         return ((char *)initial_boot_params) +
92                 initial_boot_params->off_dt_strings + offset;
93 }
94
95 /**
96  * This function is used to scan the flattened device-tree, it is
97  * used to extract the memory informations at boot before we can
98  * unflatten the tree
99  */
100 int __init of_scan_flat_dt(int (*it)(unsigned long node,
101                                      const char *uname, int depth,
102                                      void *data),
103                            void *data)
104 {
105         unsigned long p = ((unsigned long)initial_boot_params) +
106                 initial_boot_params->off_dt_struct;
107         int rc = 0;
108         int depth = -1;
109
110         do {
111                 u32 tag = *((u32 *)p);
112                 char *pathp;
113                 
114                 p += 4;
115                 if (tag == OF_DT_END_NODE) {
116                         depth --;
117                         continue;
118                 }
119                 if (tag == OF_DT_NOP)
120                         continue;
121                 if (tag == OF_DT_END)
122                         break;
123                 if (tag == OF_DT_PROP) {
124                         u32 sz = *((u32 *)p);
125                         p += 8;
126                         if (initial_boot_params->version < 0x10)
127                                 p = _ALIGN(p, sz >= 8 ? 8 : 4);
128                         p += sz;
129                         p = _ALIGN(p, 4);
130                         continue;
131                 }
132                 if (tag != OF_DT_BEGIN_NODE) {
133                         printk(KERN_WARNING "Invalid tag %x scanning flattened"
134                                " device tree !\n", tag);
135                         return -EINVAL;
136                 }
137                 depth++;
138                 pathp = (char *)p;
139                 p = _ALIGN(p + strlen(pathp) + 1, 4);
140                 if ((*pathp) == '/') {
141                         char *lp, *np;
142                         for (lp = NULL, np = pathp; *np; np++)
143                                 if ((*np) == '/')
144                                         lp = np+1;
145                         if (lp != NULL)
146                                 pathp = lp;
147                 }
148                 rc = it(p, pathp, depth, data);
149                 if (rc != 0)
150                         break;          
151         } while(1);
152
153         return rc;
154 }
155
156 unsigned long __init of_get_flat_dt_root(void)
157 {
158         unsigned long p = ((unsigned long)initial_boot_params) +
159                 initial_boot_params->off_dt_struct;
160
161         while(*((u32 *)p) == OF_DT_NOP)
162                 p += 4;
163         BUG_ON (*((u32 *)p) != OF_DT_BEGIN_NODE);
164         p += 4;
165         return _ALIGN(p + strlen((char *)p) + 1, 4);
166 }
167
168 /**
169  * This  function can be used within scan_flattened_dt callback to get
170  * access to properties
171  */
172 void* __init of_get_flat_dt_prop(unsigned long node, const char *name,
173                                  unsigned long *size)
174 {
175         unsigned long p = node;
176
177         do {
178                 u32 tag = *((u32 *)p);
179                 u32 sz, noff;
180                 const char *nstr;
181
182                 p += 4;
183                 if (tag == OF_DT_NOP)
184                         continue;
185                 if (tag != OF_DT_PROP)
186                         return NULL;
187
188                 sz = *((u32 *)p);
189                 noff = *((u32 *)(p + 4));
190                 p += 8;
191                 if (initial_boot_params->version < 0x10)
192                         p = _ALIGN(p, sz >= 8 ? 8 : 4);
193
194                 nstr = find_flat_dt_string(noff);
195                 if (nstr == NULL) {
196                         printk(KERN_WARNING "Can't find property index"
197                                " name !\n");
198                         return NULL;
199                 }
200                 if (strcmp(name, nstr) == 0) {
201                         if (size)
202                                 *size = sz;
203                         return (void *)p;
204                 }
205                 p += sz;
206                 p = _ALIGN(p, 4);
207         } while(1);
208 }
209
210 int __init of_flat_dt_is_compatible(unsigned long node, const char *compat)
211 {
212         const char* cp;
213         unsigned long cplen, l;
214
215         cp = of_get_flat_dt_prop(node, "compatible", &cplen);
216         if (cp == NULL)
217                 return 0;
218         while (cplen > 0) {
219                 if (strncasecmp(cp, compat, strlen(compat)) == 0)
220                         return 1;
221                 l = strlen(cp) + 1;
222                 cp += l;
223                 cplen -= l;
224         }
225
226         return 0;
227 }
228
229 static void *__init unflatten_dt_alloc(unsigned long *mem, unsigned long size,
230                                        unsigned long align)
231 {
232         void *res;
233
234         *mem = _ALIGN(*mem, align);
235         res = (void *)*mem;
236         *mem += size;
237
238         return res;
239 }
240
241 static unsigned long __init unflatten_dt_node(unsigned long mem,
242                                               unsigned long *p,
243                                               struct device_node *dad,
244                                               struct device_node ***allnextpp,
245                                               unsigned long fpsize)
246 {
247         struct device_node *np;
248         struct property *pp, **prev_pp = NULL;
249         char *pathp;
250         u32 tag;
251         unsigned int l, allocl;
252         int has_name = 0;
253         int new_format = 0;
254
255         tag = *((u32 *)(*p));
256         if (tag != OF_DT_BEGIN_NODE) {
257                 printk("Weird tag at start of node: %x\n", tag);
258                 return mem;
259         }
260         *p += 4;
261         pathp = (char *)*p;
262         l = allocl = strlen(pathp) + 1;
263         *p = _ALIGN(*p + l, 4);
264
265         /* version 0x10 has a more compact unit name here instead of the full
266          * path. we accumulate the full path size using "fpsize", we'll rebuild
267          * it later. We detect this because the first character of the name is
268          * not '/'.
