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
;
84 * This function is used to scan the flattened device-tree, it is
85 * used to extract the memory informations at boot before we can
88 int __init
of_scan_flat_dt(int (*it
)(unsigned long node
,
89 const char *uname
, int depth
,
93 unsigned long p
= ((unsigned long)initial_boot_params
) +
94 initial_boot_params
->off_dt_struct
;
99 u32 tag
= *((u32
*)p
);
103 if (tag
== OF_DT_END_NODE
) {
107 if (tag
== OF_DT_NOP
)
109 if (tag
== OF_DT_END
)
111 if (tag
== OF_DT_PROP
) {
112 u32 sz
= *((u32
*)p
);
114 if (initial_boot_params
->version
< 0x10)
115 p
= _ALIGN(p
, sz
>= 8 ? 8 : 4);
120 if (tag
!= OF_DT_BEGIN_NODE
) {
121 printk(KERN_WARNING
"Invalid tag %x scanning flattened"
122 " device tree !\n", tag
);
127 p
= _ALIGN(p
+ strlen(pathp
) + 1, 4);
128 if ((*pathp
) == '/') {
130 for (lp
= NULL
, np
= pathp
; *np
; np
++)
136 rc
= it(p
, pathp
, depth
, data
);
144 unsigned long __init
of_get_flat_dt_root(void)
146 unsigned long p
= ((unsigned long)initial_boot_params
) +
147 initial_boot_params
->off_dt_struct
;
149 while(*((u32
*)p
) == OF_DT_NOP
)
151 BUG_ON (*((u32
*)p
) != OF_DT_BEGIN_NODE
);
153 return _ALIGN(p
+ strlen((char *)p
) + 1, 4);
157 * This function can be used within scan_flattened_dt callback to get
158 * access to properties
160 void* __init
of_get_flat_dt_prop(unsigned long node
, const char *name
,
163 unsigned long p
= node
;
166 u32 tag
= *((u32
*)p
);
171 if (tag
== OF_DT_NOP
)
173 if (tag
!= OF_DT_PROP
)
177 noff
= *((u32
*)(p
+ 4));
179 if (initial_boot_params
->version
< 0x10)
180 p
= _ALIGN(p
, sz
>= 8 ? 8 : 4);
182 nstr
= find_flat_dt_string(noff
);
184 printk(KERN_WARNING
"Can't find property index"
188 if (strcmp(name
, nstr
) == 0) {
198 int __init
of_flat_dt_is_compatible(unsigned long node
, const char *compat
)
201 unsigned long cplen
, l
;
203 cp
= of_get_flat_dt_prop(node
, "compatible", &cplen
);
207 if (strncasecmp(cp
, compat
, strlen(compat
)) == 0)
217 static void *__init
unflatten_dt_alloc(unsigned long *mem
, unsigned long size
,
222 *mem
= _ALIGN(*mem
, align
);
229 static unsigned long __init
unflatten_dt_node(unsigned long mem
,
231 struct device_node
*dad
,
232 struct device_node
***allnextpp
,
233 unsigned long fpsize
)
235 struct device_node
*np
;
236 struct property
*pp
, **prev_pp
= NULL
;
239 unsigned int l
, allocl
;
243 tag
= *((u32
*)(*p
));
244 if (tag
!= OF_DT_BEGIN_NODE
) {
245 printk("Weird tag at start of node: %x\n", tag
);
250 l
= allocl
= strlen(pathp
) + 1;
251 *p
= _ALIGN(*p
+ l
, 4);
253 /* version 0x10 has a more compact unit name here instead of the full
254 * path. we accumulate the full path size using "fpsize", we'll rebuild
255 * it later. We detect this because the first character of the name is
258 if ((*pathp
) != '/') {
261 /* root node: special case. fpsize accounts for path
262 * plus terminating zero. root node only has '/', so
263 * fpsize should be 2, but we want to avoid the first
264 * level nodes to have two '/' so we use fpsize 1 here
269 /* account for '/' and path size minus terminal 0
278 np
= unflatten_dt_alloc(&mem
, sizeof(struct device_node
) + allocl
,
279 __alignof__(struct device_node
));
281 memset(np
, 0, sizeof(*np
));
282 np
->full_name
= ((char*)np
) + sizeof(struct device_node
);
284 char *p
= np
->full_name
;
285 /* rebuild full path for new format */
286 if (dad
&& dad
->parent
) {
287 strcpy(p
, dad
->full_name
);
289 if ((strlen(p
) + l
+ 1) != allocl
) {
290 DBG("%s: p: %d, l: %d, a: %d\n",
291 pathp
, (int)strlen(p
), l
, allocl
);
299 memcpy(np
->full_name
, pathp
, l
);
300 prev_pp
= &np
->properties
;
302 *allnextpp
= &np
->allnext
;
305 /* we temporarily use the next field as `last_child'*/
309 dad
->next
->sibling
= np
;
312 kref_init(&np
->kref
);
318 tag
= *((u32
*)(*p
));
319 if (tag
== OF_DT_NOP
) {
323 if (tag
!= OF_DT_PROP
)
327 noff
= *((u32
*)((*p
) + 4));
329 if (initial_boot_params
->version
< 0x10)
330 *p
= _ALIGN(*p
, sz
>= 8 ? 8 : 4);
332 pname
= find_flat_dt_string(noff
);
334 printk("Can't find property name in list !\n");
337 if (strcmp(pname
, "name") == 0)
339 l
= strlen(pname
) + 1;
340 pp
= unflatten_dt_alloc(&mem
, sizeof(struct property
),
341 __alignof__(struct property
));
343 if (strcmp(pname
, "linux,phandle") == 0) {
344 np
->node
= *((u32
*)*p
);
345 if (np
->linux_phandle
== 0)
346 np
->linux_phandle
= np
->node
;
348 if (strcmp(pname
, "ibm,phandle") == 0)
349 np
->linux_phandle
= *((u32
*)*p
);
352 pp
->value
= (void *)*p
;
356 *p
= _ALIGN((*p
) + sz
, 4);
358 /* with version 0x10 we may not have the name property, recreate
359 * it here from the unit name if absent
362 char *p
= pathp
, *ps
= pathp
, *pa
= NULL
;
375 pp
= unflatten_dt_alloc(&mem
, sizeof(struct property
) + sz
,
376 __alignof__(struct property
));
383 memcpy(pp
->value
, ps
, sz
- 1);
384 ((char *)pp
->value
)[sz
- 1] = 0;
385 DBG("fixed up name for %s -> %s\n", pathp
,
391 np
->name
= of_get_property(np
, "name", NULL
);
392 np
->type
= of_get_property(np
, "device_type", NULL
);
399 while (tag
== OF_DT_BEGIN_NODE
) {
400 mem
= unflatten_dt_node(mem
, p
, np
, allnextpp
, fpsize
);
401 tag
= *((u32
*)(*p
));
403 if (tag
!= OF_DT_END_NODE
) {
404 printk("Weird tag at end of node: %x\n", tag
);
411 static int __init
early_parse_mem(char *p
)
416 memory_limit
= PAGE_ALIGN(memparse(p
, &p
));
417 DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit
);
421 early_param("mem", early_parse_mem
);
424 * move_device_tree - move tree to an unused area, if needed.
