2 * Procedures for creating, accessing and interpreting the device tree.
4 * Paul Mackerras August 1996.
5 * Copyright (C) 1996-2005 Paul Mackerras.
7 * Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
8 * {engebret|bergner}@us.ibm.com
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.
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>
39 #include <asm/processor.h>
42 #include <asm/kdump.h>
44 #include <asm/system.h>
46 #include <asm/pgtable.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>
59 #define DBG(fmt...) printk(KERN_ERR fmt)
65 static int __initdata dt_root_addr_cells
;
66 static int __initdata dt_root_size_cells
;
69 int __initdata iommu_is_off
;
70 int __initdata iommu_force_on
;
71 unsigned long tce_alloc_start
, tce_alloc_end
;
76 extern struct device_node
*allnodes
; /* temporary while merging */
78 extern rwlock_t devtree_lock
; /* temporary while merging */
80 /* export that to outside world */
81 struct device_node
*of_chosen
;
83 int __init
of_flat_dt_is_compatible(unsigned long node
, const char *compat
)
86 unsigned long cplen
, l
;
88 cp
= of_get_flat_dt_prop(node
, "compatible", &cplen
);
92 if (strncasecmp(cp
, compat
, strlen(compat
)) == 0)
102 static void *__init
unflatten_dt_alloc(unsigned long *mem
, unsigned long size
,
107 *mem
= _ALIGN(*mem
, align
);
114 static unsigned long __init
unflatten_dt_node(unsigned long mem
,
116 struct device_node
*dad
,
117 struct device_node
***allnextpp
,
118 unsigned long fpsize
)
120 struct device_node
*np
;
121 struct property
*pp
, **prev_pp
= NULL
;
124 unsigned int l
, allocl
;
128 tag
= *((u32
*)(*p
));
129 if (tag
!= OF_DT_BEGIN_NODE
) {
130 printk("Weird tag at start of node: %x\n", tag
);
135 l
= allocl
= strlen(pathp
) + 1;
136 *p
= _ALIGN(*p
+ l
, 4);
138 /* version 0x10 has a more compact unit name here instead of the full
139 * path. we accumulate the full path size using "fpsize", we'll rebuild
140 * it later. We detect this because the first character of the name is
143 if ((*pathp
) != '/') {
146 /* root node: special case. fpsize accounts for path
147 * plus terminating zero. root node only has '/', so
148 * fpsize should be 2, but we want to avoid the first
149 * level nodes to have two '/' so we use fpsize 1 here
154 /* account for '/' and path size minus terminal 0
163 np
= unflatten_dt_alloc(&mem
, sizeof(struct device_node
) + allocl
,
164 __alignof__(struct device_node
));
166 memset(np
, 0, sizeof(*np
));
167 np
->full_name
= ((char*)np
) + sizeof(struct device_node
);
169 char *p
= np
->full_name
;
170 /* rebuild full path for new format */
171 if (dad
&& dad
->parent
) {
172 strcpy(p
, dad
->full_name
);
174 if ((strlen(p
) + l
+ 1) != allocl
) {
175 DBG("%s: p: %d, l: %d, a: %d\n",
176 pathp
, (int)strlen(p
), l
, allocl
);
184 memcpy(np
->full_name
, pathp
, l
);
185 prev_pp
= &np
->properties
;
187 *allnextpp
= &np
->allnext
;
190 /* we temporarily use the next field as `last_child'*/
194 dad
->next
->sibling
= np
;
197 kref_init(&np
->kref
);
203 tag
= *((u32
*)(*p
));
204 if (tag
== OF_DT_NOP
) {
208 if (tag
!= OF_DT_PROP
)
212 noff
= *((u32
*)((*p
) + 4));
214 if (initial_boot_params
->version
< 0x10)
215 *p
= _ALIGN(*p
, sz
>= 8 ? 8 : 4);
217 pname
= find_flat_dt_string(noff
);
219 printk("Can't find property name in list !\n");
222 if (strcmp(pname
, "name") == 0)
224 l
= strlen(pname
) + 1;
225 pp
= unflatten_dt_alloc(&mem
, sizeof(struct property
),
226 __alignof__(struct property
));
228 if (strcmp(pname
, "linux,phandle") == 0) {
229 np
->node
= *((u32
*)*p
);
230 if (np
->linux_phandle
== 0)
231 np
->linux_phandle
= np
->node
;
233 if (strcmp(pname
, "ibm,phandle") == 0)
234 np
->linux_phandle
= *((u32
*)*p
);
237 pp
->value
= (void *)*p
;
241 *p
= _ALIGN((*p
) + sz
, 4);
243 /* with version 0x10 we may not have the name property, recreate
244 * it here from the unit name if absent
247 char *p
= pathp
, *ps
= pathp
, *pa
= NULL
;
260 pp
= unflatten_dt_alloc(&mem
, sizeof(struct property
) + sz
,
261 __alignof__(struct property
));
268 memcpy(pp
->value
, ps
, sz
- 1);
269 ((char *)pp
->value
)[sz
- 1] = 0;
270 DBG("fixed up name for %s -> %s\n", pathp
,
276 np
->name
= of_get_property(np
, "name", NULL
);
277 np
->type
= of_get_property(np
, "device_type", NULL
);
284 while (tag
== OF_DT_BEGIN_NODE
) {
285 mem
= unflatten_dt_node(mem
, p
, np
, allnextpp
, fpsize
);
286 tag
= *((u32
*)(*p
));
288 if (tag
!= OF_DT_END_NODE
) {
289 printk("Weird tag at end of node: %x\n", tag
);
296 static int __init
early_parse_mem(char *p
)
301 memory_limit
= PAGE_ALIGN(memparse(p
, &p
));
302 DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit
);
306 early_param("mem", early_parse_mem
);
309 * move_device_tree - move tree to an unused area, if needed.