269          */
270         if ((*pathp) != '/') {
271                 new_format = 1;
272                 if (fpsize == 0) {
273                         /* root node: special case. fpsize accounts for path
274                          * plus terminating zero. root node only has '/', so
275                          * fpsize should be 2, but we want to avoid the first
276                          * level nodes to have two '/' so we use fpsize 1 here
277                          */
278                         fpsize = 1;
279                         allocl = 2;
280                 } else {
281                         /* account for '/' and path size minus terminal 0
282                          * already in 'l'
283                          */
284                         fpsize += l;
285                         allocl = fpsize;
286                 }
287         }
288
289
290         np = unflatten_dt_alloc(&mem, sizeof(struct device_node) + allocl,
291                                 __alignof__(struct device_node));
292         if (allnextpp) {
293                 memset(np, 0, sizeof(*np));
294                 np->full_name = ((char*)np) + sizeof(struct device_node);
295                 if (new_format) {
296                         char *p = np->full_name;
297                         /* rebuild full path for new format */
298                         if (dad && dad->parent) {
299                                 strcpy(p, dad->full_name);
300 #ifdef DEBUG
301                                 if ((strlen(p) + l + 1) != allocl) {
302                                         DBG("%s: p: %d, l: %d, a: %d\n",
303                                             pathp, (int)strlen(p), l, allocl);
304                                 }
305 #endif
306                                 p += strlen(p);
307                         }
308                         *(p++) = '/';
309                         memcpy(p, pathp, l);
310                 } else
311                         memcpy(np->full_name, pathp, l);
312                 prev_pp = &np->properties;
313                 **allnextpp = np;
314                 *allnextpp = &np->allnext;
315                 if (dad != NULL) {
316                         np->parent = dad;
317                         /* we temporarily use the next field as `last_child'*/
318                         if (dad->next == 0)
319                                 dad->child = np;
320                         else
321                                 dad->next->sibling = np;
322                         dad->next = np;
323                 }
324                 kref_init(&np->kref);
325         }
326         while(1) {
327                 u32 sz, noff;
328                 char *pname;
329
330                 tag = *((u32 *)(*p));
331                 if (tag == OF_DT_NOP) {
332                         *p += 4;
333                         continue;
334                 }
335                 if (tag != OF_DT_PROP)
336                         break;
337                 *p += 4;
338                 sz = *((u32 *)(*p));
339                 noff = *((u32 *)((*p) + 4));
340                 *p += 8;
341                 if (initial_boot_params->version < 0x10)
342                         *p = _ALIGN(*p, sz >= 8 ? 8 : 4);
343
344                 pname = find_flat_dt_string(noff);
345                 if (pname == NULL) {
346                         printk("Can't find property name in list !\n");
347                         break;
348                 }
349                 if (strcmp(pname, "name") == 0)
350                         has_name = 1;
351                 l = strlen(pname) + 1;
352                 pp = unflatten_dt_alloc(&mem, sizeof(struct property),
353                                         __alignof__(struct property));
354                 if (allnextpp) {
355                         if (strcmp(pname, "linux,phandle") == 0) {
356                                 np->node = *((u32 *)*p);
357                                 if (np->linux_phandle == 0)
358                                         np->linux_phandle = np->node;
359                         }
360                         if (strcmp(pname, "ibm,phandle") == 0)
361                                 np->linux_phandle = *((u32 *)*p);
362                         pp->name = pname;
363                         pp->length = sz;
364                         pp->value = (void *)*p;
365                         *prev_pp = pp;
366                         prev_pp = &pp->next;
367                 }
368                 *p = _ALIGN((*p) + sz, 4);
369         }
370         /* with version 0x10 we may not have the name property, recreate
371          * it here from the unit name if absent
372          */
373         if (!has_name) {
374                 char *p = pathp, *ps = pathp, *pa = NULL;
375                 int sz;
376
377                 while (*p) {
378                         if ((*p) == '@')
379                                 pa = p;
380                         if ((*p) == '/')
381                                 ps = p + 1;
382                         p++;
383                 }
384                 if (pa < ps)
385                         pa = p;
386                 sz = (pa - ps) + 1;
387                 pp = unflatten_dt_alloc(&mem, sizeof(struct property) + sz,
388                                         __alignof__(struct property));
389                 if (allnextpp) {
390                         pp->name = "name";
391                         pp->length = sz;
392                         pp->value = pp + 1;
393                         *prev_pp = pp;
394                         prev_pp = &pp->next;
395                         memcpy(pp->value, ps, sz - 1);
396                         ((char *)pp->value)[sz - 1] = 0;
397                         DBG("fixed up name for %s -> %s\n", pathp,
398                                 (char *)pp->value);
399                 }
400         }
401         if (allnextpp) {
402                 *prev_pp = NULL;
403                 np->name = of_get_property(np, "name", NULL);
404                 np->type = of_get_property(np, "device_type", NULL);
405
406                 if (!np->name)
407                         np->name = "<NULL>";
408                 if (!np->type)
409                         np->type = "<NULL>";
410         }
411         while (tag == OF_DT_BEGIN_NODE) {
412                 mem = unflatten_dt_node(mem, p, np, allnextpp, fpsize);
413                 tag = *((u32 *)(*p));
414         }
415         if (tag != OF_DT_END_NODE) {
416                 printk("Weird tag at end of node: %x\n", tag);
417                 return mem;
418         }
419         *p += 4;
420         return mem;
421 }
422
423 static int __init early_parse_mem(char *p)
424 {
425         if (!p)
426                 return 1;
427
428         memory_limit = PAGE_ALIGN(memparse(p, &p));
429         DBG("memory limit = 0x%lx\n", memory_limit);
430
431         return 0;
432 }
433 early_param("mem", early_parse_mem);
434
435 /**
436  * move_device_tree - move tree to an unused area, if needed.
437  *
438  * The device tree may be allocated beyond our memory limit, or inside the
439  * crash kernel region for kdump. If so, move it out of the way.
440  */
441 static void __init move_device_tree(void)
442 {
443         unsigned long start, size;
444         void *p;
445
446         DBG("-> move_device_tree\n");
447
448         start = __pa(initial_boot_params);
449         size = initial_boot_params->totalsize;
450
451         if ((memory_limit && (start + size) > memory_limit) ||
452                         overlaps_crashkernel(start, size)) {
453                 p = __va(lmb_alloc_base(size, PAGE_SIZE, lmb.rmo_size));
454                 memcpy(p, initial_boot_params, size);
455                 initial_boot_params = (struct boot_param_header *)p;
456                 DBG("Moved device tree to 0x%p\n", p);
457         }
458
459         DBG("<- move_device_tree\n");
460 }
461
462 /**
463  * unflattens the device-tree passed by the firmware, creating the
464  * tree of struct device_node. It also fills the "name" and "type"
465  * pointers of the nodes so the normal device-tree walking functions
466  * can be used (this used to be done by finish_device_tree)
467  */
468 void __init unflatten_device_tree(void)
469 {
470         unsigned long start, mem, size;
471         struct device_node **allnextp = &allnodes;
472
473         DBG(" -> unflatten_device_tree()\n");
474
475         /* First pass, scan for size */
476         start = ((unsigned long)initial_boot_params) +
477                 initial_boot_params->off_dt_struct;
478         size = unflatten_dt_node(0, &start, NULL, NULL, 0);
479         size = (size | 3) + 1;
480
481         DBG("  size is %lx, allocating...\n", size);
482
483         /* Allocate memory for the expanded device tree */
484         mem = lmb_alloc(size + 4, __alignof__(struct device_node));
485         mem = (unsigned long) __va(mem);
486
487         ((u32 *)mem)[size / 4] = 0xdeadbeef;
488
489         DBG("  unflattening %lx...\n", mem);
490
491         /* Second pass, do actual unflattening */
492         start = ((unsigned long)initial_boot_params) +
493                 initial_boot_params->off_dt_struct;
494         unflatten_dt_node(mem, &start, NULL, &allnextp, 0);
495         if (*((u32 *)start) != OF_DT_END)
496                 printk(KERN_WARNING "Weird tag at end of tree: %08x\n", *((u32 *)start));
497         if (((u32 *)mem)[size / 4] != 0xdeadbeef)
498                 printk(KERN_WARNING "End of tree marker overwritten: %08x\n",
499                        ((u32 *)mem)[size / 4] );
500         *allnextp = NULL;
501
502         /* Get pointer to OF "/chosen" node for use everywhere */
503         of_chosen = of_find_node_by_path("/chosen");
504         if (of_chosen == NULL)
505                 of_chosen = of_find_node_by_path("/chosen@0");
506
507         DBG(" <- unflatten_device_tree()\n");
508 }
509
510 /*
511  * ibm,pa-features is a per-cpu property that contains a string of
512  * attribute descriptors, each of which has a 2 byte header plus up
513  * to 254 bytes worth of processor attribute bits.  First header
514  * byte specifies the number of bytes following the header.