426 * The device tree may be allocated beyond our memory limit, or inside the
427 * crash kernel region for kdump. If so, move it out of the way.
429 static void __init
move_device_tree(void)
431 unsigned long start
, size
;
434 DBG("-> move_device_tree\n");
436 start
= __pa(initial_boot_params
);
437 size
= initial_boot_params
->totalsize
;
439 if ((memory_limit
&& (start
+ size
) > memory_limit
) ||
440 overlaps_crashkernel(start
, size
)) {
441 p
= __va(lmb_alloc_base(size
, PAGE_SIZE
, lmb
.rmo_size
));
442 memcpy(p
, initial_boot_params
, size
);
443 initial_boot_params
= (struct boot_param_header
*)p
;
444 DBG("Moved device tree to 0x%p\n", p
);
447 DBG("<- move_device_tree\n");
451 * unflattens the device-tree passed by the firmware, creating the
452 * tree of struct device_node. It also fills the "name" and "type"
453 * pointers of the nodes so the normal device-tree walking functions
454 * can be used (this used to be done by finish_device_tree)
456 void __init
unflatten_device_tree(void)
458 unsigned long start
, mem
, size
;
459 struct device_node
**allnextp
= &allnodes
;
461 DBG(" -> unflatten_device_tree()\n");
463 /* First pass, scan for size */
464 start
= ((unsigned long)initial_boot_params
) +
465 initial_boot_params
->off_dt_struct
;
466 size
= unflatten_dt_node(0, &start
, NULL
, NULL
, 0);
467 size
= (size
| 3) + 1;
469 DBG(" size is %lx, allocating...\n", size
);
471 /* Allocate memory for the expanded device tree */
472 mem
= lmb_alloc(size
+ 4, __alignof__(struct device_node
));
473 mem
= (unsigned long) __va(mem
);
475 ((u32
*)mem
)[size
/ 4] = 0xdeadbeef;
477 DBG(" unflattening %lx...\n", mem
);
479 /* Second pass, do actual unflattening */
480 start
= ((unsigned long)initial_boot_params
) +
481 initial_boot_params
->off_dt_struct
;
482 unflatten_dt_node(mem
, &start
, NULL
, &allnextp
, 0);
483 if (*((u32
*)start
) != OF_DT_END
)
484 printk(KERN_WARNING
"Weird tag at end of tree: %08x\n", *((u32
*)start
));
485 if (((u32
*)mem
)[size
/ 4] != 0xdeadbeef)
486 printk(KERN_WARNING
"End of tree marker overwritten: %08x\n",
487 ((u32
*)mem
)[size
/ 4] );
490 /* Get pointer to OF "/chosen" node for use everywhere */
491 of_chosen
= of_find_node_by_path("/chosen");
492 if (of_chosen
== NULL
)
493 of_chosen
= of_find_node_by_path("/chosen@0");
495 DBG(" <- unflatten_device_tree()\n");
499 * ibm,pa-features is a per-cpu property that contains a string of
500 * attribute descriptors, each of which has a 2 byte header plus up
501 * to 254 bytes worth of processor attribute bits. First header
502 * byte specifies the number of bytes following the header.
503 * Second header byte is an "attribute-specifier" type, of which
504 * zero is the only currently-defined value.
505 * Implementation: Pass in the byte and bit offset for the feature
506 * that we are interested in. The function will return -1 if the
507 * pa-features property is missing, or a 1/0 to indicate if the feature
508 * is supported/not supported. Note that the bit numbers are
509 * big-endian to match the definition in PAPR.