311 * The device tree may be allocated beyond our memory limit, or inside the
312 * crash kernel region for kdump. If so, move it out of the way.
314 static void __init
move_device_tree(void)
316 unsigned long start
, size
;
319 DBG("-> move_device_tree\n");
321 start
= __pa(initial_boot_params
);
322 size
= initial_boot_params
->totalsize
;
324 if ((memory_limit
&& (start
+ size
) > memory_limit
) ||
325 overlaps_crashkernel(start
, size
)) {
326 p
= __va(lmb_alloc_base(size
, PAGE_SIZE
, lmb
.rmo_size
));
327 memcpy(p
, initial_boot_params
, size
);
328 initial_boot_params
= (struct boot_param_header
*)p
;
329 DBG("Moved device tree to 0x%p\n", p
);
332 DBG("<- move_device_tree\n");
336 * unflattens the device-tree passed by the firmware, creating the
337 * tree of struct device_node. It also fills the "name" and "type"
338 * pointers of the nodes so the normal device-tree walking functions
339 * can be used (this used to be done by finish_device_tree)
341 void __init
unflatten_device_tree(void)
343 unsigned long start
, mem
, size
;
344 struct device_node
**allnextp
= &allnodes
;
346 DBG(" -> unflatten_device_tree()\n");
348 /* First pass, scan for size */
349 start
= ((unsigned long)initial_boot_params
) +
350 initial_boot_params
->off_dt_struct
;
351 size
= unflatten_dt_node(0, &start
, NULL
, NULL
, 0);
352 size
= (size
| 3) + 1;
354 DBG(" size is %lx, allocating...\n", size
);
356 /* Allocate memory for the expanded device tree */
357 mem
= lmb_alloc(size
+ 4, __alignof__(struct device_node
));
358 mem
= (unsigned long) __va(mem
);
360 ((u32
*)mem
)[size
/ 4] = 0xdeadbeef;
362 DBG(" unflattening %lx...\n", mem
);
364 /* Second pass, do actual unflattening */
365 start
= ((unsigned long)initial_boot_params
) +
366 initial_boot_params
->off_dt_struct
;
367 unflatten_dt_node(mem
, &start
, NULL
, &allnextp
, 0);
368 if (*((u32
*)start
) != OF_DT_END
)
369 printk(KERN_WARNING
"Weird tag at end of tree: %08x\n", *((u32
*)start
));
370 if (((u32
*)mem
)[size
/ 4] != 0xdeadbeef)
371 printk(KERN_WARNING
"End of tree marker overwritten: %08x\n",
372 ((u32
*)mem
)[size
/ 4] );
375 /* Get pointer to OF "/chosen" node for use everywhere */
376 of_chosen
= of_find_node_by_path("/chosen");
377 if (of_chosen
== NULL
)
378 of_chosen
= of_find_node_by_path("/chosen@0");
380 DBG(" <- unflatten_device_tree()\n");
384 * ibm,pa-features is a per-cpu property that contains a string of
385 * attribute descriptors, each of which has a 2 byte header plus up
386 * to 254 bytes worth of processor attribute bits. First header
387 * byte specifies the number of bytes following the header.
388 * Second header byte is an "attribute-specifier" type, of which
389 * zero is the only currently-defined value.
390 * Implementation: Pass in the byte and bit offset for the feature
391 * that we are interested in. The function will return -1 if the
392 * pa-features property is missing, or a 1/0 to indicate if the feature
393 * is supported/not supported. Note that the bit numbers are
394 * big-endian to match the definition in PAPR.
396 static struct ibm_pa_feature
{
397 unsigned long cpu_features
; /* CPU_FTR_xxx bit */
398 unsigned int cpu_user_ftrs
; /* PPC_FEATURE_xxx bit */
399 unsigned char pabyte
; /* byte number in ibm,pa-features */
400 unsigned char pabit
; /* bit number (big-endian) */
401 unsigned char invert
; /* if 1, pa bit set => clear feature */
402 } ibm_pa_features
[] __initdata
= {
403 {0, PPC_FEATURE_HAS_MMU
, 0, 0, 0},
404 {0, PPC_FEATURE_HAS_FPU
, 0, 1, 0},
405 {CPU_FTR_SLB
, 0, 0, 2, 0},
406 {CPU_FTR_CTRL
, 0, 0, 3, 0},
407 {CPU_FTR_NOEXECUTE
, 0, 0, 6, 0},
408 {CPU_FTR_NODSISRALIGN
, 0, 1, 1, 1},
409 {CPU_FTR_CI_LARGE_PAGE
, 0, 1, 2, 0},
410 {CPU_FTR_REAL_LE
, PPC_FEATURE_TRUE_LE
, 5, 0, 0},
413 static void __init
scan_features(unsigned long node
, unsigned char *ftrs
,
414 unsigned long tablelen
,
415 struct ibm_pa_feature
*fp
,
416 unsigned long ft_size
)
418 unsigned long i
, len
, bit
;
420 /* find descriptor with type == 0 */
426 return; /* descriptor 0 not found */
433 /* loop over bits we know about */
434 for (i
= 0; i
< ft_size
; ++i
, ++fp
) {
435 if (fp
->pabyte
>= ftrs
[0])
437 bit
= (ftrs
[2 + fp
->pabyte
] >> (7 - fp
->pabit
)) & 1;
438 if (bit
^ fp
->invert
) {
439 cur_cpu_spec
->cpu_features
|= fp
->cpu_features
;