515  * Second header byte is an "attribute-specifier" type, of which
516  * zero is the only currently-defined value.
517  * Implementation:  Pass in the byte and bit offset for the feature
518  * that we are interested in.  The function will return -1 if the
519  * pa-features property is missing, or a 1/0 to indicate if the feature
520  * is supported/not supported.  Note that the bit numbers are
521  * big-endian to match the definition in PAPR.
522  */
523 static struct ibm_pa_feature {
524         unsigned long   cpu_features;   /* CPU_FTR_xxx bit */
525         unsigned int    cpu_user_ftrs;  /* PPC_FEATURE_xxx bit */
526         unsigned char   pabyte;         /* byte number in ibm,pa-features */
527         unsigned char   pabit;          /* bit number (big-endian) */
528         unsigned char   invert;         /* if 1, pa bit set => clear feature */
529 } ibm_pa_features[] __initdata = {
530         {0, PPC_FEATURE_HAS_MMU,        0, 0, 0},
531         {0, PPC_FEATURE_HAS_FPU,        0, 1, 0},
532         {CPU_FTR_SLB, 0,                0, 2, 0},
533         {CPU_FTR_CTRL, 0,               0, 3, 0},
534         {CPU_FTR_NOEXECUTE, 0,          0, 6, 0},
535         {CPU_FTR_NODSISRALIGN, 0,       1, 1, 1},
536         {CPU_FTR_CI_LARGE_PAGE, 0,      1, 2, 0},
537         {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
538 };
539
540 static void __init scan_features(unsigned long node, unsigned char *ftrs,
541                                  unsigned long tablelen,
542                                  struct ibm_pa_feature *fp,
543                                  unsigned long ft_size)
544 {
545         unsigned long i, len, bit;
546
547         /* find descriptor with type == 0 */
548         for (;;) {
549                 if (tablelen < 3)
550                         return;
551                 len = 2 + ftrs[0];
552                 if (tablelen < len)
553                         return;         /* descriptor 0 not found */
554                 if (ftrs[1] == 0)
555                         break;
556                 tablelen -= len;
557                 ftrs += len;
558         }
559
560         /* loop over bits we know about */
561         for (i = 0; i < ft_size; ++i, ++fp) {
562                 if (fp->pabyte >= ftrs[0])
563                         continue;
564                 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
565                 if (bit ^ fp->invert) {
566                         cur_cpu_spec->cpu_features |= fp->cpu_features;
567                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
568                 } else {
569                         cur_cpu_spec->cpu_features &= ~fp->cpu_features;
570                         cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
571                 }
572         }
573 }
574
575 static void __init check_cpu_pa_features(unsigned long node)
576 {
577         unsigned char *pa_ftrs;
578         unsigned long tablelen;
579
580         pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
581         if (pa_ftrs == NULL)
582                 return;
583
584         scan_features(node, pa_ftrs, tablelen,
585                       ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
586 }
587
588 #ifdef CONFIG_PPC64
589 static void __init check_cpu_slb_size(unsigned long node)
590 {
591         u32 *slb_size_ptr;
592
593         slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
594         if (slb_size_ptr != NULL) {
595                 mmu_slb_size = *slb_size_ptr;
596                 return;
597         }
598         slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
599         if (slb_size_ptr != NULL) {
600                 mmu_slb_size = *slb_size_ptr;
601         }
602 }
603 #else
604 #define check_cpu_slb_size(node) do { } while(0)
605 #endif
606
607 static struct feature_property {
608         const char *name;
609         u32 min_value;
610         unsigned long cpu_feature;
611         unsigned long cpu_user_ftr;
612 } feature_properties[] __initdata = {
613 #ifdef CONFIG_ALTIVEC
614         {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
615         {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
616 #endif /* CONFIG_ALTIVEC */
617 #ifdef CONFIG_VSX
618         /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
619         {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
620 #endif /* CONFIG_VSX */
621 #ifdef CONFIG_PPC64
622         {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
623         {"ibm,purr", 1, CPU_FTR_PURR, 0},
624         {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
625 #endif /* CONFIG_PPC64 */
626 };
627
628 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
629 static inline void identical_pvr_fixup(unsigned long node)
630 {
631         unsigned int pvr;
632         char *model = of_get_flat_dt_prop(node, "model", NULL);
633
634         /*
635          * Since 440GR(x)/440EP(x) processors have the same pvr,
636          * we check the node path and set bit 28 in the cur_cpu_spec
637          * pvr for EP(x) processor version. This bit is always 0 in
638          * the "real" pvr. Then we call identify_cpu again with
639          * the new logical pvr to enable FPU support.
640          */
641         if (model && strstr(model, "440EP")) {
642                 pvr = cur_cpu_spec->pvr_value | 0x8;
643                 identify_cpu(0, pvr);
644                 DBG("Using logical pvr %x for %s\n", pvr, model);
645         }
646 }
647 #else
648 #define identical_pvr_fixup(node) do { } while(0)
649 #endif
650
651 static void __init check_cpu_feature_properties(unsigned long node)
652 {
653         unsigned long i;
654         struct feature_property *fp = feature_properties;
655         const u32 *prop;
656
657         for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
658                 prop = of_get_flat_dt_prop(node, fp->name, NULL);
659                 if (prop && *prop >= fp->min_value) {
660                         cur_cpu_spec->cpu_features |= fp->cpu_feature;
661                         cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
662                 }
663         }
664 }
665
666 static int __init early_init_dt_scan_cpus(unsigned long node,
667                                           const char *uname, int depth,
668                                           void *data)
669 {
670         static int logical_cpuid = 0;
671         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
672         const u32 *prop;
673         const u32 *intserv;
674         int i, nthreads;
675         unsigned long len;
676         int found = 0;
677
678         /* We are scanning "cpu" nodes only */
679         if (type == NULL || strcmp(type, "cpu") != 0)
680                 return 0;
681
682         /* Get physical cpuid */
683         intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
684         if (intserv) {
685                 nthreads = len / sizeof(int);
686         } else {
687                 intserv = of_get_flat_dt_prop(node, "reg", NULL);
688                 nthreads = 1;
689         }
690
691         /*
692          * Now see if any of these threads match our boot cpu.