511 static struct ibm_pa_feature
{
512 unsigned long cpu_features
; /* CPU_FTR_xxx bit */
513 unsigned int cpu_user_ftrs
; /* PPC_FEATURE_xxx bit */
514 unsigned char pabyte
; /* byte number in ibm,pa-features */
515 unsigned char pabit
; /* bit number (big-endian) */
516 unsigned char invert
; /* if 1, pa bit set => clear feature */
517 } ibm_pa_features
[] __initdata
= {
518 {0, PPC_FEATURE_HAS_MMU
, 0, 0, 0},
519 {0, PPC_FEATURE_HAS_FPU
, 0, 1, 0},
520 {CPU_FTR_SLB
, 0, 0, 2, 0},
521 {CPU_FTR_CTRL
, 0, 0, 3, 0},
522 {CPU_FTR_NOEXECUTE
, 0, 0, 6, 0},
523 {CPU_FTR_NODSISRALIGN
, 0, 1, 1, 1},
524 {CPU_FTR_CI_LARGE_PAGE
, 0, 1, 2, 0},
525 {CPU_FTR_REAL_LE
, PPC_FEATURE_TRUE_LE
, 5, 0, 0},
528 static void __init
scan_features(unsigned long node
, unsigned char *ftrs
,
529 unsigned long tablelen
,
530 struct ibm_pa_feature
*fp
,
531 unsigned long ft_size
)
533 unsigned long i
, len
, bit
;
535 /* find descriptor with type == 0 */
541 return; /* descriptor 0 not found */
548 /* loop over bits we know about */
549 for (i
= 0; i
< ft_size
; ++i
, ++fp
) {
550 if (fp
->pabyte
>= ftrs
[0])
552 bit
= (ftrs
[2 + fp
->pabyte
] >> (7 - fp
->pabit
)) & 1;
553 if (bit
^ fp
->invert
) {
554 cur_cpu_spec
->cpu_features
|= fp
->cpu_features
;
555 cur_cpu_spec
->cpu_user_features
|= fp
->cpu_user_ftrs
;
557 cur_cpu_spec
->cpu_features
&= ~fp
->cpu_features
;
558 cur_cpu_spec
->cpu_user_features
&= ~fp
->cpu_user_ftrs
;
563 static void __init
check_cpu_pa_features(unsigned long node
)
565 unsigned char *pa_ftrs
;
566 unsigned long tablelen
;
568 pa_ftrs
= of_get_flat_dt_prop(node
, "ibm,pa-features", &tablelen
);
572 scan_features(node
, pa_ftrs
, tablelen
,
573 ibm_pa_features
, ARRAY_SIZE(ibm_pa_features
));
576 #ifdef CONFIG_PPC_STD_MMU_64
577 static void __init
check_cpu_slb_size(unsigned long node
)
581 slb_size_ptr
= of_get_flat_dt_prop(node
, "slb-size", NULL
);
582 if (slb_size_ptr
!= NULL
) {
583 mmu_slb_size
= *slb_size_ptr
;
586 slb_size_ptr
= of_get_flat_dt_prop(node
, "ibm,slb-size", NULL
);
587 if (slb_size_ptr
!= NULL
) {
588 mmu_slb_size
= *slb_size_ptr
;
592 #define check_cpu_slb_size(node) do { } while(0)
595 static struct feature_property
{
598 unsigned long cpu_feature
;
599 unsigned long cpu_user_ftr
;
600 } feature_properties
[] __initdata
= {
601 #ifdef CONFIG_ALTIVEC
602 {"altivec", 0, CPU_FTR_ALTIVEC
, PPC_FEATURE_HAS_ALTIVEC
},
603 {"ibm,vmx", 1, CPU_FTR_ALTIVEC
, PPC_FEATURE_HAS_ALTIVEC
},
604 #endif /* CONFIG_ALTIVEC */
606 /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
607 {"ibm,vmx", 2, CPU_FTR_VSX
, PPC_FEATURE_HAS_VSX
},
608 #endif /* CONFIG_VSX */
610 {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP
},
611 {"ibm,purr", 1, CPU_FTR_PURR
, 0},
612 {"ibm,spurr", 1, CPU_FTR_SPURR
, 0},
613 #endif /* CONFIG_PPC64 */
616 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
617 static inline void identical_pvr_fixup(unsigned long node
)
620 char *model
= of_get_flat_dt_prop(node
, "model", NULL
);
623 * Since 440GR(x)/440EP(x) processors have the same pvr,
624 * we check the node path and set bit 28 in the cur_cpu_spec
625 * pvr for EP(x) processor version. This bit is always 0 in
626 * the "real" pvr. Then we call identify_cpu again with
627 * the new logical pvr to enable FPU support.
629 if (model
&& strstr(model
, "440EP")) {
630 pvr
= cur_cpu_spec
->pvr_value
| 0x8;
631 identify_cpu(0, pvr
);
632 DBG("Using logical pvr %x for %s\n", pvr
, model
);
636 #define identical_pvr_fixup(node) do { } while(0)
639 static void __init
check_cpu_feature_properties(unsigned long node
)
642 struct feature_property
*fp
= feature_properties
;
645 for (i
= 0; i
< ARRAY_SIZE(feature_properties
); ++i
, ++fp
) {
646 prop
= of_get_flat_dt_prop(node
, fp
->name
, NULL
);
647 if (prop
&& *prop
>= fp
->min_value
) {
648 cur_cpu_spec
->cpu_features
|= fp
->cpu_feature
;
649 cur_cpu_spec
->cpu_user_features
|= fp
->cpu_user_ftr
;
654 static int __init
early_init_dt_scan_cpus(unsigned long node
,
655 const char *uname
, int depth
,
658 static int logical_cpuid
= 0;
659 char *type
= of_get_flat_dt_prop(node
, "device_type", NULL
);
666 /* We are scanning "cpu" nodes only */
667 if (type
== NULL
|| strcmp(type
, "cpu") != 0)
670 /* Get physical cpuid */
671 intserv
= of_get_flat_dt_prop(node
, "ibm,ppc-interrupt-server#s", &len
);
673 nthreads
= len
/ sizeof(int);
675 intserv
= of_get_flat_dt_prop(node
, "reg", NULL
);
680 * Now see if any of these threads match our boot cpu.
