2 * Procedures for interfacing to Open Firmware.
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/proc_fs.h>
27 #include <linux/stringify.h>
28 #include <linux/delay.h>
29 #include <linux/initrd.h>
30 #include <linux/bitops.h>
34 #include <asm/processor.h>
39 #include <asm/pgtable.h>
41 #include <asm/iommu.h>
42 #include <asm/btext.h>
43 #include <asm/sections.h>
44 #include <asm/machdep.h>
47 #include <linux/linux_logo.h>
50 * Eventually bump that one up
52 #define DEVTREE_CHUNK_SIZE 0x100000
55 * This is the size of the local memory reserve map that gets copied
56 * into the boot params passed to the kernel. That size is totally
57 * flexible as the kernel just reads the list until it encounters an
58 * entry with size 0, so it can be changed without breaking binary
61 #define MEM_RESERVE_MAP_SIZE 8
64 * prom_init() is called very early on, before the kernel text
65 * and data have been mapped to KERNELBASE. At this point the code
66 * is running at whatever address it has been loaded at.
67 * On ppc32 we compile with -mrelocatable, which means that references
68 * to extern and static variables get relocated automatically.
69 * ppc64 objects are always relocatable, we just need to relocate the
72 * Because OF may have mapped I/O devices into the area starting at
73 * KERNELBASE, particularly on CHRP machines, we can't safely call
74 * OF once the kernel has been mapped to KERNELBASE. Therefore all
75 * OF calls must be done within prom_init().
77 * ADDR is used in calls to call_prom. The 4th and following
78 * arguments to call_prom should be 32-bit values.
79 * On ppc64, 64 bit values are truncated to 32 bits (and
80 * fortunately don't get interpreted as two arguments).
82 #define ADDR(x) (u32)(unsigned long)(x)
85 #define OF_WORKAROUNDS 0
87 #define OF_WORKAROUNDS of_workarounds
91 #define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
92 #define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
94 #define PROM_BUG() do { \
95 prom_printf("kernel BUG at %s line 0x%x!\n", \
96 __FILE__, __LINE__); \
97 __asm__ __volatile__(".long " BUG_ILLEGAL_INSTR); \
101 #define prom_debug(x...) prom_printf(x)
103 #define prom_debug(x...)
107 typedef u32 prom_arg_t
;
125 struct mem_map_entry
{
130 typedef __be32 cell_t
;
132 extern void __start(unsigned long r3
, unsigned long r4
, unsigned long r5
,
133 unsigned long r6
, unsigned long r7
, unsigned long r8
,
137 extern int enter_prom(struct prom_args
*args
, unsigned long entry
);
139 static inline int enter_prom(struct prom_args
*args
, unsigned long entry
)
141 return ((int (*)(struct prom_args
*))entry
)(args
);
145 extern void copy_and_flush(unsigned long dest
, unsigned long src
,
146 unsigned long size
, unsigned long offset
);
149 static struct prom_t __initdata prom
;
151 static unsigned long prom_entry __initdata
;
153 #define PROM_SCRATCH_SIZE 256
155 static char __initdata of_stdout_device
[256];
156 static char __initdata prom_scratch
[PROM_SCRATCH_SIZE
];
158 static unsigned long __initdata dt_header_start
;
159 static unsigned long __initdata dt_struct_start
, dt_struct_end
;
160 static unsigned long __initdata dt_string_start
, dt_string_end
;
162 static unsigned long __initdata prom_initrd_start
, prom_initrd_end
;
165 static int __initdata prom_iommu_force_on
;
166 static int __initdata prom_iommu_off
;
167 static unsigned long __initdata prom_tce_alloc_start
;
168 static unsigned long __initdata prom_tce_alloc_end
;
171 /* Platforms codes are now obsolete in the kernel. Now only used within this
172 * file and ultimately gone too. Feel free to change them if you need, they
173 * are not shared with anything outside of this file anymore
175 #define PLATFORM_PSERIES 0x0100
176 #define PLATFORM_PSERIES_LPAR 0x0101
177 #define PLATFORM_LPAR 0x0001
178 #define PLATFORM_POWERMAC 0x0400
179 #define PLATFORM_GENERIC 0x0500
180 #define PLATFORM_OPAL 0x0600
182 static int __initdata of_platform
;
184 static char __initdata prom_cmd_line
[COMMAND_LINE_SIZE
];
186 static unsigned long __initdata prom_memory_limit
;
188 static unsigned long __initdata alloc_top
;
189 static unsigned long __initdata alloc_top_high
;
190 static unsigned long __initdata alloc_bottom
;
191 static unsigned long __initdata rmo_top
;
192 static unsigned long __initdata ram_top
;
194 static struct mem_map_entry __initdata mem_reserve_map
[MEM_RESERVE_MAP_SIZE
];
195 static int __initdata mem_reserve_cnt
;
197 static cell_t __initdata regbuf
[1024];
199 static bool rtas_has_query_cpu_stopped
;
203 * Error results ... some OF calls will return "-1" on error, some
204 * will return 0, some will return either. To simplify, here are
205 * macros to use with any ihandle or phandle return value to check if
209 #define PROM_ERROR (-1u)
210 #define PHANDLE_VALID(p) ((p) != 0 && (p) != PROM_ERROR)
211 #define IHANDLE_VALID(i) ((i) != 0 && (i) != PROM_ERROR)
214 /* This is the one and *ONLY* place where we actually call open
218 static int __init
call_prom(const char *service
, int nargs
, int nret
, ...)
221 struct prom_args args
;
224 args
.service
= cpu_to_be32(ADDR(service
));
225 args
.nargs
= cpu_to_be32(nargs
);
226 args
.nret
= cpu_to_be32(nret
);
228 va_start(list
, nret
);
229 for (i
= 0; i
< nargs
; i
++)
230 args
.args
[i
] = cpu_to_be32(va_arg(list
, prom_arg_t
));
233 for (i
= 0; i
< nret
; i
++)
234 args
.args
[nargs
+i
] = 0;
236 if (enter_prom(&args
, prom_entry
) < 0)
239 return (nret
> 0) ? be32_to_cpu(args
.args
[nargs
]) : 0;
242 static int __init
call_prom_ret(const char *service
, int nargs
, int nret
,
243 prom_arg_t
*rets
, ...)
246 struct prom_args args
;
249 args
.service
= cpu_to_be32(ADDR(service
));
250 args
.nargs
= cpu_to_be32(nargs
);
251 args
.nret
= cpu_to_be32(nret
);
253 va_start(list
, rets
);
254 for (i
= 0; i
< nargs
; i
++)
255 args
.args
[i
] = cpu_to_be32(va_arg(list
, prom_arg_t
));
258 for (i
= 0; i
< nret
; i
++)
259 args
.args
[nargs
+i
] = 0;
261 if (enter_prom(&args
, prom_entry
) < 0)
265 for (i
= 1; i
< nret
; ++i
)
266 rets
[i
-1] = be32_to_cpu(args
.args
[nargs
+i
]);
268 return (nret
> 0) ? be32_to_cpu(args
.args
[nargs
]) : 0;
272 static void __init
prom_print(const char *msg
)
276 if (prom
.stdout
== 0)
279 for (p
= msg
; *p
!= 0; p
= q
) {
280 for (q
= p
; *q
!= 0 && *q
!= '\n'; ++q
)
283 call_prom("write", 3, 1, prom
.stdout
, p
, q
- p
);
287 call_prom("write", 3, 1, prom
.stdout
, ADDR("\r\n"), 2);
292 static void __init
prom_print_hex(unsigned long val
)
294 int i
, nibbles
= sizeof(val
)*2;
295 char buf
[sizeof(val
)*2+1];
297 for (i
= nibbles
-1; i
>= 0; i
--) {
298 buf
[i
] = (val
& 0xf) + '0';
300 buf
[i
] += ('a'-'0'-10);
304 call_prom("write", 3, 1, prom
.stdout
, buf
, nibbles
);
307 /* max number of decimal digits in an unsigned long */
309 static void __init
prom_print_dec(unsigned long val
)
312 char buf
[UL_DIGITS
+1];
314 for (i
= UL_DIGITS
-1; i
>= 0; i
--) {
315 buf
[i
] = (val
% 10) + '0';
320 /* shift stuff down */
321 size
= UL_DIGITS
- i
;
322 call_prom("write", 3, 1, prom
.stdout
, buf
+i
, size
);
325 static void __init
prom_printf(const char *format
, ...)