693          * NOTE: This must match the parsing done in smp_setup_cpu_maps.
694          */
695         for (i = 0; i < nthreads; i++) {
696                 /*
697                  * version 2 of the kexec param format adds the phys cpuid of
698                  * booted proc.
699                  */
700                 if (initial_boot_params && initial_boot_params->version >= 2) {
701                         if (intserv[i] ==
702                                         initial_boot_params->boot_cpuid_phys) {
703                                 found = 1;
704                                 break;
705                         }
706                 } else {
707                         /*
708                          * Check if it's the boot-cpu, set it's hw index now,
709                          * unfortunately this format did not support booting
710                          * off secondary threads.
711                          */
712                         if (of_get_flat_dt_prop(node,
713                                         "linux,boot-cpu", NULL) != NULL) {
714                                 found = 1;
715                                 break;
716                         }
717                 }
718
719 #ifdef CONFIG_SMP
720                 /* logical cpu id is always 0 on UP kernels */
721                 logical_cpuid++;
722 #endif
723         }
724
725         if (found) {
726                 DBG("boot cpu: logical %d physical %d\n", logical_cpuid,
727                         intserv[i]);
728                 boot_cpuid = logical_cpuid;
729                 set_hard_smp_processor_id(boot_cpuid, intserv[i]);
730
731                 /*
732                  * PAPR defines "logical" PVR values for cpus that
733                  * meet various levels of the architecture:
734                  * 0x0f000001   Architecture version 2.04
735                  * 0x0f000002   Architecture version 2.05
736                  * If the cpu-version property in the cpu node contains
737                  * such a value, we call identify_cpu again with the
738                  * logical PVR value in order to use the cpu feature
739                  * bits appropriate for the architecture level.
740                  *
741                  * A POWER6 partition in "POWER6 architected" mode
742                  * uses the 0x0f000002 PVR value; in POWER5+ mode
743                  * it uses 0x0f000001.
744                  */
745                 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
746                 if (prop && (*prop & 0xff000000) == 0x0f000000)
747                         identify_cpu(0, *prop);
748
749                 identical_pvr_fixup(node);
750         }
751
752         check_cpu_feature_properties(node);
753         check_cpu_pa_features(node);
754         check_cpu_slb_size(node);
755
756 #ifdef CONFIG_PPC_PSERIES
757         if (nthreads > 1)
758                 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
759         else
760                 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
761 #endif
762
763         return 0;
764 }
765
766 #ifdef CONFIG_BLK_DEV_INITRD
767 static void __init early_init_dt_check_for_initrd(unsigned long node)
768 {
769         unsigned long l;
770         u32 *prop;
771
772         DBG("Looking for initrd properties... ");
773
774         prop = of_get_flat_dt_prop(node, "linux,initrd-start", &l);
775         if (prop) {
776                 initrd_start = (unsigned long)__va(of_read_ulong(prop, l/4));
777
778                 prop = of_get_flat_dt_prop(node, "linux,initrd-end", &l);
779                 if (prop) {
780                         initrd_end = (unsigned long)
781                                         __va(of_read_ulong(prop, l/4));
782                         initrd_below_start_ok = 1;
783                 } else {
784                         initrd_start = 0;
785                 }
786         }
787
788         DBG("initrd_start=0x%lx  initrd_end=0x%lx\n", initrd_start, initrd_end);
789 }
790 #else
791 static inline void early_init_dt_check_for_initrd(unsigned long node)
792 {
793 }
794 #endif /* CONFIG_BLK_DEV_INITRD */
795
796 static int __init early_init_dt_scan_chosen(unsigned long node,
797                                             const char *uname, int depth, void *data)
798 {
799         unsigned long *lprop;
800         unsigned long l;
801         char *p;
802
803         DBG("search \"chosen\", depth: %d, uname: %s\n", depth, uname);
804
805         if (depth != 1 ||
806             (strcmp(uname, "chosen") != 0 && strcmp(uname, "chosen@0") != 0))
807                 return 0;
808
809 #ifdef CONFIG_PPC64
810         /* check if iommu is forced on or off */
811         if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
812                 iommu_is_off = 1;
813         if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
814                 iommu_force_on = 1;
815 #endif
816
817         /* mem=x on the command line is the preferred mechanism */
818         lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
819         if (lprop)
820                 memory_limit = *lprop;
821
822 #ifdef CONFIG_PPC64
823         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
824         if (lprop)
825                 tce_alloc_start = *lprop;
826         lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
827         if (lprop)
828                 tce_alloc_end = *lprop;
829 #endif
830
831 #ifdef CONFIG_KEXEC
832         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
833         if (lprop)
834                 crashk_res.start = *lprop;
835
836         lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
837         if (lprop)
838                 crashk_res.end = crashk_res.start + *lprop - 1;
839 #endif
840
841         early_init_dt_check_for_initrd(node);
842
843         /* Retreive command line */
844         p = of_get_flat_dt_prop(node, "bootargs", &l);
845         if (p != NULL && l > 0)
846                 strlcpy(cmd_line, p, min((int)l, COMMAND_LINE_SIZE));
847
848 #ifdef CONFIG_CMDLINE
849         if (p == NULL || l == 0 || (l == 1 && (*p) == 0))
850                 strlcpy(cmd_line, CONFIG_CMDLINE, COMMAND_LINE_SIZE);
851 #endif /* CONFIG_CMDLINE */
852
853         DBG("Command line is: %s\n", cmd_line);
854
855         /* break now */
856         return 1;
857 }
858
859 static int __init early_init_dt_scan_root(unsigned long node,
860                                           const char *uname, int depth, void *data)
861 {
862         u32 *prop;
863
864         if (depth != 0)
865                 return 0;
866
867         prop = of_get_flat_dt_prop(node, "#size-cells", NULL);
868         dt_root_size_cells = (prop == NULL) ? 1 : *prop;
869         DBG("dt_root_size_cells = %x\n", dt_root_size_cells);
870
871         prop = of_get_flat_dt_prop(node, "#address-cells", NULL);
872         dt_root_addr_cells = (prop == NULL) ? 2 : *prop;
873         DBG("dt_root_addr_cells = %x\n", dt_root_addr_cells);
874         
875         /* break now */
876         return 1;
877 }
878
879 static u64 __init dt_mem_next_cell(int s, cell_t **cellp)
880 {
881         cell_t *p = *cellp;
882
883         *cellp = p + s;
884         return of_read_number(p, s);
885 }
886
887 #ifdef CONFIG_PPC_PSERIES
888 /*
889  * Interpret the ibm,dynamic-memory property in the
890  * /ibm,dynamic-reconfiguration-memory node.
891  * This contains a list of memory blocks along with NUMA affinity
892  * information.