440 cur_cpu_spec
->cpu_user_features
|= fp
->cpu_user_ftrs
;
442 cur_cpu_spec
->cpu_features
&= ~fp
->cpu_features
;
443 cur_cpu_spec
->cpu_user_features
&= ~fp
->cpu_user_ftrs
;
448 static void __init
check_cpu_pa_features(unsigned long node
)
450 unsigned char *pa_ftrs
;
451 unsigned long tablelen
;
453 pa_ftrs
= of_get_flat_dt_prop(node
, "ibm,pa-features", &tablelen
);
457 scan_features(node
, pa_ftrs
, tablelen
,
458 ibm_pa_features
, ARRAY_SIZE(ibm_pa_features
));
461 #ifdef CONFIG_PPC_STD_MMU_64
462 static void __init
check_cpu_slb_size(unsigned long node
)
466 slb_size_ptr
= of_get_flat_dt_prop(node
, "slb-size", NULL
);
467 if (slb_size_ptr
!= NULL
) {
468 mmu_slb_size
= *slb_size_ptr
;
471 slb_size_ptr
= of_get_flat_dt_prop(node
, "ibm,slb-size", NULL
);
472 if (slb_size_ptr
!= NULL
) {
473 mmu_slb_size
= *slb_size_ptr
;
477 #define check_cpu_slb_size(node) do { } while(0)
480 static struct feature_property
{
483 unsigned long cpu_feature
;
484 unsigned long cpu_user_ftr
;
485 } feature_properties
[] __initdata
= {
486 #ifdef CONFIG_ALTIVEC
487 {"altivec", 0, CPU_FTR_ALTIVEC
, PPC_FEATURE_HAS_ALTIVEC
},
488 {"ibm,vmx", 1, CPU_FTR_ALTIVEC
, PPC_FEATURE_HAS_ALTIVEC
},
489 #endif /* CONFIG_ALTIVEC */
491 /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
492 {"ibm,vmx", 2, CPU_FTR_VSX
, PPC_FEATURE_HAS_VSX
},
493 #endif /* CONFIG_VSX */
495 {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP
},
496 {"ibm,purr", 1, CPU_FTR_PURR
, 0},
497 {"ibm,spurr", 1, CPU_FTR_SPURR
, 0},
498 #endif /* CONFIG_PPC64 */
501 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
502 static inline void identical_pvr_fixup(unsigned long node
)
505 char *model
= of_get_flat_dt_prop(node
, "model", NULL
);
508 * Since 440GR(x)/440EP(x) processors have the same pvr,
509 * we check the node path and set bit 28 in the cur_cpu_spec
510 * pvr for EP(x) processor version. This bit is always 0 in
511 * the "real" pvr. Then we call identify_cpu again with
512 * the new logical pvr to enable FPU support.
514 if (model
&& strstr(model
, "440EP")) {
515 pvr
= cur_cpu_spec
->pvr_value
| 0x8;
516 identify_cpu(0, pvr
);
517 DBG("Using logical pvr %x for %s\n", pvr
, model
);
521 #define identical_pvr_fixup(node) do { } while(0)
524 static void __init
check_cpu_feature_properties(unsigned long node
)
527 struct feature_property
*fp
= feature_properties
;
530 for (i
= 0; i
< ARRAY_SIZE(feature_properties
); ++i
, ++fp
) {
531 prop
= of_get_flat_dt_prop(node
, fp
->name
, NULL
);
532 if (prop
&& *prop
>= fp
->min_value
) {
533 cur_cpu_spec
->cpu_features
|= fp
->cpu_feature
;
534 cur_cpu_spec
->cpu_user_features
|= fp
->cpu_user_ftr
;
539 static int __init
early_init_dt_scan_cpus(unsigned long node
,
540 const char *uname
, int depth
,
543 static int logical_cpuid
= 0;
544 char *type
= of_get_flat_dt_prop(node
, "device_type", NULL
);
551 /* We are scanning "cpu" nodes only */
552 if (type
== NULL
|| strcmp(type
, "cpu") != 0)
555 /* Get physical cpuid */
556 intserv
= of_get_flat_dt_prop(node
, "ibm,ppc-interrupt-server#s", &len
);
558 nthreads
= len
/ sizeof(int);
560 intserv
= of_get_flat_dt_prop(node
, "reg", NULL
);
565 * Now see if any of these threads match our boot cpu.
566 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
568 for (i
= 0; i
< nthreads
; i
++) {
570 * version 2 of the kexec param format adds the phys cpuid of
573 if (initial_boot_params
&& initial_boot_params
->version
>= 2) {
575 initial_boot_params
->boot_cpuid_phys
) {
581 * Check if it's the boot-cpu, set it's hw index now,
582 * unfortunately this format did not support booting
583 * off secondary threads.
585 if (of_get_flat_dt_prop(node
,
586 "linux,boot-cpu", NULL
) != NULL
) {
593 /* logical cpu id is always 0 on UP kernels */
599 DBG("boot cpu: logical %d physical %d\n", logical_cpuid
,
601 boot_cpuid
= logical_cpuid
;
602 set_hard_smp_processor_id(boot_cpuid
, intserv
[i
]);
605 * PAPR defines "logical" PVR values for cpus that
606 * meet various levels of the architecture:
607 * 0x0f000001 Architecture version 2.04
608 * 0x0f000002 Architecture version 2.05
609 * If the cpu-version property in the cpu node contains
610 * such a value, we call identify_cpu again with the
611 * logical PVR value in order to use the cpu feature
612 * bits appropriate for the architecture level.