681 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
683 for (i
= 0; i
< nthreads
; i
++) {
685 * version 2 of the kexec param format adds the phys cpuid of
688 if (initial_boot_params
&& initial_boot_params
->version
>= 2) {
690 initial_boot_params
->boot_cpuid_phys
) {
696 * Check if it's the boot-cpu, set it's hw index now,
697 * unfortunately this format did not support booting
698 * off secondary threads.
700 if (of_get_flat_dt_prop(node
,
701 "linux,boot-cpu", NULL
) != NULL
) {
708 /* logical cpu id is always 0 on UP kernels */
714 DBG("boot cpu: logical %d physical %d\n", logical_cpuid
,
716 boot_cpuid
= logical_cpuid
;
717 set_hard_smp_processor_id(boot_cpuid
, intserv
[i
]);
720 * PAPR defines "logical" PVR values for cpus that
721 * meet various levels of the architecture:
722 * 0x0f000001 Architecture version 2.04
723 * 0x0f000002 Architecture version 2.05
724 * If the cpu-version property in the cpu node contains
725 * such a value, we call identify_cpu again with the
726 * logical PVR value in order to use the cpu feature
727 * bits appropriate for the architecture level.
729 * A POWER6 partition in "POWER6 architected" mode
730 * uses the 0x0f000002 PVR value; in POWER5+ mode
731 * it uses 0x0f000001.
733 prop
= of_get_flat_dt_prop(node
, "cpu-version", NULL
);
734 if (prop
&& (*prop
& 0xff000000) == 0x0f000000)
735 identify_cpu(0, *prop
);
737 identical_pvr_fixup(node
);
740 check_cpu_feature_properties(node
);
741 check_cpu_pa_features(node
);
742 check_cpu_slb_size(node
);
744 #ifdef CONFIG_PPC_PSERIES
746 cur_cpu_spec
->cpu_features
|= CPU_FTR_SMT
;
748 cur_cpu_spec
->cpu_features
&= ~CPU_FTR_SMT
;
754 #ifdef CONFIG_BLK_DEV_INITRD
755 static void __init
early_init_dt_check_for_initrd(unsigned long node
)
760 DBG("Looking for initrd properties... ");
762 prop
= of_get_flat_dt_prop(node
, "linux,initrd-start", &l
);
764 initrd_start
= (unsigned long)__va(of_read_ulong(prop
, l
/4));
766 prop
= of_get_flat_dt_prop(node
, "linux,initrd-end", &l
);
768 initrd_end
= (unsigned long)
769 __va(of_read_ulong(prop
, l
/4));
770 initrd_below_start_ok
= 1;
776 DBG("initrd_start=0x%lx initrd_end=0x%lx\n", initrd_start
, initrd_end
);
779 static inline void early_init_dt_check_for_initrd(unsigned long node
)
782 #endif /* CONFIG_BLK_DEV_INITRD */
784 static int __init
early_init_dt_scan_chosen(unsigned long node
,
785 const char *uname
, int depth
, void *data
)
787 unsigned long *lprop
;
791 DBG("search \"chosen\", depth: %d, uname: %s\n", depth
, uname
);
794 (strcmp(uname
, "chosen") != 0 && strcmp(uname
, "chosen@0") != 0))
798 /* check if iommu is forced on or off */
799 if (of_get_flat_dt_prop(node
, "linux,iommu-off", NULL
) != NULL
)
801 if (of_get_flat_dt_prop(node
, "linux,iommu-force-on", NULL
) != NULL
)
805 /* mem=x on the command line is the preferred mechanism */
806 lprop
= of_get_flat_dt_prop(node
, "linux,memory-limit", NULL
);
808 memory_limit
= *lprop
;
811 lprop
= of_get_flat_dt_prop(node
, "linux,tce-alloc-start", NULL
);
813 tce_alloc_start
= *lprop
;
814 lprop
= of_get_flat_dt_prop(node
, "linux,tce-alloc-end", NULL
);
816 tce_alloc_end
= *lprop
;
820 lprop
= of_get_flat_dt_prop(node
, "linux,crashkernel-base", NULL
);
822 crashk_res
.start
= *lprop
;
824 lprop
= of_get_flat_dt_prop(node
, "linux,crashkernel-size", NULL
);
826 crashk_res
.end
= crashk_res
.start
+ *lprop
- 1;
829 early_init_dt_check_for_initrd(node
);
831 /* Retreive command line */
832 p
= of_get_flat_dt_prop(node
, "bootargs", &l
);
833 if (p
!= NULL
&& l
> 0)
834 strlcpy(cmd_line
, p
, min((int)l
, COMMAND_LINE_SIZE
));
836 #ifdef CONFIG_CMDLINE
837 if (p
== NULL
|| l
== 0 || (l
== 1 && (*p
) == 0))
838 strlcpy(cmd_line
, CONFIG_CMDLINE
, COMMAND_LINE_SIZE
);
839 #endif /* CONFIG_CMDLINE */
841 DBG("Command line is: %s\n", cmd_line
);
847 static int __init
early_init_dt_scan_root(unsigned long node
,
848 const char *uname
, int depth
, void *data
)
855 prop
= of_get_flat_dt_prop(node
, "#size-cells", NULL
);
856 dt_root_size_cells
= (prop
== NULL
) ? 1 : *prop
;
857 DBG("dt_root_size_cells = %x\n", dt_root_size_cells
);
859 prop
= of_get_flat_dt_prop(node
, "#address-cells", NULL
);
860 dt_root_addr_cells
= (prop
== NULL
) ? 2 : *prop
;
861 DBG("dt_root_addr_cells = %x\n", dt_root_addr_cells
);
867 static u64 __init
dt_mem_next_cell(int s
, cell_t
**cellp
)
872 return of_read_number(p
, s
);
875 #ifdef CONFIG_PPC_PSERIES
877 * Interpret the ibm,dynamic-memory property in the
878 * /ibm,dynamic-reconfiguration-memory node.