893  */
894 static int __init early_init_dt_scan_drconf_memory(unsigned long node)
895 {
896         cell_t *dm, *ls, *usm;
897         unsigned long l, n, flags;
898         u64 base, size, lmb_size;
899         unsigned int is_kexec_kdump = 0, rngs;
900
901         ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
902         if (ls == NULL || l < dt_root_size_cells * sizeof(cell_t))
903                 return 0;
904         lmb_size = dt_mem_next_cell(dt_root_size_cells, &ls);
905
906         dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
907         if (dm == NULL || l < sizeof(cell_t))
908                 return 0;
909
910         n = *dm++;      /* number of entries */
911         if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(cell_t))
912                 return 0;
913
914         /* check if this is a kexec/kdump kernel. */
915         usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
916                                                  &l);
917         if (usm != NULL)
918                 is_kexec_kdump = 1;
919
920         for (; n != 0; --n) {
921                 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
922                 flags = dm[3];
923                 /* skip DRC index, pad, assoc. list index, flags */
924                 dm += 4;
925                 /* skip this block if the reserved bit is set in flags (0x80)
926                    or if the block is not assigned to this partition (0x8) */
927                 if ((flags & 0x80) || !(flags & 0x8))
928                         continue;
929                 size = lmb_size;
930                 rngs = 1;
931                 if (is_kexec_kdump) {
932                         /*
933                          * For each lmb in ibm,dynamic-memory, a corresponding
934                          * entry in linux,drconf-usable-memory property contains
935                          * a counter 'p' followed by 'p' (base, size) duple.
936                          * Now read the counter from
937                          * linux,drconf-usable-memory property
938                          */
939                         rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
940                         if (!rngs) /* there are no (base, size) duple */
941                                 continue;
942                 }
943                 do {
944                         if (is_kexec_kdump) {
945                                 base = dt_mem_next_cell(dt_root_addr_cells,
946                                                          &usm);
947                                 size = dt_mem_next_cell(dt_root_size_cells,
948                                                          &usm);
949                         }
950                         if (iommu_is_off) {
951                                 if (base >= 0x80000000ul)
952                                         continue;
953                                 if ((base + size) > 0x80000000ul)
954                                         size = 0x80000000ul - base;
955                         }
956                         lmb_add(base, size);
957                 } while (--rngs);
958         }
959         lmb_dump_all();
960         return 0;
961 }
962 #else
963 #define early_init_dt_scan_drconf_memory(node)  0
964 #endif /* CONFIG_PPC_PSERIES */
965
966 static int __init early_init_dt_scan_memory(unsigned long node,
967                                             const char *uname, int depth, void *data)
968 {
969         char *type = of_get_flat_dt_prop(node, "device_type", NULL);
970         cell_t *reg, *endp;
971         unsigned long l;
972
973         /* Look for the ibm,dynamic-reconfiguration-memory node */
974         if (depth == 1 &&
975             strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
976                 return early_init_dt_scan_drconf_memory(node);
977
978         /* We are scanning "memory" nodes only */
979         if (type == NULL) {
980                 /*
981                  * The longtrail doesn't have a device_type on the
982                  * /memory node, so look for the node called /memory@0.
983                  */
984                 if (depth != 1 || strcmp(uname, "memory@0") != 0)
985                         return 0;
986         } else if (strcmp(type, "memory") != 0)
987                 return 0;
988
989         reg = of_get_flat_dt_prop(node, "linux,usable-memory", &l);
990         if (reg == NULL)
991                 reg = of_get_flat_dt_prop(node, "reg", &l);
992         if (reg == NULL)
993                 return 0;
994
995         endp = reg + (l / sizeof(cell_t));
996
997         DBG("memory scan node %s, reg size %ld, data: %x %x %x %x,\n",
998             uname, l, reg[0], reg[1], reg[2], reg[3]);
999
1000         while ((endp - reg) >= (dt_root_addr_cells + dt_root_size_cells)) {
1001                 u64 base, size;
1002
1003                 base = dt_mem_next_cell(dt_root_addr_cells, &reg);
1004                 size = dt_mem_next_cell(dt_root_size_cells, &reg);
1005
1006                 if (size == 0)
1007                         continue;
1008                 DBG(" - %llx ,  %llx\n", (unsigned long long)base,
1009                     (unsigned long long)size);
1010 #ifdef CONFIG_PPC64
1011                 if (iommu_is_off) {
1012                         if (base >= 0x80000000ul)
1013                                 continue;
1014                         if ((base + size) > 0x80000000ul)
1015                                 size = 0x80000000ul - base;
1016                 }
1017 #endif
1018                 lmb_add(base, size);
1019
1020                 memstart_addr = min((u64)memstart_addr, base);
1021         }
1022
1023         return 0;
1024 }
1025
1026 static void __init early_reserve_mem(void)
1027 {
1028         u64 base, size;
1029         u64 *reserve_map;
1030         unsigned long self_base;
1031         unsigned long self_size;
1032
1033         reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
1034                                         initial_boot_params->off_mem_rsvmap);
1035
1036         /* before we do anything, lets reserve the dt blob */
1037         self_base = __pa((unsigned long)initial_boot_params);
1038         self_size = initial_boot_params->totalsize;
1039         lmb_reserve(self_base, self_size);
1040
1041 #ifdef CONFIG_BLK_DEV_INITRD
1042         /* then reserve the initrd, if any */
1043         if (initrd_start && (initrd_end > initrd_start))
1044                 lmb_reserve(__pa(initrd_start), initrd_end - initrd_start);
1045 #endif /* CONFIG_BLK_DEV_INITRD */
1046
1047 #ifdef CONFIG_PPC32
1048         /* 
1049          * Handle the case where we might be booting from an old kexec
1050          * image that setup the mem_rsvmap as pairs of 32-bit values
1051          */
1052         if (*reserve_map > 0xffffffffull) {
1053                 u32 base_32, size_32;
1054                 u32 *reserve_map_32 = (u32 *)reserve_map;
1055
1056                 while (1) {
1057                         base_32 = *(reserve_map_32++);
1058                         size_32 = *(reserve_map_32++);
1059                         if (size_32 == 0)
1060                                 break;
1061                         /* skip if the reservation is for the blob */
1062                         if (base_32 == self_base && size_32 == self_size)
1063                                 continue;
1064                         DBG("reserving: %x -> %x\n", base_32, size_32);
1065                         lmb_reserve(base_32, size_32);
1066                 }
1067                 return;
1068         }
1069 #endif
1070         while (1) {
1071                 base = *(reserve_map++);
1072                 size = *(reserve_map++);
1073                 if (size == 0)
1074                         break;
1075                 DBG("reserving: %llx -> %llx\n", base, size);
1076                 lmb_reserve(base, size);
1077         }
1078
1079 #if 0
1080         DBG("memory reserved, lmbs :\n");
1081         lmb_dump_all();
1082 #endif
1083 }
1084
1085 #ifdef CONFIG_PHYP_DUMP
1086 /**
1087  * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
1088  *
1089  * Function to find the largest size we need to reserve
1090  * during early boot process.
1091  *
1092  * It either looks for boot param and returns that OR
1093  * returns larger of 256 or 5% rounded down to multiples of 256MB.