614 * A POWER6 partition in "POWER6 architected" mode
615 * uses the 0x0f000002 PVR value; in POWER5+ mode
616 * it uses 0x0f000001.
618 prop
= of_get_flat_dt_prop(node
, "cpu-version", NULL
);
619 if (prop
&& (*prop
& 0xff000000) == 0x0f000000)
620 identify_cpu(0, *prop
);
622 identical_pvr_fixup(node
);
625 check_cpu_feature_properties(node
);
626 check_cpu_pa_features(node
);
627 check_cpu_slb_size(node
);
629 #ifdef CONFIG_PPC_PSERIES
631 cur_cpu_spec
->cpu_features
|= CPU_FTR_SMT
;
633 cur_cpu_spec
->cpu_features
&= ~CPU_FTR_SMT
;
639 #ifdef CONFIG_BLK_DEV_INITRD
640 static void __init
early_init_dt_check_for_initrd(unsigned long node
)
645 DBG("Looking for initrd properties... ");
647 prop
= of_get_flat_dt_prop(node
, "linux,initrd-start", &l
);
649 initrd_start
= (unsigned long)__va(of_read_ulong(prop
, l
/4));
651 prop
= of_get_flat_dt_prop(node
, "linux,initrd-end", &l
);
653 initrd_end
= (unsigned long)
654 __va(of_read_ulong(prop
, l
/4));
655 initrd_below_start_ok
= 1;
661 DBG("initrd_start=0x%lx initrd_end=0x%lx\n", initrd_start
, initrd_end
);
664 static inline void early_init_dt_check_for_initrd(unsigned long node
)
667 #endif /* CONFIG_BLK_DEV_INITRD */
669 static int __init
early_init_dt_scan_chosen(unsigned long node
,
670 const char *uname
, int depth
, void *data
)
672 unsigned long *lprop
;
676 DBG("search \"chosen\", depth: %d, uname: %s\n", depth
, uname
);
679 (strcmp(uname
, "chosen") != 0 && strcmp(uname
, "chosen@0") != 0))
683 /* check if iommu is forced on or off */
684 if (of_get_flat_dt_prop(node
, "linux,iommu-off", NULL
) != NULL
)
686 if (of_get_flat_dt_prop(node
, "linux,iommu-force-on", NULL
) != NULL
)
690 /* mem=x on the command line is the preferred mechanism */
691 lprop
= of_get_flat_dt_prop(node
, "linux,memory-limit", NULL
);
693 memory_limit
= *lprop
;
696 lprop
= of_get_flat_dt_prop(node
, "linux,tce-alloc-start", NULL
);
698 tce_alloc_start
= *lprop
;
699 lprop
= of_get_flat_dt_prop(node
, "linux,tce-alloc-end", NULL
);
701 tce_alloc_end
= *lprop
;
705 lprop
= of_get_flat_dt_prop(node
, "linux,crashkernel-base", NULL
);
707 crashk_res
.start
= *lprop
;
709 lprop
= of_get_flat_dt_prop(node
, "linux,crashkernel-size", NULL
);
711 crashk_res
.end
= crashk_res
.start
+ *lprop
- 1;
714 early_init_dt_check_for_initrd(node
);
716 /* Retreive command line */
717 p
= of_get_flat_dt_prop(node
, "bootargs", &l
);
718 if (p
!= NULL
&& l
> 0)
719 strlcpy(cmd_line
, p
, min((int)l
, COMMAND_LINE_SIZE
));
721 #ifdef CONFIG_CMDLINE
722 if (p
== NULL
|| l
== 0 || (l
== 1 && (*p
) == 0))
723 strlcpy(cmd_line
, CONFIG_CMDLINE
, COMMAND_LINE_SIZE
);
724 #endif /* CONFIG_CMDLINE */
726 DBG("Command line is: %s\n", cmd_line
);
732 static int __init
early_init_dt_scan_root(unsigned long node
,
733 const char *uname
, int depth
, void *data
)
740 prop
= of_get_flat_dt_prop(node
, "#size-cells", NULL
);
741 dt_root_size_cells
= (prop
== NULL
) ? 1 : *prop
;
742 DBG("dt_root_size_cells = %x\n", dt_root_size_cells
);
744 prop
= of_get_flat_dt_prop(node
, "#address-cells", NULL
);
745 dt_root_addr_cells
= (prop
== NULL
) ? 2 : *prop
;
746 DBG("dt_root_addr_cells = %x\n", dt_root_addr_cells
);
752 static u64 __init
dt_mem_next_cell(int s
, cell_t
**cellp
)
757 return of_read_number(p
, s
);
760 #ifdef CONFIG_PPC_PSERIES
762 * Interpret the ibm,dynamic-memory property in the
763 * /ibm,dynamic-reconfiguration-memory node.