879 * This contains a list of memory blocks along with NUMA affinity
882 static int __init
early_init_dt_scan_drconf_memory(unsigned long node
)
884 cell_t
*dm
, *ls
, *usm
;
885 unsigned long l
, n
, flags
;
886 u64 base
, size
, lmb_size
;
887 unsigned int is_kexec_kdump
= 0, rngs
;
889 ls
= of_get_flat_dt_prop(node
, "ibm,lmb-size", &l
);
890 if (ls
== NULL
|| l
< dt_root_size_cells
* sizeof(cell_t
))
892 lmb_size
= dt_mem_next_cell(dt_root_size_cells
, &ls
);
894 dm
= of_get_flat_dt_prop(node
, "ibm,dynamic-memory", &l
);
895 if (dm
== NULL
|| l
< sizeof(cell_t
))
898 n
= *dm
++; /* number of entries */
899 if (l
< (n
* (dt_root_addr_cells
+ 4) + 1) * sizeof(cell_t
))
902 /* check if this is a kexec/kdump kernel. */
903 usm
= of_get_flat_dt_prop(node
, "linux,drconf-usable-memory",
908 for (; n
!= 0; --n
) {
909 base
= dt_mem_next_cell(dt_root_addr_cells
, &dm
);
911 /* skip DRC index, pad, assoc. list index, flags */
913 /* skip this block if the reserved bit is set in flags (0x80)
914 or if the block is not assigned to this partition (0x8) */
915 if ((flags
& 0x80) || !(flags
& 0x8))
919 if (is_kexec_kdump
) {
921 * For each lmb in ibm,dynamic-memory, a corresponding
922 * entry in linux,drconf-usable-memory property contains
923 * a counter 'p' followed by 'p' (base, size) duple.
924 * Now read the counter from
925 * linux,drconf-usable-memory property
927 rngs
= dt_mem_next_cell(dt_root_size_cells
, &usm
);
928 if (!rngs
) /* there are no (base, size) duple */
932 if (is_kexec_kdump
) {
933 base
= dt_mem_next_cell(dt_root_addr_cells
,
935 size
= dt_mem_next_cell(dt_root_size_cells
,
939 if (base
>= 0x80000000ul
)
941 if ((base
+ size
) > 0x80000000ul
)
942 size
= 0x80000000ul
- base
;
951 #define early_init_dt_scan_drconf_memory(node) 0
952 #endif /* CONFIG_PPC_PSERIES */
954 static int __init
early_init_dt_scan_memory(unsigned long node
,
955 const char *uname
, int depth
, void *data
)
957 char *type
= of_get_flat_dt_prop(node
, "device_type", NULL
);
961 /* Look for the ibm,dynamic-reconfiguration-memory node */
963 strcmp(uname
, "ibm,dynamic-reconfiguration-memory") == 0)
964 return early_init_dt_scan_drconf_memory(node
);
966 /* We are scanning "memory" nodes only */
969 * The longtrail doesn't have a device_type on the
970 * /memory node, so look for the node called /memory@0.
972 if (depth
!= 1 || strcmp(uname
, "memory@0") != 0)
974 } else if (strcmp(type
, "memory") != 0)
977 reg
= of_get_flat_dt_prop(node
, "linux,usable-memory", &l
);
979 reg
= of_get_flat_dt_prop(node
, "reg", &l
);
983 endp
= reg
+ (l
/ sizeof(cell_t
));
985 DBG("memory scan node %s, reg size %ld, data: %x %x %x %x,\n",
986 uname
, l
, reg
[0], reg
[1], reg
[2], reg
[3]);
988 while ((endp
- reg
) >= (dt_root_addr_cells
+ dt_root_size_cells
)) {
991 base
= dt_mem_next_cell(dt_root_addr_cells
, ®
);
992 size
= dt_mem_next_cell(dt_root_size_cells
, ®
);
996 DBG(" - %llx , %llx\n", (unsigned long long)base
,
997 (unsigned long long)size
);
1000 if (base
>= 0x80000000ul
)
1002 if ((base
+ size
) > 0x80000000ul
)
1003 size
= 0x80000000ul
- base
;
1006 lmb_add(base
, size
);
1008 memstart_addr
= min((u64
)memstart_addr
, base
);
1014 static void __init
early_reserve_mem(void)
1018 unsigned long self_base
;
1019 unsigned long self_size
;
1021 reserve_map
= (u64
*)(((unsigned long)initial_boot_params
) +
1022 initial_boot_params
->off_mem_rsvmap
);
1024 /* before we do anything, lets reserve the dt blob */
1025 self_base
= __pa((unsigned long)initial_boot_params
);
1026 self_size
= initial_boot_params
->totalsize
;
1027 lmb_reserve(self_base
, self_size
);
1029 #ifdef CONFIG_BLK_DEV_INITRD
1030 /* then reserve the initrd, if any */
1031 if (initrd_start
&& (initrd_end
> initrd_start
))
1032 lmb_reserve(__pa(initrd_start
), initrd_end
- initrd_start
);
1033 #endif /* CONFIG_BLK_DEV_INITRD */
1037 * Handle the case where we might be booting from an old kexec
1038 * image that setup the mem_rsvmap as pairs of 32-bit values
1040 if (*reserve_map
> 0xffffffffull
) {
1041 u32 base_32
, size_32
;
1042 u32
*reserve_map_32
= (u32
*)reserve_map
;
1045 base_32
= *(reserve_map_32
++);
1046 size_32
= *(reserve_map_32
++);
1049 /* skip if the reservation is for the blob */
1050 if (base_32
== self_base
&& size_32
== self_size
)
1052 DBG("reserving: %x -> %x\n", base_32
, size_32
);
1053 lmb_reserve(base_32
, size_32
);
1059 base
= *(reserve_map
++);
1060 size
= *(reserve_map
++);
1063 DBG("reserving: %llx -> %llx\n", base
, size
);
1064 lmb_reserve(base
, size
);
1068 #ifdef CONFIG_PHYP_DUMP
1070 * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
1072 * Function to find the largest size we need to reserve
1073 * during early boot process.