1094  *
1095  */
1096 static inline unsigned long phyp_dump_calculate_reserve_size(void)
1097 {
1098         unsigned long tmp;
1099
1100         if (phyp_dump_info->reserve_bootvar)
1101                 return phyp_dump_info->reserve_bootvar;
1102
1103         /* divide by 20 to get 5% of value */
1104         tmp = lmb_end_of_DRAM();
1105         do_div(tmp, 20);
1106
1107         /* round it down in multiples of 256 */
1108         tmp = tmp & ~0x0FFFFFFFUL;
1109
1110         return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
1111 }
1112
1113 /**
1114  * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
1115  *
1116  * This routine may reserve memory regions in the kernel only
1117  * if the system is supported and a dump was taken in last
1118  * boot instance or if the hardware is supported and the
1119  * scratch area needs to be setup. In other instances it returns
1120  * without reserving anything. The memory in case of dump being
1121  * active is freed when the dump is collected (by userland tools).
1122  */
1123 static void __init phyp_dump_reserve_mem(void)
1124 {
1125         unsigned long base, size;
1126         unsigned long variable_reserve_size;
1127
1128         if (!phyp_dump_info->phyp_dump_configured) {
1129                 printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
1130                 return;
1131         }
1132
1133         if (!phyp_dump_info->phyp_dump_at_boot) {
1134                 printk(KERN_INFO "Phyp-dump disabled at boot time\n");
1135                 return;
1136         }
1137
1138         variable_reserve_size = phyp_dump_calculate_reserve_size();
1139
1140         if (phyp_dump_info->phyp_dump_is_active) {
1141                 /* Reserve *everything* above RMR.Area freed by userland tools*/
1142                 base = variable_reserve_size;
1143                 size = lmb_end_of_DRAM() - base;
1144
1145                 /* XXX crashed_ram_end is wrong, since it may be beyond
1146                  * the memory_limit, it will need to be adjusted. */
1147                 lmb_reserve(base, size);
1148
1149                 phyp_dump_info->init_reserve_start = base;
1150                 phyp_dump_info->init_reserve_size = size;
1151         } else {
1152                 size = phyp_dump_info->cpu_state_size +
1153                         phyp_dump_info->hpte_region_size +
1154                         variable_reserve_size;
1155                 base = lmb_end_of_DRAM() - size;
1156                 lmb_reserve(base, size);
1157                 phyp_dump_info->init_reserve_start = base;
1158                 phyp_dump_info->init_reserve_size = size;
1159         }
1160 }
1161 #else
1162 static inline void __init phyp_dump_reserve_mem(void) {}
1163 #endif /* CONFIG_PHYP_DUMP  && CONFIG_PPC_RTAS */
1164
1165
1166 void __init early_init_devtree(void *params)
1167 {
1168         unsigned long limit;
1169
1170         DBG(" -> early_init_devtree(%p)\n", params);
1171
1172         /* Setup flat device-tree pointer */
1173         initial_boot_params = params;
1174
1175 #ifdef CONFIG_PPC_RTAS
1176         /* Some machines might need RTAS info for debugging, grab it now. */
1177         of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
1178 #endif
1179
1180 #ifdef CONFIG_PHYP_DUMP
1181         /* scan tree to see if dump occured during last boot */
1182         of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
1183 #endif
1184
1185         /* Retrieve various informations from the /chosen node of the
1186          * device-tree, including the platform type, initrd location and
1187          * size, TCE reserve, and more ...
1188          */
1189         of_scan_flat_dt(early_init_dt_scan_chosen, NULL);
1190
1191         /* Scan memory nodes and rebuild LMBs */
1192         lmb_init();
1193         of_scan_flat_dt(early_init_dt_scan_root, NULL);
1194         of_scan_flat_dt(early_init_dt_scan_memory, NULL);
1195
1196         /* Save command line for /proc/cmdline and then parse parameters */
1197         strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
1198         parse_early_param();
1199
1200         /* Reserve LMB regions used by kernel, initrd, dt, etc... */
1201         lmb_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
1202         /* If relocatable, reserve first 32k for interrupt vectors etc. */
1203         if (PHYSICAL_START > MEMORY_START)
1204                 lmb_reserve(MEMORY_START, 0x8000);
1205         reserve_kdump_trampoline();
1206         reserve_crashkernel();
1207         early_reserve_mem();
1208         phyp_dump_reserve_mem();
1209
1210         limit = memory_limit;
1211         if (! limit) {
1212                 unsigned long memsize;
1213
1214                 /* Ensure that total memory size is page-aligned, because
1215                  * otherwise mark_bootmem() gets upset. */
1216                 lmb_analyze();
1217                 memsize = lmb_phys_mem_size();
1218                 if ((memsize & PAGE_MASK) != memsize)
1219                         limit = memsize & PAGE_MASK;
1220         }
1221         lmb_enforce_memory_limit(limit);
1222
1223         lmb_analyze();
1224
1225         DBG("Phys. mem: %lx\n", lmb_phys_mem_size());
1226
1227         /* We may need to relocate the flat tree, do it now.
1228          * FIXME .. and the initrd too? */
1229         move_device_tree();
1230
1231         DBG("Scanning CPUs ...\n");
1232
1233         /* Retreive CPU related informations from the flat tree
1234          * (altivec support, boot CPU ID, ...)
1235          */
1236         of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
1237
1238         DBG(" <- early_init_devtree()\n");
1239 }
1240
1241
1242 /**
1243  * Indicates whether the root node has a given value in its
1244  * compatible property.
1245  */
1246 int machine_is_compatible(const char *compat)
1247 {
1248         struct device_node *root;
1249         int rc = 0;
1250
1251         root = of_find_node_by_path("/");
1252         if (root) {
1253                 rc = of_device_is_compatible(root, compat);
1254                 of_node_put(root);
1255         }
1256         return rc;
1257 }
1258 EXPORT_SYMBOL(machine_is_compatible);
1259
1260 /*******
1261  *
1262  * New implementation of the OF "find" APIs, return a refcounted
1263  * object, call of_node_put() when done.  The device tree and list
1264  * are protected by a rw_lock.
1265  *
1266  * Note that property management will need some locking as well,
1267  * this isn't dealt with yet.
1268  *
1269  *******/
1270
1271 /**
1272  *      of_find_node_by_phandle - Find a node given a phandle
1273  *      @handle:        phandle of the node to find
1274  *
1275  *      Returns a node pointer with refcount incremented, use
1276  *      of_node_put() on it when done.