764 * This contains a list of memory blocks along with NUMA affinity
767 static int __init
early_init_dt_scan_drconf_memory(unsigned long node
)
769 cell_t
*dm
, *ls
, *usm
;
770 unsigned long l
, n
, flags
;
771 u64 base
, size
, lmb_size
;
772 unsigned int is_kexec_kdump
= 0, rngs
;
774 ls
= of_get_flat_dt_prop(node
, "ibm,lmb-size", &l
);
775 if (ls
== NULL
|| l
< dt_root_size_cells
* sizeof(cell_t
))
777 lmb_size
= dt_mem_next_cell(dt_root_size_cells
, &ls
);
779 dm
= of_get_flat_dt_prop(node
, "ibm,dynamic-memory", &l
);
780 if (dm
== NULL
|| l
< sizeof(cell_t
))
783 n
= *dm
++; /* number of entries */
784 if (l
< (n
* (dt_root_addr_cells
+ 4) + 1) * sizeof(cell_t
))
787 /* check if this is a kexec/kdump kernel. */
788 usm
= of_get_flat_dt_prop(node
, "linux,drconf-usable-memory",
793 for (; n
!= 0; --n
) {
794 base
= dt_mem_next_cell(dt_root_addr_cells
, &dm
);
796 /* skip DRC index, pad, assoc. list index, flags */
798 /* skip this block if the reserved bit is set in flags (0x80)
799 or if the block is not assigned to this partition (0x8) */
800 if ((flags
& 0x80) || !(flags
& 0x8))
804 if (is_kexec_kdump
) {
806 * For each lmb in ibm,dynamic-memory, a corresponding
807 * entry in linux,drconf-usable-memory property contains
808 * a counter 'p' followed by 'p' (base, size) duple.
809 * Now read the counter from
810 * linux,drconf-usable-memory property
812 rngs
= dt_mem_next_cell(dt_root_size_cells
, &usm
);
813 if (!rngs
) /* there are no (base, size) duple */
817 if (is_kexec_kdump
) {
818 base
= dt_mem_next_cell(dt_root_addr_cells
,
820 size
= dt_mem_next_cell(dt_root_size_cells
,
824 if (base
>= 0x80000000ul
)
826 if ((base
+ size
) > 0x80000000ul
)
827 size
= 0x80000000ul
- base
;
836 #define early_init_dt_scan_drconf_memory(node) 0
837 #endif /* CONFIG_PPC_PSERIES */
839 static int __init
early_init_dt_scan_memory(unsigned long node
,
840 const char *uname
, int depth
, void *data
)
842 char *type
= of_get_flat_dt_prop(node
, "device_type", NULL
);
846 /* Look for the ibm,dynamic-reconfiguration-memory node */
848 strcmp(uname
, "ibm,dynamic-reconfiguration-memory") == 0)
849 return early_init_dt_scan_drconf_memory(node
);
851 /* We are scanning "memory" nodes only */
854 * The longtrail doesn't have a device_type on the
855 * /memory node, so look for the node called /memory@0.
857 if (depth
!= 1 || strcmp(uname
, "memory@0") != 0)
859 } else if (strcmp(type
, "memory") != 0)
862 reg
= of_get_flat_dt_prop(node
, "linux,usable-memory", &l
);
864 reg
= of_get_flat_dt_prop(node
, "reg", &l
);
868 endp
= reg
+ (l
/ sizeof(cell_t
));
870 DBG("memory scan node %s, reg size %ld, data: %x %x %x %x,\n",
871 uname
, l
, reg
[0], reg
[1], reg
[2], reg
[3]);
873 while ((endp
- reg
) >= (dt_root_addr_cells
+ dt_root_size_cells
)) {
876 base
= dt_mem_next_cell(dt_root_addr_cells
, ®
);
877 size
= dt_mem_next_cell(dt_root_size_cells
, ®
);
881 DBG(" - %llx , %llx\n", (unsigned long long)base
,
882 (unsigned long long)size
);
885 if (base
>= 0x80000000ul
)
887 if ((base
+ size
) > 0x80000000ul
)
888 size
= 0x80000000ul
- base
;
893 memstart_addr
= min((u64
)memstart_addr
, base
);
899 static void __init
early_reserve_mem(void)
903 unsigned long self_base
;
904 unsigned long self_size
;
906 reserve_map
= (u64
*)(((unsigned long)initial_boot_params
) +
907 initial_boot_params
->off_mem_rsvmap
);
909 /* before we do anything, lets reserve the dt blob */
910 self_base
= __pa((unsigned long)initial_boot_params
);
911 self_size
= initial_boot_params
->totalsize
;
912 lmb_reserve(self_base
, self_size
);
914 #ifdef CONFIG_BLK_DEV_INITRD
915 /* then reserve the initrd, if any */
916 if (initrd_start
&& (initrd_end
> initrd_start
))
917 lmb_reserve(__pa(initrd_start
), initrd_end
- initrd_start
);
918 #endif /* CONFIG_BLK_DEV_INITRD */
922 * Handle the case where we might be booting from an old kexec
923 * image that setup the mem_rsvmap as pairs of 32-bit values
925 if (*reserve_map
> 0xffffffffull
) {
926 u32 base_32
, size_32
;
927 u32
*reserve_map_32
= (u32
*)reserve_map
;
930 base_32
= *(reserve_map_32
++);
931 size_32
= *(reserve_map_32
++);
934 /* skip if the reservation is for the blob */
935 if (base_32
== self_base
&& size_32
== self_size
)
937 DBG("reserving: %x -> %x\n", base_32
, size_32
);
938 lmb_reserve(base_32
, size_32
);
944 base
= *(reserve_map
++);
945 size
= *(reserve_map
++);
948 DBG("reserving: %llx -> %llx\n", base
, size
);
949 lmb_reserve(base
, size
);
953 #ifdef CONFIG_PHYP_DUMP
955 * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
957 * Function to find the largest size we need to reserve
958 * during early boot process.
960 * It either looks for boot param and returns that OR
961 * returns larger of 256 or 5% rounded down to multiples of 256MB.