327 const char *p
, *q
, *s
;
332 va_start(args
, format
);
333 for (p
= format
; *p
!= 0; p
= q
) {
334 for (q
= p
; *q
!= 0 && *q
!= '\n' && *q
!= '%'; ++q
)
337 call_prom("write", 3, 1, prom
.stdout
, p
, q
- p
);
342 call_prom("write", 3, 1, prom
.stdout
,
352 s
= va_arg(args
, const char *);
357 v
= va_arg(args
, unsigned long);
362 vs
= va_arg(args
, int);
373 else if (*q
== 'x') {
375 v
= va_arg(args
, unsigned long);
377 } else if (*q
== 'u') { /* '%lu' */
379 v
= va_arg(args
, unsigned long);
381 } else if (*q
== 'd') { /* %ld */
383 vs
= va_arg(args
, long);
396 static unsigned int __init
prom_claim(unsigned long virt
, unsigned long size
,
400 if (align
== 0 && (OF_WORKAROUNDS
& OF_WA_CLAIM
)) {
402 * Old OF requires we claim physical and virtual separately
403 * and then map explicitly (assuming virtual mode)
408 ret
= call_prom_ret("call-method", 5, 2, &result
,
409 ADDR("claim"), prom
.memory
,
411 if (ret
!= 0 || result
== -1)
413 ret
= call_prom_ret("call-method", 5, 2, &result
,
414 ADDR("claim"), prom
.mmumap
,
417 call_prom("call-method", 4, 1, ADDR("release"),
418 prom
.memory
, size
, virt
);
421 /* the 0x12 is M (coherence) + PP == read/write */
422 call_prom("call-method", 6, 1,
423 ADDR("map"), prom
.mmumap
, 0x12, size
, virt
, virt
);
426 return call_prom("claim", 3, 1, (prom_arg_t
)virt
, (prom_arg_t
)size
,
430 static void __init
__attribute__((noreturn
)) prom_panic(const char *reason
)
433 /* Do not call exit because it clears the screen on pmac
434 * it also causes some sort of double-fault on early pmacs */
435 if (of_platform
== PLATFORM_POWERMAC
)
438 /* ToDo: should put up an SRC here on pSeries */
439 call_prom("exit", 0, 0);
441 for (;;) /* should never get here */
446 static int __init
prom_next_node(phandle
*nodep
)
450 if ((node
= *nodep
) != 0
451 && (*nodep
= call_prom("child", 1, 1, node
)) != 0)
453 if ((*nodep
= call_prom("peer", 1, 1, node
)) != 0)
456 if ((node
= call_prom("parent", 1, 1, node
)) == 0)
458 if ((*nodep
= call_prom("peer", 1, 1, node
)) != 0)
463 static int inline prom_getprop(phandle node
, const char *pname
,
464 void *value
, size_t valuelen
)
466 return call_prom("getprop", 4, 1, node
, ADDR(pname
),
467 (u32
)(unsigned long) value
, (u32
) valuelen
);
470 static int inline prom_getproplen(phandle node
, const char *pname
)
472 return call_prom("getproplen", 2, 1, node
, ADDR(pname
));
475 static void add_string(char **str
, const char *q
)
485 static char *tohex(unsigned int x
)
487 static char digits
[] = "0123456789abcdef";
488 static char result
[9];
495 result
[i
] = digits
[x
& 0xf];
497 } while (x
!= 0 && i
> 0);
501 static int __init
prom_setprop(phandle node
, const char *nodename
,
502 const char *pname
, void *value
, size_t valuelen
)
506 if (!(OF_WORKAROUNDS
& OF_WA_LONGTRAIL
))
507 return call_prom("setprop", 4, 1, node
, ADDR(pname
),
508 (u32
)(unsigned long) value
, (u32
) valuelen
);
510 /* gah... setprop doesn't work on longtrail, have to use interpret */
512 add_string(&p
, "dev");
513 add_string(&p
, nodename
);
514 add_string(&p
, tohex((u32
)(unsigned long) value
));
515 add_string(&p
, tohex(valuelen
));
516 add_string(&p
, tohex(ADDR(pname
)));
517 add_string(&p
, tohex(strlen(pname
)));
518 add_string(&p
, "property");
520 return call_prom("interpret", 1, 1, (u32
)(unsigned long) cmd
);
523 /* We can't use the standard versions because of relocation headaches. */
524 #define isxdigit(c) (('0' <= (c) && (c) <= '9') \
525 || ('a' <= (c) && (c) <= 'f') \
526 || ('A' <= (c) && (c) <= 'F'))
528 #define isdigit(c) ('0' <= (c) && (c) <= '9')
529 #define islower(c) ('a' <= (c) && (c) <= 'z')
530 #define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
532 static unsigned long prom_strtoul(const char *cp
, const char **endp
)
534 unsigned long result
= 0, base
= 10, value
;
539 if (toupper(*cp
) == 'X') {
545 while (isxdigit(*cp
) &&
546 (value
= isdigit(*cp
) ? *cp
- '0' : toupper(*cp
) - 'A' + 10) < base
) {
547 result
= result
* base
+ value
;
557 static unsigned long prom_memparse(const char *ptr
, const char **retptr
)
559 unsigned long ret
= prom_strtoul(ptr
, retptr
);
563 * We can't use a switch here because GCC *may* generate a
564 * jump table which won't work, because we're not running at
565 * the address we're linked at.
567 if ('G' == **retptr
|| 'g' == **retptr
)
570 if ('M' == **retptr
|| 'm' == **retptr
)
573 if ('K' == **retptr
|| 'k' == **retptr
)
585 * Early parsing of the command line passed to the kernel, used for
586 * "mem=x" and the options that affect the iommu
588 static void __init
early_cmdline_parse(void)
595 prom_cmd_line
[0] = 0;
597 if ((long)prom
.chosen
> 0)
598 l
= prom_getprop(prom
.chosen
, "bootargs", p
, COMMAND_LINE_SIZE
-1);
599 #ifdef CONFIG_CMDLINE
600 if (l
<= 0 || p
[0] == '\0') /* dbl check */
601 strlcpy(prom_cmd_line
,
602 CONFIG_CMDLINE
, sizeof(prom_cmd_line
));
603 #endif /* CONFIG_CMDLINE */
604 prom_printf("command line: %s\n", prom_cmd_line
);
607 opt
= strstr(prom_cmd_line
, "iommu=");
609 prom_printf("iommu opt is: %s\n", opt
);
611 while (*opt
&& *opt
== ' ')
613 if (!strncmp(opt
, "off", 3))
615 else if (!strncmp(opt
, "force", 5))
616 prom_iommu_force_on
= 1;
619 opt
= strstr(prom_cmd_line
, "mem=");
622 prom_memory_limit
= prom_memparse(opt
, (const char **)&opt
);
624 /* Align to 16 MB == size of ppc64 large page */
625 prom_memory_limit
= ALIGN(prom_memory_limit
, 0x1000000);
630 #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
632 * The architecture vector has an array of PVR mask/value pairs,
633 * followed by # option vectors - 1, followed by the option vectors.
635 * See prom.h for the definition of the bits specified in the
636 * architecture vector.
638 * Because the description vector contains a mix of byte and word
639 * values, we declare it as an unsigned char array, and use this
640 * macro to put word values in.
642 #define W(x) ((x) >> 24) & 0xff, ((x) >> 16) & 0xff, \
643 ((x) >> 8) & 0xff, (x) & 0xff
645 unsigned char ibm_architecture_vec
[] = {
646 W(0xfffe0000), W(0x003a0000), /* POWER5/POWER5+ */
647 W(0xffff0000), W(0x003e0000), /* POWER6 */
648 W(0xffff0000), W(0x003f0000), /* POWER7 */
649 W(0xffff0000), W(0x004b0000), /* POWER8E */
650 W(0xffff0000), W(0x004d0000), /* POWER8 */
651 W(0xffffffff), W(0x0f000004), /* all 2.07-compliant */
652 W(0xffffffff), W(0x0f000003), /* all 2.06-compliant */
653 W(0xffffffff), W(0x0f000002), /* all 2.05-compliant */
654 W(0xfffffffe), W(0x0f000001), /* all 2.04-compliant and earlier */
655 6 - 1, /* 6 option vectors */
657 /* option vector 1: processor architectures supported */
659 0, /* don't ignore, don't halt */
660 OV1_PPC_2_00
| OV1_PPC_2_01
| OV1_PPC_2_02
| OV1_PPC_2_03
|
661 OV1_PPC_2_04
| OV1_PPC_2_05
| OV1_PPC_2_06
| OV1_PPC_2_07
,
663 /* option vector 2: Open Firmware options supported */
667 W(0xffffffff), /* real_base */
668 W(0xffffffff), /* real_size */
669 W(0xffffffff), /* virt_base */
670 W(0xffffffff), /* virt_size */
671 W(0xffffffff), /* load_base */
672 W(256), /* 256MB min RMA */
673 W(0xffffffff), /* full client load */
674 0, /* min RMA percentage of total RAM */
675 48, /* max log_2(hash table size) */
677 /* option vector 3: processor options supported */
679 0, /* don't ignore, don't halt */
680 OV3_FP
| OV3_VMX
| OV3_DFP
,
682 /* option vector 4: IBM PAPR implementation */
685 OV4_MIN_ENT_CAP
, /* minimum VP entitled capacity */
687 /* option vector 5: PAPR/OF options */
689 0, /* don't ignore, don't halt */
690 OV5_FEAT(OV5_LPAR
) | OV5_FEAT(OV5_SPLPAR
) | OV5_FEAT(OV5_LARGE_PAGES
) |
691 OV5_FEAT(OV5_DRCONF_MEMORY
) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU
) |
692 #ifdef CONFIG_PCI_MSI
693 /* PCIe/MSI support. Without MSI full PCIe is not supported */
699 #ifdef CONFIG_PPC_SMLPAR
700 OV5_FEAT(OV5_CMO
) | OV5_FEAT(OV5_XCMO
),
704 OV5_FEAT(OV5_TYPE1_AFFINITY
) | OV5_FEAT(OV5_PRRN
),
708 /* WARNING: The offset of the "number of cores" field below
709 * must match by the macro below. Update the definition if
710 * the structure layout changes.