1277  */
1278 struct device_node *of_find_node_by_phandle(phandle handle)
1279 {
1280         struct device_node *np;
1281
1282         read_lock(&devtree_lock);
1283         for (np = allnodes; np != 0; np = np->allnext)
1284                 if (np->linux_phandle == handle)
1285                         break;
1286         of_node_get(np);
1287         read_unlock(&devtree_lock);
1288         return np;
1289 }
1290 EXPORT_SYMBOL(of_find_node_by_phandle);
1291
1292 /**
1293  *      of_find_next_cache_node - Find a node's subsidiary cache
1294  *      @np:    node of type "cpu" or "cache"
1295  *
1296  *      Returns a node pointer with refcount incremented, use
1297  *      of_node_put() on it when done.  Caller should hold a reference
1298  *      to np.
1299  */
1300 struct device_node *of_find_next_cache_node(struct device_node *np)
1301 {
1302         struct device_node *child;
1303         const phandle *handle;
1304
1305         handle = of_get_property(np, "l2-cache", NULL);
1306         if (!handle)
1307                 handle = of_get_property(np, "next-level-cache", NULL);
1308
1309         if (handle)
1310                 return of_find_node_by_phandle(*handle);
1311
1312         /* OF on pmac has nodes instead of properties named "l2-cache"
1313          * beneath CPU nodes.
1314          */
1315         if (!strcmp(np->type, "cpu"))
1316                 for_each_child_of_node(np, child)
1317                         if (!strcmp(child->type, "cache"))
1318                                 return child;
1319
1320         return NULL;
1321 }
1322
1323 /**
1324  *      of_find_all_nodes - Get next node in global list
1325  *      @prev:  Previous node or NULL to start iteration
1326  *              of_node_put() will be called on it
1327  *
1328  *      Returns a node pointer with refcount incremented, use
1329  *      of_node_put() on it when done.
1330  */
1331 struct device_node *of_find_all_nodes(struct device_node *prev)
1332 {
1333         struct device_node *np;
1334
1335         read_lock(&devtree_lock);
1336         np = prev ? prev->allnext : allnodes;
1337         for (; np != 0; np = np->allnext)
1338                 if (of_node_get(np))
1339                         break;
1340         of_node_put(prev);
1341         read_unlock(&devtree_lock);
1342         return np;
1343 }
1344 EXPORT_SYMBOL(of_find_all_nodes);
1345
1346 /**
1347  *      of_node_get - Increment refcount of a node
1348  *      @node:  Node to inc refcount, NULL is supported to
1349  *              simplify writing of callers
1350  *
1351  *      Returns node.
1352  */
1353 struct device_node *of_node_get(struct device_node *node)
1354 {
1355         if (node)
1356                 kref_get(&node->kref);
1357         return node;
1358 }
1359 EXPORT_SYMBOL(of_node_get);
1360
1361 static inline struct device_node * kref_to_device_node(struct kref *kref)
1362 {
1363         return container_of(kref, struct device_node, kref);
1364 }
1365
1366 /**
1367  *      of_node_release - release a dynamically allocated node
1368  *      @kref:  kref element of the node to be released
1369  *
1370  *      In of_node_put() this function is passed to kref_put()
1371  *      as the destructor.
1372  */
1373 static void of_node_release(struct kref *kref)
1374 {
1375         struct device_node *node = kref_to_device_node(kref);
1376         struct property *prop = node->properties;
1377
1378         /* We should never be releasing nodes that haven't been detached. */
1379         if (!of_node_check_flag(node, OF_DETACHED)) {
1380                 printk("WARNING: Bad of_node_put() on %s\n", node->full_name);
1381                 dump_stack();
1382                 kref_init(&node->kref);
1383                 return;
1384         }
1385
1386         if (!of_node_check_flag(node, OF_DYNAMIC))
1387                 return;
1388
1389         while (prop) {
1390                 struct property *next = prop->next;
1391                 kfree(prop->name);
1392                 kfree(prop->value);
1393                 kfree(prop);
1394                 prop = next;
1395
1396                 if (!prop) {
1397                         prop = node->deadprops;
1398                         node->deadprops = NULL;
1399                 }
1400         }
1401         kfree(node->full_name);
1402         kfree(node->data);
1403         kfree(node);
1404 }
1405
1406 /**
1407  *      of_node_put - Decrement refcount of a node
1408  *      @node:  Node to dec refcount, NULL is supported to
1409  *              simplify writing of callers
1410  *
1411  */
1412 void of_node_put(struct device_node *node)
1413 {
1414         if (node)
1415                 kref_put(&node->kref, of_node_release);
1416 }
1417 EXPORT_SYMBOL(of_node_put);
1418
1419 /*
1420  * Plug a device node into the tree and global list.
1421  */
1422 void of_attach_node(struct device_node *np)
1423 {
1424         unsigned long flags;
1425
1426         write_lock_irqsave(&devtree_lock, flags);
1427         np->sibling = np->parent->child;
1428         np->allnext = allnodes;
1429         np->parent->child = np;
1430         allnodes = np;
1431         write_unlock_irqrestore(&devtree_lock, flags);
1432 }
1433
1434 /*
1435  * "Unplug" a node from the device tree.  The caller must hold
1436  * a reference to the node.  The memory associated with the node
1437  * is not freed until its refcount goes to zero.