964 static inline unsigned long phyp_dump_calculate_reserve_size(void)
968 if (phyp_dump_info
->reserve_bootvar
)
969 return phyp_dump_info
->reserve_bootvar
;
971 /* divide by 20 to get 5% of value */
972 tmp
= lmb_end_of_DRAM();
975 /* round it down in multiples of 256 */
976 tmp
= tmp
& ~0x0FFFFFFFUL
;
978 return (tmp
> PHYP_DUMP_RMR_END
? tmp
: PHYP_DUMP_RMR_END
);
982 * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
984 * This routine may reserve memory regions in the kernel only
985 * if the system is supported and a dump was taken in last
986 * boot instance or if the hardware is supported and the
987 * scratch area needs to be setup. In other instances it returns
988 * without reserving anything. The memory in case of dump being
989 * active is freed when the dump is collected (by userland tools).
991 static void __init
phyp_dump_reserve_mem(void)
993 unsigned long base
, size
;
994 unsigned long variable_reserve_size
;
996 if (!phyp_dump_info
->phyp_dump_configured
) {
997 printk(KERN_ERR
"Phyp-dump not supported on this hardware\n");
1001 if (!phyp_dump_info
->phyp_dump_at_boot
) {
1002 printk(KERN_INFO
"Phyp-dump disabled at boot time\n");
1006 variable_reserve_size
= phyp_dump_calculate_reserve_size();
1008 if (phyp_dump_info
->phyp_dump_is_active
) {
1009 /* Reserve *everything* above RMR.Area freed by userland tools*/
1010 base
= variable_reserve_size
;
1011 size
= lmb_end_of_DRAM() - base
;
1013 /* XXX crashed_ram_end is wrong, since it may be beyond
1014 * the memory_limit, it will need to be adjusted. */
1015 lmb_reserve(base
, size
);
1017 phyp_dump_info
->init_reserve_start
= base
;
1018 phyp_dump_info
->init_reserve_size
= size
;
1020 size
= phyp_dump_info
->cpu_state_size
+
1021 phyp_dump_info
->hpte_region_size
+
1022 variable_reserve_size
;
1023 base
= lmb_end_of_DRAM() - size
;
1024 lmb_reserve(base
, size
);
1025 phyp_dump_info
->init_reserve_start
= base
;
1026 phyp_dump_info
->init_reserve_size
= size
;
1030 static inline void __init
phyp_dump_reserve_mem(void) {}
1031 #endif /* CONFIG_PHYP_DUMP && CONFIG_PPC_RTAS */
1034 void __init
early_init_devtree(void *params
)
1038 DBG(" -> early_init_devtree(%p)\n", params
);
1040 /* Setup flat device-tree pointer */
1041 initial_boot_params
= params
;
1043 #ifdef CONFIG_PPC_RTAS
1044 /* Some machines might need RTAS info for debugging, grab it now. */
1045 of_scan_flat_dt(early_init_dt_scan_rtas
, NULL
);
1048 #ifdef CONFIG_PHYP_DUMP
1049 /* scan tree to see if dump occured during last boot */
1050 of_scan_flat_dt(early_init_dt_scan_phyp_dump
, NULL
);
1053 /* Retrieve various informations from the /chosen node of the
1054 * device-tree, including the platform type, initrd location and
1055 * size, TCE reserve, and more ...
1057 of_scan_flat_dt(early_init_dt_scan_chosen
, NULL
);
1059 /* Scan memory nodes and rebuild LMBs */
1061 of_scan_flat_dt(early_init_dt_scan_root
, NULL
);
1062 of_scan_flat_dt(early_init_dt_scan_memory
, NULL
);
1064 /* Save command line for /proc/cmdline and then parse parameters */
1065 strlcpy(boot_command_line
, cmd_line
, COMMAND_LINE_SIZE
);
1066 parse_early_param();
1068 /* Reserve LMB regions used by kernel, initrd, dt, etc... */
1069 lmb_reserve(PHYSICAL_START
, __pa(klimit
) - PHYSICAL_START
);
1070 /* If relocatable, reserve first 32k for interrupt vectors etc. */
1071 if (PHYSICAL_START
> MEMORY_START
)
1072 lmb_reserve(MEMORY_START
, 0x8000);
1073 reserve_kdump_trampoline();
1074 reserve_crashkernel();
1075 early_reserve_mem();
1076 phyp_dump_reserve_mem();
1078 limit
= memory_limit
;
1080 phys_addr_t memsize
;
1082 /* Ensure that total memory size is page-aligned, because
1083 * otherwise mark_bootmem() gets upset. */
1085 memsize
= lmb_phys_mem_size();
1086 if ((memsize
& PAGE_MASK
) != memsize
)
1087 limit
= memsize
& PAGE_MASK
;
1089 lmb_enforce_memory_limit(limit
);
1094 DBG("Phys. mem: %llx\n", lmb_phys_mem_size());
1096 /* We may need to relocate the flat tree, do it now.
1097 * FIXME .. and the initrd too? */
1100 DBG("Scanning CPUs ...\n");
1102 /* Retreive CPU related informations from the flat tree
1103 * (altivec support, boot CPU ID, ...)
1105 of_scan_flat_dt(early_init_dt_scan_cpus
, NULL
);
1107 DBG(" <- early_init_devtree()\n");
1112 * Indicates whether the root node has a given value in its
1113 * compatible property.