712 #define IBM_ARCH_VEC_NRCORES_OFFSET 125
713 W(NR_CPUS
), /* number of cores supported */
718 OV5_FEAT(OV5_PFO_HW_RNG
) | OV5_FEAT(OV5_PFO_HW_ENCR
) |
719 OV5_FEAT(OV5_PFO_HW_842
),
720 OV5_FEAT(OV5_SUB_PROCESSORS
),
721 /* option vector 6: IBM PAPR hints */
729 /* Old method - ELF header with PT_NOTE sections only works on BE */
730 #ifdef __BIG_ENDIAN__
731 static struct fake_elf
{
738 char name
[8]; /* "PowerPC" */
752 char name
[24]; /* "IBM,RPA-Client-Config" */
766 .e_ident
= { 0x7f, 'E', 'L', 'F',
767 ELFCLASS32
, ELFDATA2MSB
, EV_CURRENT
},
768 .e_type
= ET_EXEC
, /* yeah right */
770 .e_version
= EV_CURRENT
,
771 .e_phoff
= offsetof(struct fake_elf
, phdr
),
772 .e_phentsize
= sizeof(Elf32_Phdr
),
778 .p_offset
= offsetof(struct fake_elf
, chrpnote
),
779 .p_filesz
= sizeof(struct chrpnote
)
782 .p_offset
= offsetof(struct fake_elf
, rpanote
),
783 .p_filesz
= sizeof(struct rpanote
)
787 .namesz
= sizeof("PowerPC"),
788 .descsz
= sizeof(struct chrpdesc
),
792 .real_mode
= ~0U, /* ~0 means "don't care" */
801 .namesz
= sizeof("IBM,RPA-Client-Config"),
802 .descsz
= sizeof(struct rpadesc
),
804 .name
= "IBM,RPA-Client-Config",
807 .min_rmo_size
= 64, /* in megabytes */
808 .min_rmo_percent
= 0,
809 .max_pft_size
= 48, /* 2^48 bytes max PFT size */
816 #endif /* __BIG_ENDIAN__ */
818 static int __init
prom_count_smt_threads(void)
824 /* Pick up th first CPU node we can find */
825 for (node
= 0; prom_next_node(&node
); ) {
827 prom_getprop(node
, "device_type", type
, sizeof(type
));
829 if (strcmp(type
, "cpu"))
832 * There is an entry for each smt thread, each entry being
833 * 4 bytes long. All cpus should have the same number of
834 * smt threads, so return after finding the first.
836 plen
= prom_getproplen(node
, "ibm,ppc-interrupt-server#s");
837 if (plen
== PROM_ERROR
)
840 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen
);
843 if (plen
< 1 || plen
> 64) {
844 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
845 (unsigned long)plen
);
850 prom_debug("No threads found, assuming 1 per core\n");
857 static void __init
prom_send_capabilities(void)
862 unsigned char *ptcores
;
864 root
= call_prom("open", 1, 1, ADDR("/"));
866 /* We need to tell the FW about the number of cores we support.
868 * To do that, we count the number of threads on the first core
869 * (we assume this is the same for all cores) and use it to
873 /* The core value may start at an odd address. If such a word
874 * access is made at a cache line boundary, this leads to an
875 * exception which may not be handled at this time.
876 * Forcing a per byte access to avoid exception.
878 ptcores
= &ibm_architecture_vec
[IBM_ARCH_VEC_NRCORES_OFFSET
];
880 cores
|= ptcores
[0] << 24;
881 cores
|= ptcores
[1] << 16;
882 cores
|= ptcores
[2] << 8;
884 if (cores
!= NR_CPUS
) {
885 prom_printf("WARNING ! "
886 "ibm_architecture_vec structure inconsistent: %lu!\n",
889 cores
= DIV_ROUND_UP(NR_CPUS
, prom_count_smt_threads());
890 prom_printf("Max number of cores passed to firmware: %lu (NR_CPUS = %lu)\n",
892 ptcores
[0] = (cores
>> 24) & 0xff;
893 ptcores
[1] = (cores
>> 16) & 0xff;
894 ptcores
[2] = (cores
>> 8) & 0xff;
895 ptcores
[3] = cores
& 0xff;
898 /* try calling the ibm,client-architecture-support method */
899 prom_printf("Calling ibm,client-architecture-support...");
900 if (call_prom_ret("call-method", 3, 2, &ret
,
901 ADDR("ibm,client-architecture-support"),
903 ADDR(ibm_architecture_vec
)) == 0) {
904 /* the call exists... */
906 prom_printf("\nWARNING: ibm,client-architecture"
907 "-support call FAILED!\n");
908 call_prom("close", 1, 0, root
);
909 prom_printf(" done\n");
912 call_prom("close", 1, 0, root
);
913 prom_printf(" not implemented\n");
916 #ifdef __BIG_ENDIAN__
920 /* no ibm,client-architecture-support call, try the old way */
921 elfloader
= call_prom("open", 1, 1,
922 ADDR("/packages/elf-loader"));
923 if (elfloader
== 0) {
924 prom_printf("couldn't open /packages/elf-loader\n");
927 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
928 elfloader
, ADDR(&fake_elf
));
929 call_prom("close", 1, 0, elfloader
);
931 #endif /* __BIG_ENDIAN__ */
933 #endif /* #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV) */
936 * Memory allocation strategy... our layout is normally:
938 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
939 * rare cases, initrd might end up being before the kernel though.
940 * We assume this won't override the final kernel at 0, we have no
941 * provision to handle that in this version, but it should hopefully
944 * alloc_top is set to the top of RMO, eventually shrink down if the
947 * alloc_bottom is set to the top of kernel/initrd
949 * from there, allocations are done this way : rtas is allocated
950 * topmost, and the device-tree is allocated from the bottom. We try
951 * to grow the device-tree allocation as we progress. If we can't,
952 * then we fail, we don't currently have a facility to restart
953 * elsewhere, but that shouldn't be necessary.
955 * Note that calls to reserve_mem have to be done explicitly, memory
956 * allocated with either alloc_up or alloc_down isn't automatically
962 * Allocates memory in the RMO upward from the kernel/initrd
964 * When align is 0, this is a special case, it means to allocate in place
965 * at the current location of alloc_bottom or fail (that is basically
966 * extending the previous allocation). Used for the device-tree flattening
968 static unsigned long __init
alloc_up(unsigned long size
, unsigned long align
)
970 unsigned long base
= alloc_bottom
;
971 unsigned long addr
= 0;
974 base
= _ALIGN_UP(base
, align
);
975 prom_debug("alloc_up(%x, %x)\n", size
, align
);
977 prom_panic("alloc_up() called with mem not initialized\n");
980 base
= _ALIGN_UP(alloc_bottom
, align
);
984 for(; (base
+ size
) <= alloc_top
;
985 base
= _ALIGN_UP(base
+ 0x100000, align
)) {
986 prom_debug(" trying: 0x%x\n\r", base
);
987 addr
= (unsigned long)prom_claim(base
, size
, 0);
988 if (addr
!= PROM_ERROR
&& addr
!= 0)
996 alloc_bottom
= addr
+ size
;
998 prom_debug(" -> %x\n", addr
);
999 prom_debug(" alloc_bottom : %x\n", alloc_bottom
);
1000 prom_debug(" alloc_top : %x\n", alloc_top
);
1001 prom_debug(" alloc_top_hi : %x\n", alloc_top_high
);
1002 prom_debug(" rmo_top : %x\n", rmo_top
);
1003 prom_debug(" ram_top : %x\n", ram_top
);
1009 * Allocates memory downward, either from top of RMO, or if highmem
1010 * is set, from the top of RAM. Note that this one doesn't handle
1011 * failures. It does claim memory if highmem is not set.