1075 * It either looks for boot param and returns that OR
1076 * returns larger of 256 or 5% rounded down to multiples of 256MB.
1079 static inline unsigned long phyp_dump_calculate_reserve_size(void)
1083 if (phyp_dump_info
->reserve_bootvar
)
1084 return phyp_dump_info
->reserve_bootvar
;
1086 /* divide by 20 to get 5% of value */
1087 tmp
= lmb_end_of_DRAM();
1090 /* round it down in multiples of 256 */
1091 tmp
= tmp
& ~0x0FFFFFFFUL
;
1093 return (tmp
> PHYP_DUMP_RMR_END
? tmp
: PHYP_DUMP_RMR_END
);
1097 * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
1099 * This routine may reserve memory regions in the kernel only
1100 * if the system is supported and a dump was taken in last
1101 * boot instance or if the hardware is supported and the
1102 * scratch area needs to be setup. In other instances it returns
1103 * without reserving anything. The memory in case of dump being
1104 * active is freed when the dump is collected (by userland tools).
1106 static void __init
phyp_dump_reserve_mem(void)
1108 unsigned long base
, size
;
1109 unsigned long variable_reserve_size
;
1111 if (!phyp_dump_info
->phyp_dump_configured
) {
1112 printk(KERN_ERR
"Phyp-dump not supported on this hardware\n");
1116 if (!phyp_dump_info
->phyp_dump_at_boot
) {
1117 printk(KERN_INFO
"Phyp-dump disabled at boot time\n");
1121 variable_reserve_size
= phyp_dump_calculate_reserve_size();
1123 if (phyp_dump_info
->phyp_dump_is_active
) {
1124 /* Reserve *everything* above RMR.Area freed by userland tools*/
1125 base
= variable_reserve_size
;
1126 size
= lmb_end_of_DRAM() - base
;
1128 /* XXX crashed_ram_end is wrong, since it may be beyond
1129 * the memory_limit, it will need to be adjusted. */
1130 lmb_reserve(base
, size
);
1132 phyp_dump_info
->init_reserve_start
= base
;
1133 phyp_dump_info
->init_reserve_size
= size
;
1135 size
= phyp_dump_info
->cpu_state_size
+
1136 phyp_dump_info
->hpte_region_size
+
1137 variable_reserve_size
;
1138 base
= lmb_end_of_DRAM() - size
;
1139 lmb_reserve(base
, size
);
1140 phyp_dump_info
->init_reserve_start
= base
;
1141 phyp_dump_info
->init_reserve_size
= size
;
1145 static inline void __init
phyp_dump_reserve_mem(void) {}
1146 #endif /* CONFIG_PHYP_DUMP && CONFIG_PPC_RTAS */
1149 void __init
early_init_devtree(void *params
)
1153 DBG(" -> early_init_devtree(%p)\n", params
);
1155 /* Setup flat device-tree pointer */
1156 initial_boot_params
= params
;
1158 #ifdef CONFIG_PPC_RTAS
1159 /* Some machines might need RTAS info for debugging, grab it now. */
1160 of_scan_flat_dt(early_init_dt_scan_rtas
, NULL
);
1163 #ifdef CONFIG_PHYP_DUMP
1164 /* scan tree to see if dump occured during last boot */
1165 of_scan_flat_dt(early_init_dt_scan_phyp_dump
, NULL
);
1168 /* Retrieve various informations from the /chosen node of the
1169 * device-tree, including the platform type, initrd location and
1170 * size, TCE reserve, and more ...
1172 of_scan_flat_dt(early_init_dt_scan_chosen
, NULL
);
1174 /* Scan memory nodes and rebuild LMBs */
1176 of_scan_flat_dt(early_init_dt_scan_root
, NULL
);
1177 of_scan_flat_dt(early_init_dt_scan_memory
, NULL
);
1179 /* Save command line for /proc/cmdline and then parse parameters */
1180 strlcpy(boot_command_line
, cmd_line
, COMMAND_LINE_SIZE
);
1181 parse_early_param();
1183 /* Reserve LMB regions used by kernel, initrd, dt, etc... */
1184 lmb_reserve(PHYSICAL_START
, __pa(klimit
) - PHYSICAL_START
);
1185 /* If relocatable, reserve first 32k for interrupt vectors etc. */
1186 if (PHYSICAL_START
> MEMORY_START
)
1187 lmb_reserve(MEMORY_START
, 0x8000);
1188 reserve_kdump_trampoline();
1189 reserve_crashkernel();
1190 early_reserve_mem();
1191 phyp_dump_reserve_mem();
1193 limit
= memory_limit
;
1195 phys_addr_t memsize
;
1197 /* Ensure that total memory size is page-aligned, because
1198 * otherwise mark_bootmem() gets upset. */
1200 memsize
= lmb_phys_mem_size();
1201 if ((memsize
& PAGE_MASK
) != memsize
)
1202 limit
= memsize
& PAGE_MASK
;
1204 lmb_enforce_memory_limit(limit
);
1209 DBG("Phys. mem: %llx\n", lmb_phys_mem_size());
1211 /* We may need to relocate the flat tree, do it now.