1115 int machine_is_compatible(const char *compat
)
1117 struct device_node
*root
;
1120 root
= of_find_node_by_path("/");
1122 rc
= of_device_is_compatible(root
, compat
);
1127 EXPORT_SYMBOL(machine_is_compatible
);
1131 * New implementation of the OF "find" APIs, return a refcounted
1132 * object, call of_node_put() when done. The device tree and list
1133 * are protected by a rw_lock.
1135 * Note that property management will need some locking as well,
1136 * this isn't dealt with yet.
1141 * of_find_node_by_phandle - Find a node given a phandle
1142 * @handle: phandle of the node to find
1144 * Returns a node pointer with refcount incremented, use
1145 * of_node_put() on it when done.
1147 struct device_node
*of_find_node_by_phandle(phandle handle
)
1149 struct device_node
*np
;
1151 read_lock(&devtree_lock
);
1152 for (np
= allnodes
; np
!= 0; np
= np
->allnext
)
1153 if (np
->linux_phandle
== handle
)
1156 read_unlock(&devtree_lock
);
1159 EXPORT_SYMBOL(of_find_node_by_phandle
);
1162 * of_find_next_cache_node - Find a node's subsidiary cache
1163 * @np: node of type "cpu" or "cache"
1165 * Returns a node pointer with refcount incremented, use
1166 * of_node_put() on it when done. Caller should hold a reference
1169 struct device_node
*of_find_next_cache_node(struct device_node
*np
)
1171 struct device_node
*child
;
1172 const phandle
*handle
;
1174 handle
= of_get_property(np
, "l2-cache", NULL
);
1176 handle
= of_get_property(np
, "next-level-cache", NULL
);
1179 return of_find_node_by_phandle(*handle
);
1181 /* OF on pmac has nodes instead of properties named "l2-cache"
1182 * beneath CPU nodes.
1184 if (!strcmp(np
->type
, "cpu"))
1185 for_each_child_of_node(np
, child
)
1186 if (!strcmp(child
->type
, "cache"))
1193 * of_node_get - Increment refcount of a node
1194 * @node: Node to inc refcount, NULL is supported to
1195 * simplify writing of callers
1199 struct device_node
*of_node_get(struct device_node
*node
)
1202 kref_get(&node
->kref
);
1205 EXPORT_SYMBOL(of_node_get
);
1207 static inline struct device_node
* kref_to_device_node(struct kref
*kref
)
1209 return container_of(kref
, struct device_node
, kref
);
1213 * of_node_release - release a dynamically allocated node
1214 * @kref: kref element of the node to be released
1216 * In of_node_put() this function is passed to kref_put()
1217 * as the destructor.
1219 static void of_node_release(struct kref
*kref
)
1221 struct device_node
*node
= kref_to_device_node(kref
);
1222 struct property
*prop
= node
->properties
;
1224 /* We should never be releasing nodes that haven't been detached. */
1225 if (!of_node_check_flag(node
, OF_DETACHED
)) {
1226 printk("WARNING: Bad of_node_put() on %s\n", node
->full_name
);
1228 kref_init(&node
->kref
);
1232 if (!of_node_check_flag(node
, OF_DYNAMIC
))
1236 struct property
*next
= prop
->next
;
1243 prop
= node
->deadprops
;
1244 node
->deadprops
= NULL
;
1247 kfree(node
->full_name
);
1253 * of_node_put - Decrement refcount of a node
1254 * @node: Node to dec refcount, NULL is supported to
1255 * simplify writing of callers
1258 void of_node_put(struct device_node
*node
)
1261 kref_put(&node
->kref
, of_node_release
);
1263 EXPORT_SYMBOL(of_node_put
);
1266 * Plug a device node into the tree and global list.
1268 void of_attach_node(struct device_node
*np
)
1270 unsigned long flags
;
1272 write_lock_irqsave(&devtree_lock
, flags
);
1273 np
->sibling
= np
->parent
->child
;
1274 np
->allnext
= allnodes
;
1275 np
->parent
->child
= np
;
1277 write_unlock_irqrestore(&devtree_lock
, flags
);
1281 * "Unplug" a node from the device tree. The caller must hold
1282 * a reference to the node. The memory associated with the node
1283 * is not freed until its refcount goes to zero.