1013 static unsigned long __init
alloc_down(unsigned long size
, unsigned long align
,
1016 unsigned long base
, addr
= 0;
1018 prom_debug("alloc_down(%x, %x, %s)\n", size
, align
,
1019 highmem
? "(high)" : "(low)");
1021 prom_panic("alloc_down() called with mem not initialized\n");
1024 /* Carve out storage for the TCE table. */
1025 addr
= _ALIGN_DOWN(alloc_top_high
- size
, align
);
1026 if (addr
<= alloc_bottom
)
1028 /* Will we bump into the RMO ? If yes, check out that we
1029 * didn't overlap existing allocations there, if we did,
1030 * we are dead, we must be the first in town !
1032 if (addr
< rmo_top
) {
1033 /* Good, we are first */
1034 if (alloc_top
== rmo_top
)
1035 alloc_top
= rmo_top
= addr
;
1039 alloc_top_high
= addr
;
1043 base
= _ALIGN_DOWN(alloc_top
- size
, align
);
1044 for (; base
> alloc_bottom
;
1045 base
= _ALIGN_DOWN(base
- 0x100000, align
)) {
1046 prom_debug(" trying: 0x%x\n\r", base
);
1047 addr
= (unsigned long)prom_claim(base
, size
, 0);
1048 if (addr
!= PROM_ERROR
&& addr
!= 0)
1057 prom_debug(" -> %x\n", addr
);
1058 prom_debug(" alloc_bottom : %x\n", alloc_bottom
);
1059 prom_debug(" alloc_top : %x\n", alloc_top
);
1060 prom_debug(" alloc_top_hi : %x\n", alloc_top_high
);
1061 prom_debug(" rmo_top : %x\n", rmo_top
);
1062 prom_debug(" ram_top : %x\n", ram_top
);
1068 * Parse a "reg" cell
1070 static unsigned long __init
prom_next_cell(int s
, cell_t
**cellp
)
1073 unsigned long r
= 0;
1075 /* Ignore more than 2 cells */
1076 while (s
> sizeof(unsigned long) / 4) {
1080 r
= be32_to_cpu(*p
++);
1084 r
|= be32_to_cpu(*(p
++));
1092 * Very dumb function for adding to the memory reserve list, but
1093 * we don't need anything smarter at this point
1095 * XXX Eventually check for collisions. They should NEVER happen.
1096 * If problems seem to show up, it would be a good start to track
1099 static void __init
reserve_mem(u64 base
, u64 size
)
1101 u64 top
= base
+ size
;
1102 unsigned long cnt
= mem_reserve_cnt
;
1107 /* We need to always keep one empty entry so that we
1108 * have our terminator with "size" set to 0 since we are
1109 * dumb and just copy this entire array to the boot params
1111 base
= _ALIGN_DOWN(base
, PAGE_SIZE
);
1112 top
= _ALIGN_UP(top
, PAGE_SIZE
);
1115 if (cnt
>= (MEM_RESERVE_MAP_SIZE
- 1))
1116 prom_panic("Memory reserve map exhausted !\n");
1117 mem_reserve_map
[cnt
].base
= cpu_to_be64(base
);
1118 mem_reserve_map
[cnt
].size
= cpu_to_be64(size
);
1119 mem_reserve_cnt
= cnt
+ 1;
1123 * Initialize memory allocation mechanism, parse "memory" nodes and
1124 * obtain that way the top of memory and RMO to setup out local allocator
1126 static void __init
prom_init_mem(void)
1129 char *path
, type
[64];
1136 * We iterate the memory nodes to find
1137 * 1) top of RMO (first node)
1140 val
= cpu_to_be32(2);
1141 prom_getprop(prom
.root
, "#address-cells", &val
, sizeof(val
));
1142 rac
= be32_to_cpu(val
);
1143 val
= cpu_to_be32(1);
1144 prom_getprop(prom
.root
, "#size-cells", &val
, sizeof(rsc
));
1145 rsc
= be32_to_cpu(val
);
1146 prom_debug("root_addr_cells: %x\n", rac
);
1147 prom_debug("root_size_cells: %x\n", rsc
);
1149 prom_debug("scanning memory:\n");
1150 path
= prom_scratch
;
1152 for (node
= 0; prom_next_node(&node
); ) {
1154 prom_getprop(node
, "device_type", type
, sizeof(type
));
1158 * CHRP Longtrail machines have no device_type
1159 * on the memory node, so check the name instead...
1161 prom_getprop(node
, "name", type
, sizeof(type
));
1163 if (strcmp(type
, "memory"))
1166 plen
= prom_getprop(node
, "reg", regbuf
, sizeof(regbuf
));
1167 if (plen
> sizeof(regbuf
)) {
1168 prom_printf("memory node too large for buffer !\n");
1169 plen
= sizeof(regbuf
);
1172 endp
= p
+ (plen
/ sizeof(cell_t
));
1175 memset(path
, 0, PROM_SCRATCH_SIZE
);
1176 call_prom("package-to-path", 3, 1, node
, path
, PROM_SCRATCH_SIZE
-1);
1177 prom_debug(" node %s :\n", path
);
1178 #endif /* DEBUG_PROM */
1180 while ((endp
- p
) >= (rac
+ rsc
)) {
1181 unsigned long base
, size
;
1183 base
= prom_next_cell(rac
, &p
);
1184 size
= prom_next_cell(rsc
, &p
);
1188 prom_debug(" %x %x\n", base
, size
);
1189 if (base
== 0 && (of_platform
& PLATFORM_LPAR
))
1191 if ((base
+ size
) > ram_top
)
1192 ram_top
= base
+ size
;
1196 alloc_bottom
= PAGE_ALIGN((unsigned long)&_end
+ 0x4000);
1199 * If prom_memory_limit is set we reduce the upper limits *except* for
1200 * alloc_top_high. This must be the real top of RAM so we can put
1204 alloc_top_high
= ram_top
;
1206 if (prom_memory_limit
) {
1207 if (prom_memory_limit
<= alloc_bottom
) {
1208 prom_printf("Ignoring mem=%x <= alloc_bottom.\n",
1210 prom_memory_limit
= 0;
1211 } else if (prom_memory_limit
>= ram_top
) {
1212 prom_printf("Ignoring mem=%x >= ram_top.\n",
1214 prom_memory_limit
= 0;
1216 ram_top
= prom_memory_limit
;
1217 rmo_top
= min(rmo_top
, prom_memory_limit
);
1222 * Setup our top alloc point, that is top of RMO or top of
1223 * segment 0 when running non-LPAR.
1224 * Some RS64 machines have buggy firmware where claims up at
1225 * 1GB fail. Cap at 768MB as a workaround.
1226 * Since 768MB is plenty of room, and we need to cap to something
1227 * reasonable on 32-bit, cap at 768MB on all machines.
1231 rmo_top
= min(0x30000000ul
, rmo_top
);
1232 alloc_top
= rmo_top
;
1233 alloc_top_high
= ram_top
;
1236 * Check if we have an initrd after the kernel but still inside
1237 * the RMO. If we do move our bottom point to after it.