1212 * FIXME .. and the initrd too? */
1215 DBG("Scanning CPUs ...\n");
1217 /* Retreive CPU related informations from the flat tree
1218 * (altivec support, boot CPU ID, ...)
1220 of_scan_flat_dt(early_init_dt_scan_cpus
, NULL
);
1222 DBG(" <- early_init_devtree()\n");
1227 * Indicates whether the root node has a given value in its
1228 * compatible property.
1230 int machine_is_compatible(const char *compat
)
1232 struct device_node
*root
;
1235 root
= of_find_node_by_path("/");
1237 rc
= of_device_is_compatible(root
, compat
);
1242 EXPORT_SYMBOL(machine_is_compatible
);
1246 * New implementation of the OF "find" APIs, return a refcounted
1247 * object, call of_node_put() when done. The device tree and list
1248 * are protected by a rw_lock.
1250 * Note that property management will need some locking as well,
1251 * this isn't dealt with yet.
1256 * of_find_node_by_phandle - Find a node given a phandle
1257 * @handle: phandle of the node to find
1259 * Returns a node pointer with refcount incremented, use
1260 * of_node_put() on it when done.
1262 struct device_node
*of_find_node_by_phandle(phandle handle
)
1264 struct device_node
*np
;
1266 read_lock(&devtree_lock
);
1267 for (np
= allnodes
; np
!= 0; np
= np
->allnext
)
1268 if (np
->linux_phandle
== handle
)
1271 read_unlock(&devtree_lock
);
1274 EXPORT_SYMBOL(of_find_node_by_phandle
);
1277 * of_find_next_cache_node - Find a node's subsidiary cache
1278 * @np: node of type "cpu" or "cache"
1280 * Returns a node pointer with refcount incremented, use
1281 * of_node_put() on it when done. Caller should hold a reference
1284 struct device_node
*of_find_next_cache_node(struct device_node
*np
)
1286 struct device_node
*child
;
1287 const phandle
*handle
;
1289 handle
= of_get_property(np
, "l2-cache", NULL
);
1291 handle
= of_get_property(np
, "next-level-cache", NULL
);
1294 return of_find_node_by_phandle(*handle
);
1296 /* OF on pmac has nodes instead of properties named "l2-cache"
1297 * beneath CPU nodes.
1299 if (!strcmp(np
->type
, "cpu"))
1300 for_each_child_of_node(np
, child
)
1301 if (!strcmp(child
->type
, "cache"))
1308 * of_node_get - Increment refcount of a node
1309 * @node: Node to inc refcount, NULL is supported to
1310 * simplify writing of callers
1314 struct device_node
*of_node_get(struct device_node
*node
)
1317 kref_get(&node
->kref
);
1320 EXPORT_SYMBOL(of_node_get
);
1322 static inline struct device_node
* kref_to_device_node(struct kref
*kref
)
1324 return container_of(kref
, struct device_node
, kref
);
1328 * of_node_release - release a dynamically allocated node
1329 * @kref: kref element of the node to be released
1331 * In of_node_put() this function is passed to kref_put()
1332 * as the destructor.
1334 static void of_node_release(struct kref
*kref
)
1336 struct device_node
*node
= kref_to_device_node(kref
);
1337 struct property
*prop
= node
->properties
;
1339 /* We should never be releasing nodes that haven't been detached. */
1340 if (!of_node_check_flag(node
, OF_DETACHED
)) {
1341 printk("WARNING: Bad of_node_put() on %s\n", node
->full_name
);
1343 kref_init(&node
->kref
);
1347 if (!of_node_check_flag(node
, OF_DYNAMIC
))
1351 struct property
*next
= prop
->next
;
1358 prop
= node
->deadprops
;
1359 node
->deadprops
= NULL
;
1362 kfree(node
->full_name
);
1368 * of_node_put - Decrement refcount of a node
1369 * @node: Node to dec refcount, NULL is supported to
1370 * simplify writing of callers
1373 void of_node_put(struct device_node
*node
)
1376 kref_put(&node
->kref
, of_node_release
);
1378 EXPORT_SYMBOL(of_node_put
);
1381 * Plug a device node into the tree and global list.
1383 void of_attach_node(struct device_node
*np
)
1385 unsigned long flags
;
1387 write_lock_irqsave(&devtree_lock
, flags
);
1388 np
->sibling
= np
->parent
->child
;
1389 np
->allnext
= allnodes
;
1390 np
->parent
->child
= np
;
1392 write_unlock_irqrestore(&devtree_lock
, flags
);
1396 * "Unplug" a node from the device tree. The caller must hold
1397 * a reference to the node. The memory associated with the node
1398 * is not freed until its refcount goes to zero.