1285 void of_detach_node(struct device_node
*np
)
1287 struct device_node
*parent
;
1288 unsigned long flags
;
1290 write_lock_irqsave(&devtree_lock
, flags
);
1292 parent
= np
->parent
;
1297 allnodes
= np
->allnext
;
1299 struct device_node
*prev
;
1300 for (prev
= allnodes
;
1301 prev
->allnext
!= np
;
1302 prev
= prev
->allnext
)
1304 prev
->allnext
= np
->allnext
;
1307 if (parent
->child
== np
)
1308 parent
->child
= np
->sibling
;
1310 struct device_node
*prevsib
;
1311 for (prevsib
= np
->parent
->child
;
1312 prevsib
->sibling
!= np
;
1313 prevsib
= prevsib
->sibling
)
1315 prevsib
->sibling
= np
->sibling
;
1318 of_node_set_flag(np
, OF_DETACHED
);
1321 write_unlock_irqrestore(&devtree_lock
, flags
);
1324 #ifdef CONFIG_PPC_PSERIES
1326 * Fix up the uninitialized fields in a new device node:
1327 * name, type and pci-specific fields
1330 static int of_finish_dynamic_node(struct device_node
*node
)
1332 struct device_node
*parent
= of_get_parent(node
);
1334 const phandle
*ibm_phandle
;
1336 node
->name
= of_get_property(node
, "name", NULL
);
1337 node
->type
= of_get_property(node
, "device_type", NULL
);
1340 node
->name
= "<NULL>";
1342 node
->type
= "<NULL>";
1349 /* We don't support that function on PowerMac, at least
1352 if (machine_is(powermac
))
1355 /* fix up new node's linux_phandle field */
1356 if ((ibm_phandle
= of_get_property(node
, "ibm,phandle", NULL
)))
1357 node
->linux_phandle
= *ibm_phandle
;
1360 of_node_put(parent
);
1364 static int prom_reconfig_notifier(struct notifier_block
*nb
,
1365 unsigned long action
, void *node
)
1370 case PSERIES_RECONFIG_ADD
:
1371 err
= of_finish_dynamic_node(node
);
1373 printk(KERN_ERR
"finish_node returned %d\n", err
);
1384 static struct notifier_block prom_reconfig_nb
= {
1385 .notifier_call
= prom_reconfig_notifier
,
1386 .priority
= 10, /* This one needs to run first */
1389 static int __init
prom_reconfig_setup(void)
1391 return pSeries_reconfig_notifier_register(&prom_reconfig_nb
);
1393 __initcall(prom_reconfig_setup
);
1397 * Add a property to a node
1399 int prom_add_property(struct device_node
* np
, struct property
* prop
)
1401 struct property
**next
;
1402 unsigned long flags
;
1405 write_lock_irqsave(&devtree_lock
, flags
);
1406 next
= &np
->properties
;
1408 if (strcmp(prop
->name
, (*next
)->name
) == 0) {
1409 /* duplicate ! don't insert it */
1410 write_unlock_irqrestore(&devtree_lock
, flags
);
1413 next
= &(*next
)->next
;
1416 write_unlock_irqrestore(&devtree_lock
, flags
);
1418 #ifdef CONFIG_PROC_DEVICETREE
1419 /* try to add to proc as well if it was initialized */
1421 proc_device_tree_add_prop(np
->pde
, prop
);
1422 #endif /* CONFIG_PROC_DEVICETREE */
1428 * Remove a property from a node. Note that we don't actually
1429 * remove it, since we have given out who-knows-how-many pointers
1430 * to the data using get-property. Instead we just move the property
1431 * to the "dead properties" list, so it won't be found any more.
1433 int prom_remove_property(struct device_node
*np
, struct property
*prop
)
1435 struct property
**next
;
1436 unsigned long flags
;
1439 write_lock_irqsave(&devtree_lock
, flags
);
1440 next
= &np
->properties
;
1442 if (*next
== prop
) {
1443 /* found the node */
1445 prop
->next
= np
->deadprops
;
1446 np
->deadprops
= prop
;
1450 next
= &(*next
)->next
;
1452 write_unlock_irqrestore(&devtree_lock
, flags
);
1457 #ifdef CONFIG_PROC_DEVICETREE
1458 /* try to remove the proc node as well */
1460 proc_device_tree_remove_prop(np
->pde
, prop
);
1461 #endif /* CONFIG_PROC_DEVICETREE */
1467 * Update a property in a node. Note that we don't actually
1468 * remove it, since we have given out who-knows-how-many pointers
1469 * to the data using get-property. Instead we just move the property
1470 * to the "dead properties" list, and add the new property to the
1473 int prom_update_property(struct device_node
*np
,
1474 struct property
*newprop
,
1475 struct property
*oldprop
)
1477 struct property
**next
;
1478 unsigned long flags
;
1481 write_lock_irqsave(&devtree_lock
, flags
);
1482 next
= &np
->properties
;
1484 if (*next
== oldprop
) {
1485 /* found the node */
1486 newprop
->next
= oldprop
->next
;
1488 oldprop
->next
= np
->deadprops
;
1489 np
->deadprops
= oldprop
;
1493 next
= &(*next
)->next
;
1495 write_unlock_irqrestore(&devtree_lock
, flags
);
1500 #ifdef CONFIG_PROC_DEVICETREE
1501 /* try to add to proc as well if it was initialized */
1503 proc_device_tree_update_prop(np
->pde
, newprop
, oldprop
);
1504 #endif /* CONFIG_PROC_DEVICETREE */
1510 /* Find the device node for a given logical cpu number, also returns the cpu
1511 * local thread number (index in ibm,interrupt-server#s) if relevant and
1512 * asked for (non NULL)
1514 struct device_node
*of_get_cpu_node(int cpu
, unsigned int *thread
)
1517 struct device_node
*np
;
1519 hardid
= get_hard_smp_processor_id(cpu
);
1521 for_each_node_by_type(np
, "cpu") {
1523 unsigned int plen
, t
;
1525 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
1526 * fallback to "reg" property and assume no threads
1528 intserv
= of_get_property(np
, "ibm,ppc-interrupt-server#s",
1530 if (intserv
== NULL
) {
1531 const u32
*reg
= of_get_property(np
, "reg", NULL
);
1534 if (*reg
== hardid
) {
1540 plen
/= sizeof(u32
);
1541 for (t
= 0; t
< plen
; t
++) {
1542 if (hardid
== intserv
[t
]) {
1552 EXPORT_SYMBOL(of_get_cpu_node
);
1554 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
1555 static struct debugfs_blob_wrapper flat_dt_blob
;
1557 static int __init
export_flat_device_tree(void)
1561 flat_dt_blob
.data
= initial_boot_params
;
1562 flat_dt_blob
.size
= initial_boot_params
->totalsize
;
1564 d
= debugfs_create_blob("flat-device-tree", S_IFREG
| S_IRUSR
,
1565 powerpc_debugfs_root
, &flat_dt_blob
);
1571 __initcall(export_flat_device_tree
);