1239 if (prom_initrd_start
&&
1240 prom_initrd_start
< rmo_top
&&
1241 prom_initrd_end
> alloc_bottom
)
1242 alloc_bottom
= PAGE_ALIGN(prom_initrd_end
);
1244 prom_printf("memory layout at init:\n");
1245 prom_printf(" memory_limit : %x (16 MB aligned)\n", prom_memory_limit
);
1246 prom_printf(" alloc_bottom : %x\n", alloc_bottom
);
1247 prom_printf(" alloc_top : %x\n", alloc_top
);
1248 prom_printf(" alloc_top_hi : %x\n", alloc_top_high
);
1249 prom_printf(" rmo_top : %x\n", rmo_top
);
1250 prom_printf(" ram_top : %x\n", ram_top
);
1253 static void __init
prom_close_stdin(void)
1258 if (prom_getprop(prom
.chosen
, "stdin", &val
, sizeof(val
)) > 0) {
1259 stdin
= be32_to_cpu(val
);
1260 call_prom("close", 1, 0, stdin
);
1264 #ifdef CONFIG_PPC_POWERNV
1266 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1267 static u64 __initdata prom_opal_base
;
1268 static u64 __initdata prom_opal_entry
;
1271 #ifdef __BIG_ENDIAN__
1272 /* XXX Don't change this structure without updating opal-takeover.S */
1273 static struct opal_secondary_data
{
1276 struct opal_takeover_args args
; /* 16 */
1277 } opal_secondary_data
;
1279 static u64 __initdata prom_opal_align
;
1280 static u64 __initdata prom_opal_size
;
1281 static int __initdata prom_rtas_start_cpu
;
1282 static u64 __initdata prom_rtas_data
;
1283 static u64 __initdata prom_rtas_entry
;
1285 extern char opal_secondary_entry
;
1287 static void __init
prom_query_opal(void)
1291 /* We must not query for OPAL presence on a machine that
1292 * supports TNK takeover (970 blades), as this uses the same
1293 * h-call with different arguments and will crash
1295 if (PHANDLE_VALID(call_prom("finddevice", 1, 1,
1296 ADDR("/tnk-memory-map")))) {
1297 prom_printf("TNK takeover detected, skipping OPAL check\n");
1301 prom_printf("Querying for OPAL presence... ");
1303 rc
= opal_query_takeover(&prom_opal_size
,
1305 prom_debug("(rc = %ld) ", rc
);
1307 prom_printf("not there.\n");
1310 of_platform
= PLATFORM_OPAL
;
1311 prom_printf(" there !\n");
1312 prom_debug(" opal_size = 0x%lx\n", prom_opal_size
);
1313 prom_debug(" opal_align = 0x%lx\n", prom_opal_align
);
1314 if (prom_opal_align
< 0x10000)
1315 prom_opal_align
= 0x10000;
1318 static int __init
prom_rtas_call(int token
, int nargs
, int nret
,
1321 struct rtas_args rtas_args
;
1325 rtas_args
.token
= token
;
1326 rtas_args
.nargs
= nargs
;
1327 rtas_args
.nret
= nret
;
1328 rtas_args
.rets
= (rtas_arg_t
*)&(rtas_args
.args
[nargs
]);
1329 va_start(list
, outputs
);
1330 for (i
= 0; i
< nargs
; ++i
)
1331 rtas_args
.args
[i
] = va_arg(list
, rtas_arg_t
);
1334 for (i
= 0; i
< nret
; ++i
)
1335 rtas_args
.rets
[i
] = 0;
1337 opal_enter_rtas(&rtas_args
, prom_rtas_data
,
1340 if (nret
> 1 && outputs
!= NULL
)
1341 for (i
= 0; i
< nret
-1; ++i
)
1342 outputs
[i
] = rtas_args
.rets
[i
+1];
1343 return (nret
> 0)? rtas_args
.rets
[0]: 0;
1346 static void __init
prom_opal_hold_cpus(void)
1348 int i
, cnt
, cpu
, rc
;
1353 void *entry
= (unsigned long *)&opal_secondary_entry
;
1354 struct opal_secondary_data
*data
= &opal_secondary_data
;
1356 prom_debug("prom_opal_hold_cpus: start...\n");
1357 prom_debug(" - entry = 0x%x\n", entry
);
1358 prom_debug(" - data = 0x%x\n", data
);
1364 for (node
= 0; prom_next_node(&node
); ) {
1366 prom_getprop(node
, "device_type", type
, sizeof(type
));
1367 if (strcmp(type
, "cpu") != 0)
1370 /* Skip non-configured cpus. */
1371 if (prom_getprop(node
, "status", type
, sizeof(type
)) > 0)
1372 if (strcmp(type
, "okay") != 0)
1375 cnt
= prom_getprop(node
, "ibm,ppc-interrupt-server#s", servers
,
1377 if (cnt
== PROM_ERROR
)
1380 for (i
= 0; i
< cnt
; i
++) {
1382 prom_debug("CPU %d ... ", cpu
);
1383 if (cpu
== prom
.cpu
) {
1384 prom_debug("booted !\n");
1387 prom_debug("starting ... ");
1389 /* Init the acknowledge var which will be reset by
1390 * the secondary cpu when it awakens from its OF
1394 rc
= prom_rtas_call(prom_rtas_start_cpu
, 3, 1,
1395 NULL
, cpu
, entry
, data
);
1396 prom_debug("rtas rc=%d ...", rc
);
1398 for (j
= 0; j
< 100000000 && data
->ack
== -1; j
++) {
1403 if (data
->ack
!= -1)
1404 prom_debug("done, PIR=0x%x\n", data
->ack
);
1406 prom_debug("timeout !\n");
1409 prom_debug("prom_opal_hold_cpus: end...\n");
1412 static void __init
prom_opal_takeover(void)
1414 struct opal_secondary_data
*data
= &opal_secondary_data
;
1415 struct opal_takeover_args
*args
= &data
->args
;
1416 u64 align
= prom_opal_align
;
1417 u64 top_addr
, opal_addr
;
1419 args
->k_image
= (u64
)_stext
;
1420 args
->k_size
= _end
- _stext
;
1422 args
->k_entry2
= 0x60;
1424 top_addr
= _ALIGN_UP(args
->k_size
, align
);
1426 if (prom_initrd_start
!= 0) {
1427 args
->rd_image
= prom_initrd_start
;
1428 args
->rd_size
= prom_initrd_end
- args
->rd_image
;
1429 args
->rd_loc
= top_addr
;
1430 top_addr
= _ALIGN_UP(args
->rd_loc
+ args
->rd_size
, align
);
1433 /* Pickup an address for the HAL. We want to go really high
1434 * up to avoid problem with future kexecs. On the other hand
1435 * we don't want to be all over the TCEs on P5IOC2 machines
1436 * which are going to be up there too. We assume the machine
1437 * has plenty of memory, and we ask for the HAL for now to
1438 * be just below the 1G point, or above the initrd
1440 opal_addr
= _ALIGN_DOWN(0x40000000 - prom_opal_size
, align
);
1441 if (opal_addr
< top_addr
)
1442 opal_addr
= top_addr
;
1443 args
->hal_addr
= opal_addr
;
1445 /* Copy the command line to the kernel image */
1446 strlcpy(boot_command_line
, prom_cmd_line
,
1449 prom_debug(" k_image = 0x%lx\n", args
->k_image
);
1450 prom_debug(" k_size = 0x%lx\n", args
->k_size
);
1451 prom_debug(" k_entry = 0x%lx\n", args
->k_entry
);
1452 prom_debug(" k_entry2 = 0x%lx\n", args
->k_entry2
);
1453 prom_debug(" hal_addr = 0x%lx\n", args
->hal_addr
);
1454 prom_debug(" rd_image = 0x%lx\n", args
->rd_image
);
1455 prom_debug(" rd_size = 0x%lx\n", args
->rd_size
);
1456 prom_debug(" rd_loc = 0x%lx\n", args
->rd_loc
);
1457 prom_printf("Performing OPAL takeover,this can take a few minutes..\n");
1462 opal_do_takeover(args
);
1464 #endif /* __BIG_ENDIAN__ */
1467 * Allocate room for and instantiate OPAL
1469 static void __init
prom_instantiate_opal(void)
1474 u64 size
= 0, align
= 0x10000;
1478 prom_debug("prom_instantiate_opal: start...\n");
1480 opal_node
= call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1481 prom_debug("opal_node: %x\n", opal_node
);
1482 if (!PHANDLE_VALID(opal_node
))
1486 prom_getprop(opal_node
, "opal-runtime-size", &val64
, sizeof(val64
));
1487 size
= be64_to_cpu(val64
);
1491 prom_getprop(opal_node
, "opal-runtime-alignment", &val64
,sizeof(val64
));
1492 align
= be64_to_cpu(val64
);
1494 base
= alloc_down(size
, align
, 0);
1496 prom_printf("OPAL allocation failed !\n");
1500 opal_inst
= call_prom("open", 1, 1, ADDR("/ibm,opal"));
1501 if (!IHANDLE_VALID(opal_inst
)) {
1502 prom_printf("opening opal package failed (%x)\n", opal_inst
);
1506 prom_printf("instantiating opal at 0x%x...", base
);
1508 if (call_prom_ret("call-method", 4, 3, rets
,
1509 ADDR("load-opal-runtime"),
1511 base
>> 32, base
& 0xffffffff) != 0
1512 || (rets
[0] == 0 && rets
[1] == 0)) {