1400 void of_detach_node(struct device_node
*np
)
1402 struct device_node
*parent
;
1403 unsigned long flags
;
1405 write_lock_irqsave(&devtree_lock
, flags
);
1407 parent
= np
->parent
;
1412 allnodes
= np
->allnext
;
1414 struct device_node
*prev
;
1415 for (prev
= allnodes
;
1416 prev
->allnext
!= np
;
1417 prev
= prev
->allnext
)
1419 prev
->allnext
= np
->allnext
;
1422 if (parent
->child
== np
)
1423 parent
->child
= np
->sibling
;
1425 struct device_node
*prevsib
;
1426 for (prevsib
= np
->parent
->child
;
1427 prevsib
->sibling
!= np
;
1428 prevsib
= prevsib
->sibling
)
1430 prevsib
->sibling
= np
->sibling
;
1433 of_node_set_flag(np
, OF_DETACHED
);
1436 write_unlock_irqrestore(&devtree_lock
, flags
);
1439 #ifdef CONFIG_PPC_PSERIES
1441 * Fix up the uninitialized fields in a new device node:
1442 * name, type and pci-specific fields
1445 static int of_finish_dynamic_node(struct device_node
*node
)
1447 struct device_node
*parent
= of_get_parent(node
);
1449 const phandle
*ibm_phandle
;
1451 node
->name
= of_get_property(node
, "name", NULL
);
1452 node
->type
= of_get_property(node
, "device_type", NULL
);
1455 node
->name
= "<NULL>";
1457 node
->type
= "<NULL>";
1464 /* We don't support that function on PowerMac, at least
1467 if (machine_is(powermac
))
1470 /* fix up new node's linux_phandle field */
1471 if ((ibm_phandle
= of_get_property(node
, "ibm,phandle", NULL
)))
1472 node
->linux_phandle
= *ibm_phandle
;
1475 of_node_put(parent
);
1479 static int prom_reconfig_notifier(struct notifier_block
*nb
,
1480 unsigned long action
, void *node
)
1485 case PSERIES_RECONFIG_ADD
:
1486 err
= of_finish_dynamic_node(node
);
1488 printk(KERN_ERR
"finish_node returned %d\n", err
);
1499 static struct notifier_block prom_reconfig_nb
= {
1500 .notifier_call
= prom_reconfig_notifier
,
1501 .priority
= 10, /* This one needs to run first */
1504 static int __init
prom_reconfig_setup(void)
1506 return pSeries_reconfig_notifier_register(&prom_reconfig_nb
);
1508 __initcall(prom_reconfig_setup
);
1512 * Add a property to a node
1514 int prom_add_property(struct device_node
* np
, struct property
* prop
)
1516 struct property
**next
;
1517 unsigned long flags
;
1520 write_lock_irqsave(&devtree_lock
, flags
);
1521 next
= &np
->properties
;
1523 if (strcmp(prop
->name
, (*next
)->name
) == 0) {
1524 /* duplicate ! don't insert it */
1525 write_unlock_irqrestore(&devtree_lock
, flags
);
1528 next
= &(*next
)->next
;
1531 write_unlock_irqrestore(&devtree_lock
, flags
);
1533 #ifdef CONFIG_PROC_DEVICETREE
1534 /* try to add to proc as well if it was initialized */
1536 proc_device_tree_add_prop(np
->pde
, prop
);
1537 #endif /* CONFIG_PROC_DEVICETREE */
1543 * Remove a property from a node. Note that we don't actually
1544 * remove it, since we have given out who-knows-how-many pointers
1545 * to the data using get-property. Instead we just move the property
1546 * to the "dead properties" list, so it won't be found any more.
1548 int prom_remove_property(struct device_node
*np
, struct property
*prop
)
1550 struct property
**next
;
1551 unsigned long flags
;
1554 write_lock_irqsave(&devtree_lock
, flags
);
1555 next
= &np
->properties
;
1557 if (*next
== prop
) {
1558 /* found the node */
1560 prop
->next
= np
->deadprops
;
1561 np
->deadprops
= prop
;
1565 next
= &(*next
)->next
;
1567 write_unlock_irqrestore(&devtree_lock
, flags
);
1572 #ifdef CONFIG_PROC_DEVICETREE
1573 /* try to remove the proc node as well */
1575 proc_device_tree_remove_prop(np
->pde
, prop
);
1576 #endif /* CONFIG_PROC_DEVICETREE */
1582 * Update a property in 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, and add the new property to the
1588 int prom_update_property(struct device_node
*np
,
1589 struct property
*newprop
,
1590 struct property
*oldprop
)
1592 struct property
**next
;
1593 unsigned long flags
;
1596 write_lock_irqsave(&devtree_lock
, flags
);
1597 next
= &np
->properties
;
1599 if (*next
== oldprop
) {
1600 /* found the node */
1601 newprop
->next
= oldprop
->next
;
1603 oldprop
->next
= np
->deadprops
;
1604 np
->deadprops
= oldprop
;
1608 next
= &(*next
)->next
;
1610 write_unlock_irqrestore(&devtree_lock
, flags
);
1615 #ifdef CONFIG_PROC_DEVICETREE
1616 /* try to add to proc as well if it was initialized */
1618 proc_device_tree_update_prop(np
->pde
, newprop
, oldprop
);
1619 #endif /* CONFIG_PROC_DEVICETREE */
1625 /* Find the device node for a given logical cpu number, also returns the cpu
1626 * local thread number (index in ibm,interrupt-server#s) if relevant and
1627 * asked for (non NULL)
1629 struct device_node
*of_get_cpu_node(int cpu
, unsigned int *thread
)
1632 struct device_node
*np
;
1634 hardid
= get_hard_smp_processor_id(cpu
);
1636 for_each_node_by_type(np
, "cpu") {
1638 unsigned int plen
, t
;
1640 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
1641 * fallback to "reg" property and assume no threads
1643 intserv
= of_get_property(np
, "ibm,ppc-interrupt-server#s",
1645 if (intserv
== NULL
) {
1646 const u32
*reg
= of_get_property(np
, "reg", NULL
);
1649 if (*reg
== hardid
) {
1655 plen
/= sizeof(u32
);
1656 for (t
= 0; t
< plen
; t
++) {
1657 if (hardid
== intserv
[t
]) {
1667 EXPORT_SYMBOL(of_get_cpu_node
);
1669 #if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
1670 static struct debugfs_blob_wrapper flat_dt_blob
;
1672 static int __init
export_flat_device_tree(void)
1676 flat_dt_blob
.data
= initial_boot_params
;
1677 flat_dt_blob
.size
= initial_boot_params
->totalsize
;
1679 d
= debugfs_create_blob("flat-device-tree", S_IFREG
| S_IRUSR
,
1680 powerpc_debugfs_root
, &flat_dt_blob
);
1686 __initcall(export_flat_device_tree
);