1438  */
1439 void of_detach_node(struct device_node *np)
1440 {
1441         struct device_node *parent;
1442         unsigned long flags;
1443
1444         write_lock_irqsave(&devtree_lock, flags);
1445
1446         parent = np->parent;
1447         if (!parent)
1448                 goto out_unlock;
1449
1450         if (allnodes == np)
1451                 allnodes = np->allnext;
1452         else {
1453                 struct device_node *prev;
1454                 for (prev = allnodes;
1455                      prev->allnext != np;
1456                      prev = prev->allnext)
1457                         ;
1458                 prev->allnext = np->allnext;
1459         }
1460
1461         if (parent->child == np)
1462                 parent->child = np->sibling;
1463         else {
1464                 struct device_node *prevsib;
1465                 for (prevsib = np->parent->child;
1466                      prevsib->sibling != np;
1467                      prevsib = prevsib->sibling)
1468                         ;
1469                 prevsib->sibling = np->sibling;
1470         }
1471
1472         of_node_set_flag(np, OF_DETACHED);
1473
1474 out_unlock:
1475         write_unlock_irqrestore(&devtree_lock, flags);
1476 }
1477
1478 #ifdef CONFIG_PPC_PSERIES
1479 /*
1480  * Fix up the uninitialized fields in a new device node:
1481  * name, type and pci-specific fields
1482  */
1483
1484 static int of_finish_dynamic_node(struct device_node *node)
1485 {
1486         struct device_node *parent = of_get_parent(node);
1487         int err = 0;
1488         const phandle *ibm_phandle;
1489
1490         node->name = of_get_property(node, "name", NULL);
1491         node->type = of_get_property(node, "device_type", NULL);
1492
1493         if (!node->name)
1494                 node->name = "<NULL>";
1495         if (!node->type)
1496                 node->type = "<NULL>";
1497
1498         if (!parent) {
1499                 err = -ENODEV;
1500                 goto out;
1501         }
1502
1503         /* We don't support that function on PowerMac, at least
1504          * not yet
1505          */
1506         if (machine_is(powermac))
1507                 return -ENODEV;
1508
1509         /* fix up new node's linux_phandle field */
1510         if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
1511                 node->linux_phandle = *ibm_phandle;
1512
1513 out:
1514         of_node_put(parent);
1515         return err;
1516 }
1517
1518 static int prom_reconfig_notifier(struct notifier_block *nb,
1519                                   unsigned long action, void *node)
1520 {
1521         int err;
1522
1523         switch (action) {
1524         case PSERIES_RECONFIG_ADD:
1525                 err = of_finish_dynamic_node(node);
1526                 if (err < 0) {
1527                         printk(KERN_ERR "finish_node returned %d\n", err);
1528                         err = NOTIFY_BAD;
1529                 }
1530                 break;
1531         default:
1532                 err = NOTIFY_DONE;
1533                 break;
1534         }
1535         return err;
1536 }
1537
1538 static struct notifier_block prom_reconfig_nb = {
1539         .notifier_call = prom_reconfig_notifier,
1540         .priority = 10, /* This one needs to run first */
1541 };
1542
1543 static int __init prom_reconfig_setup(void)
1544 {
1545         return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
1546 }
1547 __initcall(prom_reconfig_setup);
1548 #endif
1549
1550 /*
1551  * Add a property to a node
1552  */
1553 int prom_add_property(struct device_node* np, struct property* prop)
1554 {
1555         struct property **next;
1556         unsigned long flags;
1557
1558         prop->next = NULL;      
1559         write_lock_irqsave(&devtree_lock, flags);
1560         next = &np->properties;
1561         while (*next) {
1562                 if (strcmp(prop->name, (*next)->name) == 0) {
1563                         /* duplicate ! don't insert it */
1564                         write_unlock_irqrestore(&devtree_lock, flags);
1565                         return -1;
1566                 }
1567                 next = &(*next)->next;
1568         }
1569         *next = prop;
1570         write_unlock_irqrestore(&devtree_lock, flags);
1571
1572 #ifdef CONFIG_PROC_DEVICETREE
1573         /* try to add to proc as well if it was initialized */
1574         if (np->pde)
1575                 proc_device_tree_add_prop(np->pde, prop);
1576 #endif /* CONFIG_PROC_DEVICETREE */
1577
1578         return 0;
1579 }
1580
1581 /*
1582  * Remove a property from a node.  Note that we don't actually
1583  * remove it, since we have given out who-knows-how-many pointers
1584  * to the data using get-property.  Instead we just move the property
1585  * to the "dead properties" list, so it won't be found any more.
1586  */
1587 int prom_remove_property(struct device_node *np, struct property *prop)
1588 {
1589         struct property **next;
1590         unsigned long flags;
1591         int found = 0;
1592
1593         write_lock_irqsave(&devtree_lock, flags);
1594         next = &np->properties;
1595         while (*next) {
1596                 if (*next == prop) {
1597                         /* found the node */
1598                         *next = prop->next;
1599                         prop->next = np->deadprops;
1600                         np->deadprops = prop;
1601                         found = 1;
1602                         break;
1603                 }
1604                 next = &(*next)->next;
1605         }
1606         write_unlock_irqrestore(&devtree_lock, flags);
1607
1608         if (!found)
1609                 return -ENODEV;
1610
1611 #ifdef CONFIG_PROC_DEVICETREE
1612         /* try to remove the proc node as well */
1613         if (np->pde)
1614                 proc_device_tree_remove_prop(np->pde, prop);
1615 #endif /* CONFIG_PROC_DEVICETREE */
1616
1617         return 0;
1618 }
1619
1620 /*
1621  * Update a property in a node.  Note that we don't actually
1622  * remove it, since we have given out who-knows-how-many pointers
1623  * to the data using get-property.  Instead we just move the property
1624  * to the "dead properties" list, and add the new property to the
1625  * property list
1626  */
1627 int prom_update_property(struct device_node *np,
1628                          struct property *newprop,
1629                          struct property *oldprop)
1630 {
1631         struct property **next;
1632         unsigned long flags;
1633         int found = 0;
1634
1635         write_lock_irqsave(&devtree_lock, flags);
1636         next = &np->properties;
1637         while (*next) {
1638                 if (*next == oldprop) {
1639                         /* found the node */
1640                         newprop->next = oldprop->next;
1641                         *next = newprop;
1642                         oldprop->next = np->deadprops;
1643                         np->deadprops = oldprop;
1644                         found = 1;
1645                         break;
1646                 }
1647                 next = &(*next)->next;
1648         }
1649         write_unlock_irqrestore(&devtree_lock, flags);
1650
1651         if (!found)
1652                 return -ENODEV;
1653
1654 #ifdef CONFIG_PROC_DEVICETREE
1655         /* try to add to proc as well if it was initialized */
1656         if (np->pde)
1657                 proc_device_tree_update_prop(np->pde, newprop, oldprop);
1658 #endif /* CONFIG_PROC_DEVICETREE */
1659
1660         return 0;
1661 }
1662
1663
1664 /* Find the device node for a given logical cpu number, also returns the cpu
1665  * local thread number (index in ibm,interrupt-server#s) if relevant and
1666  * asked for (non NULL)
1667  */
1668 struct device_node *of_get_cpu_node(int cpu, unsigned int *thread)
1669 {
1670         int hardid;
1671         struct device_node *np;
1672
1673         hardid = get_hard_smp_processor_id(cpu);
1674
1675         for_each_node_by_type(np, "cpu") {
1676                 const u32 *intserv;
1677                 unsigned int plen, t;
1678
1679                 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
1680                  * fallback to "reg" property and assume no threads
1681                  */
1682                 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
1683                                 &plen);
1684                 if (intserv == NULL) {
1685                         const u32 *reg = of_get_property(np, "reg", NULL);
1686                         if (reg == NULL)
1687                                 continue;
1688                         if (*reg == hardid) {
1689                                 if (thread)
1690                                         *thread = 0;
1691                                 return np;
1692                         }
1693                 } else {
1694                         plen /= sizeof(u32);
1695                         for (t = 0; t < plen; t++) {
1696                                 if (hardid == intserv[t]) {
1697                                         if (thread)
1698                                                 *thread = t;
1699                                         return np;
1700                                 }
1701                         }
1702                 }
1703         }
1704         return NULL;
1705 }
1706 EXPORT_SYMBOL(of_get_cpu_node);
1707
1708 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
1709 static struct debugfs_blob_wrapper flat_dt_blob;
1710
1711 static int __init export_flat_device_tree(void)
1712 {
1713         struct dentry *d;
1714
1715         flat_dt_blob.data = initial_boot_params;
1716         flat_dt_blob.size = initial_boot_params->totalsize;
1717
1718         d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
1719                                 powerpc_debugfs_root, &flat_dt_blob);
1720         if (!d)
1721                 return 1;
1722
1723         return 0;
1724 }
1725 __initcall(export_flat_device_tree);
1726 #endif