1513 prom_printf(" failed\n");
1516 entry
= (((u64
)rets
[0]) << 32) | rets
[1];
1518 prom_printf(" done\n");
1520 reserve_mem(base
, size
);
1522 prom_debug("opal base = 0x%x\n", base
);
1523 prom_debug("opal align = 0x%x\n", align
);
1524 prom_debug("opal entry = 0x%x\n", entry
);
1525 prom_debug("opal size = 0x%x\n", (long)size
);
1527 prom_setprop(opal_node
, "/ibm,opal", "opal-base-address",
1528 &base
, sizeof(base
));
1529 prom_setprop(opal_node
, "/ibm,opal", "opal-entry-address",
1530 &entry
, sizeof(entry
));
1532 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1533 prom_opal_base
= base
;
1534 prom_opal_entry
= entry
;
1536 prom_debug("prom_instantiate_opal: end...\n");
1539 #endif /* CONFIG_PPC_POWERNV */
1542 * Allocate room for and instantiate RTAS
1544 static void __init
prom_instantiate_rtas(void)
1548 u32 base
, entry
= 0;
1552 prom_debug("prom_instantiate_rtas: start...\n");
1554 rtas_node
= call_prom("finddevice", 1, 1, ADDR("/rtas"));
1555 prom_debug("rtas_node: %x\n", rtas_node
);
1556 if (!PHANDLE_VALID(rtas_node
))
1560 prom_getprop(rtas_node
, "rtas-size", &val
, sizeof(size
));
1561 size
= be32_to_cpu(val
);
1565 base
= alloc_down(size
, PAGE_SIZE
, 0);
1567 prom_panic("Could not allocate memory for RTAS\n");
1569 rtas_inst
= call_prom("open", 1, 1, ADDR("/rtas"));
1570 if (!IHANDLE_VALID(rtas_inst
)) {
1571 prom_printf("opening rtas package failed (%x)\n", rtas_inst
);
1575 prom_printf("instantiating rtas at 0x%x...", base
);
1577 if (call_prom_ret("call-method", 3, 2, &entry
,
1578 ADDR("instantiate-rtas"),
1579 rtas_inst
, base
) != 0
1581 prom_printf(" failed\n");
1584 prom_printf(" done\n");
1586 reserve_mem(base
, size
);
1588 val
= cpu_to_be32(base
);
1589 prom_setprop(rtas_node
, "/rtas", "linux,rtas-base",
1591 val
= cpu_to_be32(entry
);
1592 prom_setprop(rtas_node
, "/rtas", "linux,rtas-entry",
1595 /* Check if it supports "query-cpu-stopped-state" */
1596 if (prom_getprop(rtas_node
, "query-cpu-stopped-state",
1597 &val
, sizeof(val
)) != PROM_ERROR
)
1598 rtas_has_query_cpu_stopped
= true;
1600 #if defined(CONFIG_PPC_POWERNV) && defined(__BIG_ENDIAN__)
1601 /* PowerVN takeover hack */
1602 prom_rtas_data
= base
;
1603 prom_rtas_entry
= entry
;
1604 prom_getprop(rtas_node
, "start-cpu", &prom_rtas_start_cpu
, 4);
1606 prom_debug("rtas base = 0x%x\n", base
);
1607 prom_debug("rtas entry = 0x%x\n", entry
);
1608 prom_debug("rtas size = 0x%x\n", (long)size
);
1610 prom_debug("prom_instantiate_rtas: end...\n");
1615 * Allocate room for and instantiate Stored Measurement Log (SML)
1617 static void __init
prom_instantiate_sml(void)
1619 phandle ibmvtpm_node
;
1620 ihandle ibmvtpm_inst
;
1621 u32 entry
= 0, size
= 0;
1624 prom_debug("prom_instantiate_sml: start...\n");
1626 ibmvtpm_node
= call_prom("finddevice", 1, 1, ADDR("/ibm,vtpm"));
1627 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node
);
1628 if (!PHANDLE_VALID(ibmvtpm_node
))
1631 ibmvtpm_inst
= call_prom("open", 1, 1, ADDR("/ibm,vtpm"));
1632 if (!IHANDLE_VALID(ibmvtpm_inst
)) {
1633 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst
);
1637 if (call_prom_ret("call-method", 2, 2, &size
,
1638 ADDR("sml-get-handover-size"),
1639 ibmvtpm_inst
) != 0 || size
== 0) {
1640 prom_printf("SML get handover size failed\n");
1644 base
= alloc_down(size
, PAGE_SIZE
, 0);
1646 prom_panic("Could not allocate memory for sml\n");
1648 prom_printf("instantiating sml at 0x%x...", base
);
1650 if (call_prom_ret("call-method", 4, 2, &entry
,
1651 ADDR("sml-handover"),
1652 ibmvtpm_inst
, size
, base
) != 0 || entry
== 0) {
1653 prom_printf("SML handover failed\n");
1656 prom_printf(" done\n");
1658 reserve_mem(base
, size
);
1660 prom_setprop(ibmvtpm_node
, "/ibm,vtpm", "linux,sml-base",
1661 &base
, sizeof(base
));
1662 prom_setprop(ibmvtpm_node
, "/ibm,vtpm", "linux,sml-size",
1663 &size
, sizeof(size
));
1665 prom_debug("sml base = 0x%x\n", base
);
1666 prom_debug("sml size = 0x%x\n", (long)size
);
1668 prom_debug("prom_instantiate_sml: end...\n");
1672 * Allocate room for and initialize TCE tables
1674 #ifdef __BIG_ENDIAN__
1675 static void __init
prom_initialize_tce_table(void)
1679 char compatible
[64], type
[64], model
[64];
1680 char *path
= prom_scratch
;
1682 u32 minalign
, minsize
;
1683 u64 tce_entry
, *tce_entryp
;
1684 u64 local_alloc_top
, local_alloc_bottom
;
1690 prom_debug("starting prom_initialize_tce_table\n");
1692 /* Cache current top of allocs so we reserve a single block */
1693 local_alloc_top
= alloc_top_high
;
1694 local_alloc_bottom
= local_alloc_top
;
1696 /* Search all nodes looking for PHBs. */
1697 for (node
= 0; prom_next_node(&node
); ) {
1701 prom_getprop(node
, "compatible",
1702 compatible
, sizeof(compatible
));
1703 prom_getprop(node
, "device_type", type
, sizeof(type
));
1704 prom_getprop(node
, "model", model
, sizeof(model
));
1706 if ((type
[0] == 0) || (strstr(type
, "pci") == NULL
))
1709 /* Keep the old logic intact to avoid regression. */
1710 if (compatible
[0] != 0) {
1711 if ((strstr(compatible
, "python") == NULL
) &&
1712 (strstr(compatible
, "Speedwagon") == NULL
) &&
1713 (strstr(compatible
, "Winnipeg") == NULL
))
1715 } else if (model
[0] != 0) {
1716 if ((strstr(model
, "ython") == NULL
) &&
1717 (strstr(model
, "peedwagon") == NULL
) &&
1718 (strstr(model
, "innipeg") == NULL
))
1722 if (prom_getprop(node
, "tce-table-minalign", &minalign
,
1723 sizeof(minalign
)) == PROM_ERROR
)
1725 if (prom_getprop(node
, "tce-table-minsize", &minsize
,
1726 sizeof(minsize
)) == PROM_ERROR
)
1727 minsize
= 4UL << 20;
1730 * Even though we read what OF wants, we just set the table
1731 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
1732 * By doing this, we avoid the pitfalls of trying to DMA to
1733 * MMIO space and the DMA alias hole.
1735 * On POWER4, firmware sets the TCE region by assuming
1736 * each TCE table is 8MB. Using this memory for anything
1737 * else will impact performance, so we always allocate 8MB.
1740 if (pvr_version_is(PVR_POWER4
) || pvr_version_is(PVR_POWER4p
))
1741 minsize
= 8UL << 20;
1743 minsize
= 4UL << 20;
1745 /* Align to the greater of the align or size */
1746 align
= max(minalign
, minsize
);
1747 base
= alloc_down(minsize
, align
, 1);
1749 prom_panic("ERROR, cannot find space for TCE table.\n");
1750 if (base
< local_alloc_bottom
)
1751 local_alloc_bottom
= base
;
1753 /* It seems OF doesn't null-terminate the path :-( */
1754 memset(path
, 0, PROM_SCRATCH_SIZE
);
1755 /* Call OF to setup the TCE hardware */
1756 if (call_prom("package-to-path", 3, 1, node
,
1757 path
, PROM_SCRATCH_SIZE
-1) == PROM_ERROR
) {
1758 prom_printf("package-to-path failed\n");
1761 /* Save away the TCE table attributes for later use. */
1762 prom_setprop(node
, path
, "linux,tce-base", &base
, sizeof(base
));
1763 prom_setprop(node
, path
, "linux,tce-size", &minsize
, sizeof(minsize
));
1765 prom_debug("TCE table: %s\n", path
);
1766 prom_debug("\tnode = 0x%x\n", node
);
1767 prom_debug("\tbase = 0x%x\n", base
);
1768 prom_debug("\tsize = 0x%x\n", minsize
);
1770 /* Initialize the table to have a one-to-one mapping
1771 * over the allocated size.
1773 tce_entryp
= (u64
*)base
;
1774 for (i
= 0; i
< (minsize
>> 3) ;tce_entryp
++, i
++) {
1775 tce_entry
= (i
<< PAGE_SHIFT
);
1777 *tce_entryp
= tce_entry
;
1780 prom_printf("opening PHB %s", path
);
1781 phb_node
= call_prom("open", 1, 1, path
);
1783 prom_printf("... failed\n");
1785 prom_printf("... done\n");
1787 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1788 phb_node
, -1, minsize
,
1789 (u32
) base
, (u32
) (base
>> 32));
1790 call_prom("close", 1, 0, phb_node
);
1793 reserve_mem(local_alloc_bottom
, local_alloc_top
- local_alloc_bottom
);
1795 /* These are only really needed if there is a memory limit in
1796 * effect, but we don't know so export them always. */
1797 prom_tce_alloc_start
= local_alloc_bottom
;
1798 prom_tce_alloc_end
= local_alloc_top
;
1800 /* Flag the first invalid entry */
1801 prom_debug("ending prom_initialize_tce_table\n");
1803 #endif /* __BIG_ENDIAN__ */
1804 #endif /* CONFIG_PPC64 */
1807 * With CHRP SMP we need to use the OF to start the other processors.
1808 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1809 * so we have to put the processors into a holding pattern controlled
1810 * by the kernel (not OF) before we destroy the OF.
1812 * This uses a chunk of low memory, puts some holding pattern
1813 * code there and sends the other processors off to there until
1814 * smp_boot_cpus tells them to do something. The holding pattern
1815 * checks that address until its cpu # is there, when it is that
1816 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
1817 * of setting those values.
1819 * We also use physical address 0x4 here to tell when a cpu
1820 * is in its holding pattern code.
1825 * We want to reference the copy of __secondary_hold_* in the
1826 * 0 - 0x100 address range
1828 #define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
1830 static void __init
prom_hold_cpus(void)
1835 unsigned long *spinloop
1836 = (void *) LOW_ADDR(__secondary_hold_spinloop
);
1837 unsigned long *acknowledge
1838 = (void *) LOW_ADDR(__secondary_hold_acknowledge
);
1839 unsigned long secondary_hold
= LOW_ADDR(__secondary_hold
);
1842 * On pseries, if RTAS supports "query-cpu-stopped-state",
1843 * we skip this stage, the CPUs will be started by the
1844 * kernel using RTAS.
1846 if ((of_platform
== PLATFORM_PSERIES
||
1847 of_platform
== PLATFORM_PSERIES_LPAR
) &&
1848 rtas_has_query_cpu_stopped
) {
1849 prom_printf("prom_hold_cpus: skipped\n");
1853 prom_debug("prom_hold_cpus: start...\n");
1854 prom_debug(" 1) spinloop = 0x%x\n", (unsigned long)spinloop
);
1855 prom_debug(" 1) *spinloop = 0x%x\n", *spinloop
);
1856 prom_debug(" 1) acknowledge = 0x%x\n",
1857 (unsigned long)acknowledge
);
1858 prom_debug(" 1) *acknowledge = 0x%x\n", *acknowledge
);
1859 prom_debug(" 1) secondary_hold = 0x%x\n", secondary_hold
);
1861 /* Set the common spinloop variable, so all of the secondary cpus
1862 * will block when they are awakened from their OF spinloop.
1863 * This must occur for both SMP and non SMP kernels, since OF will
1864 * be trashed when we move the kernel.
1869 for (node
= 0; prom_next_node(&node
); ) {
1870 unsigned int cpu_no
;
1874 prom_getprop(node
, "device_type", type
, sizeof(type
));
1875 if (strcmp(type
, "cpu") != 0)
1878 /* Skip non-configured cpus. */
1879 if (prom_getprop(node
, "status", type
, sizeof(type
)) > 0)
1880 if (strcmp(type
, "okay") != 0)
1883 reg
= cpu_to_be32(-1); /* make sparse happy */
1884 prom_getprop(node
, "reg", ®
, sizeof(reg
));
1885 cpu_no
= be32_to_cpu(reg
);
1887 prom_debug("cpu hw idx = %lu\n", cpu_no
);
1889 /* Init the acknowledge var which will be reset by
1890 * the secondary cpu when it awakens from its OF
1893 *acknowledge
= (unsigned long)-1;
1895 if (cpu_no
!= prom
.cpu
) {
1896 /* Primary Thread of non-boot cpu or any thread */
1897 prom_printf("starting cpu hw idx %lu... ", cpu_no
);
1898 call_prom("start-cpu", 3, 0, node
,
1899 secondary_hold
, cpu_no
);
1901 for (i
= 0; (i
< 100000000) &&
1902 (*acknowledge
== ((unsigned long)-1)); i
++ )
1905 if (*acknowledge
== cpu_no
)
1906 prom_printf("done\n");
1908 prom_printf("failed: %x\n", *acknowledge
);
1912 prom_printf("boot cpu hw idx %lu\n", cpu_no
);
1913 #endif /* CONFIG_SMP */
1916 prom_debug("prom_hold_cpus: end...\n");
1920 static void __init
prom_init_client_services(unsigned long pp
)
1922 /* Get a handle to the prom entry point before anything else */
1925 /* get a handle for the stdout device */
1926 prom
.chosen
= call_prom("finddevice", 1, 1, ADDR("/chosen"));
1927 if (!PHANDLE_VALID(prom
.chosen
))
1928 prom_panic("cannot find chosen"); /* msg won't be printed :( */
1930 /* get device tree root */
1931 prom
.root
= call_prom("finddevice", 1, 1, ADDR("/"));
1932 if (!PHANDLE_VALID(prom
.root
))
1933 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1940 * For really old powermacs, we need to map things we claim.
1941 * For that, we need the ihandle of the mmu.
1942 * Also, on the longtrail, we need to work around other bugs.
1944 static void __init
prom_find_mmu(void)
1949 oprom
= call_prom("finddevice", 1, 1, ADDR("/openprom"));
1950 if (!PHANDLE_VALID(oprom
))
1952 if (prom_getprop(oprom
, "model", version
, sizeof(version
)) <= 0)
1954 version
[sizeof(version
) - 1] = 0;
1955 /* XXX might need to add other versions here */
1956 if (strcmp(version
, "Open Firmware, 1.0.5") == 0)
1957 of_workarounds
= OF_WA_CLAIM
;
1958 else if (strncmp(version
, "FirmWorks,3.", 12) == 0) {
1959 of_workarounds
= OF_WA_CLAIM
| OF_WA_LONGTRAIL
;
1960 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1963 prom
.memory
= call_prom("open", 1, 1, ADDR("/memory"));
1964 prom_getprop(prom
.chosen
, "mmu", &prom
.mmumap
,
1965 sizeof(prom
.mmumap
));
1966 prom
.mmumap
= be32_to_cpu(prom
.mmumap
);
1967 if (!IHANDLE_VALID(prom
.memory
) || !IHANDLE_VALID(prom
.mmumap
))
1968 of_workarounds
&= ~OF_WA_CLAIM
; /* hmmm */
1971 #define prom_find_mmu()
1974 static void __init
prom_init_stdout(void)
1976 char *path
= of_stdout_device
;
1978 phandle stdout_node
;
1981 if (prom_getprop(prom
.chosen
, "stdout", &val
, sizeof(val
)) <= 0)
1982 prom_panic("cannot find stdout");
1984 prom
.stdout
= be32_to_cpu(val
);
1986 /* Get the full OF pathname of the stdout device */
1987 memset(path
, 0, 256);
1988 call_prom("instance-to-path", 3, 1, prom
.stdout
, path
, 255);
1989 prom_printf("OF stdout device is: %s\n", of_stdout_device
);
1990 prom_setprop(prom
.chosen
, "/chosen", "linux,stdout-path",
1991 path
, strlen(path
) + 1);
1993 /* instance-to-package fails on PA-Semi */
1994 stdout_node
= call_prom("instance-to-package", 1, 1, prom
.stdout
);
1995 if (stdout_node
!= PROM_ERROR
) {
1996 val
= cpu_to_be32(stdout_node
);
1997 prom_setprop(prom
.chosen
, "/chosen", "linux,stdout-package",
2000 /* If it's a display, note it */
2001 memset(type
, 0, sizeof(type
));
2002 prom_getprop(stdout_node
, "device_type", type
, sizeof(type
));
2003 if (strcmp(type
, "display") == 0)
2004 prom_setprop(stdout_node
, path
, "linux,boot-display", NULL
, 0);
2008 static int __init
prom_find_machine_type(void)
2017 /* Look for a PowerMac or a Cell */
2018 len
= prom_getprop(prom
.root
, "compatible",
2019 compat
, sizeof(compat
)-1);
2023 char *p
= &compat
[i
];
2027 if (strstr(p
, "Power Macintosh") ||
2028 strstr(p
, "MacRISC"))
2029 return PLATFORM_POWERMAC
;
2031 /* We must make sure we don't detect the IBM Cell
2032 * blades as pSeries due to some firmware issues,
2035 if (strstr(p
, "IBM,CBEA") ||
2036 strstr(p
, "IBM,CPBW-1.0"))
2037 return PLATFORM_GENERIC
;
2038 #endif /* CONFIG_PPC64 */
2043 /* Try to detect OPAL */
2044 if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
2045 return PLATFORM_OPAL
;
2047 /* Try to figure out if it's an IBM pSeries or any other
2048 * PAPR compliant platform. We assume it is if :
2049 * - /device_type is "chrp" (please, do NOT use that for future
2053 len
= prom_getprop(prom
.root
, "device_type",
2054 compat
, sizeof(compat
)-1);
2056 return PLATFORM_GENERIC
;
2057 if (strcmp(compat
, "chrp"))
2058 return PLATFORM_GENERIC
;
2060 /* Default to pSeries. We need to know if we are running LPAR */
2061 rtas
= call_prom("finddevice", 1, 1, ADDR("/rtas"));
2062 if (!PHANDLE_VALID(rtas
))
2063 return PLATFORM_GENERIC
;
2064 x
= prom_getproplen(rtas
, "ibm,hypertas-functions");
2065 if (x
!= PROM_ERROR
) {
2066 prom_debug("Hypertas detected, assuming LPAR !\n");
2067 return PLATFORM_PSERIES_LPAR
;
2069 return PLATFORM_PSERIES
;
2071 return PLATFORM_GENERIC
;
2075 static int __init
prom_set_color(ihandle ih
, int i
, int r
, int g
, int b
)
2077 return call_prom("call-method", 6, 1, ADDR("color!"), ih
, i
, b
, g
, r
);
2081 * If we have a display that we don't know how to drive,
2082 * we will want to try to execute OF's open method for it
2083 * later. However, OF will probably fall over if we do that
2084 * we've taken over the MMU.
2085 * So we check whether we will need to open the display,
2086 * and if so, open it now.
2088 static void __init
prom_check_displays(void)
2090 char type
[16], *path
;
2095 static unsigned char default_colors
[] = {
2113 const unsigned char *clut
;
2115 prom_debug("Looking for displays\n");
2116 for (node
= 0; prom_next_node(&node
); ) {
2117 memset(type
, 0, sizeof(type
));
2118 prom_getprop(node
, "device_type", type
, sizeof(type
));
2119 if (strcmp(type
, "display") != 0)
2122 /* It seems OF doesn't null-terminate the path :-( */
2123 path
= prom_scratch
;
2124 memset(path
, 0, PROM_SCRATCH_SIZE
);
2127 * leave some room at the end of the path for appending extra
2130 if (call_prom("package-to-path", 3, 1, node
, path
,
2131 PROM_SCRATCH_SIZE
-10) == PROM_ERROR
)
2133 prom_printf("found display : %s, opening... ", path
);
2135 ih
= call_prom("open", 1, 1, path
);
2137 prom_printf("failed\n");
2142 prom_printf("done\n");
2143 prom_setprop(node
, path
, "linux,opened", NULL
, 0);
2145 /* Setup a usable color table when the appropriate
2146 * method is available. Should update this to set-colors */
2147 clut
= default_colors
;
2148 for (i
= 0; i
< 16; i
++, clut
+= 3)
2149 if (prom_set_color(ih
, i
, clut
[0], clut
[1],
2153 #ifdef CONFIG_LOGO_LINUX_CLUT224
2154 clut
= PTRRELOC(logo_linux_clut224
.clut
);
2155 for (i
= 0; i
< logo_linux_clut224
.clutsize
; i
++, clut
+= 3)
2156 if (prom_set_color(ih
, i
+ 32, clut
[0], clut
[1],
2159 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
2161 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2162 if (prom_getprop(node
, "linux,boot-display", NULL
, 0) !=
2164 u32 width
, height
, pitch
, addr
;
2166 prom_printf("Setting btext !\n");
2167 prom_getprop(node
, "width", &width
, 4);
2168 prom_getprop(node
, "height", &height
, 4);
2169 prom_getprop(node
, "linebytes", &pitch
, 4);
2170 prom_getprop(node
, "address", &addr
, 4);
2171 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2172 width
, height
, pitch
, addr
);
2173 btext_setup_display(width
, height
, 8, pitch
, addr
);
2175 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2180 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2181 static void __init
*make_room(unsigned long *mem_start
, unsigned long *mem_end
,
2182 unsigned long needed
, unsigned long align
)
2186 *mem_start
= _ALIGN(*mem_start
, align
);
2187 while ((*mem_start
+ needed
) > *mem_end
) {
2188 unsigned long room
, chunk
;
2190 prom_debug("Chunk exhausted, claiming more at %x...\n",
2192 room
= alloc_top
- alloc_bottom
;
2193 if (room
> DEVTREE_CHUNK_SIZE
)
2194 room
= DEVTREE_CHUNK_SIZE
;
2195 if (room
< PAGE_SIZE
)
2196 prom_panic("No memory for flatten_device_tree "
2198 chunk
= alloc_up(room
, 0);
2200 prom_panic("No memory for flatten_device_tree "
2201 "(claim failed)\n");
2202 *mem_end
= chunk
+ room
;
2205 ret
= (void *)*mem_start
;
2206 *mem_start
+= needed
;
2211 #define dt_push_token(token, mem_start, mem_end) do { \
2212 void *room = make_room(mem_start, mem_end, 4, 4); \
2213 *(__be32 *)room = cpu_to_be32(token); \
2216 static unsigned long __init
dt_find_string(char *str
)
2220 s
= os
= (char *)dt_string_start
;
2222 while (s
< (char *)dt_string_end
) {
2223 if (strcmp(s
, str
) == 0)
2231 * The Open Firmware 1275 specification states properties must be 31 bytes or
2232 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2234 #define MAX_PROPERTY_NAME 64
2236 static void __init
scan_dt_build_strings(phandle node
,
2237 unsigned long *mem_start
,
2238 unsigned long *mem_end
)
2240 char *prev_name
, *namep
, *sstart
;
2244 sstart
= (char *)dt_string_start
;
2246 /* get and store all property names */
2249 /* 64 is max len of name including nul. */
2250 namep
= make_room(mem_start
, mem_end
, MAX_PROPERTY_NAME
, 1);
2251 if (call_prom("nextprop", 3, 1, node
, prev_name
, namep
) != 1) {
2252 /* No more nodes: unwind alloc */
2253 *mem_start
= (unsigned long)namep
;
2258 if (strcmp(namep
, "name") == 0) {
2259 *mem_start
= (unsigned long)namep
;
2263 /* get/create string entry */
2264 soff
= dt_find_string(namep
);
2266 *mem_start
= (unsigned long)namep
;
2267 namep
= sstart
+ soff
;
2269 /* Trim off some if we can */
2270 *mem_start
= (unsigned long)namep
+ strlen(namep
) + 1;
2271 dt_string_end
= *mem_start
;
2276 /* do all our children */
2277 child
= call_prom("child", 1, 1, node
);
2278 while (child
!= 0) {
2279 scan_dt_build_strings(child
, mem_start
, mem_end
);
2280 child
= call_prom("peer", 1, 1, child
);
2284 static void __init
scan_dt_build_struct(phandle node
, unsigned long *mem_start
,
2285 unsigned long *mem_end
)
2288 char *namep
, *prev_name
, *sstart
, *p
, *ep
, *lp
, *path
;
2290 unsigned char *valp
;
2291 static char pname
[MAX_PROPERTY_NAME
];
2292 int l
, room
, has_phandle
= 0;
2294 dt_push_token(OF_DT_BEGIN_NODE
, mem_start
, mem_end
);
2296 /* get the node's full name */
2297 namep
= (char *)*mem_start
;
2298 room
= *mem_end
- *mem_start
;
2301 l
= call_prom("package-to-path", 3, 1, node
, namep
, room
);
2303 /* Didn't fit? Get more room. */
2305 if (l
>= *mem_end
- *mem_start
)
2306 namep
= make_room(mem_start
, mem_end
, l
+1, 1);
2307 call_prom("package-to-path", 3, 1, node
, namep
, l
);
2311 /* Fixup an Apple bug where they have bogus \0 chars in the
2312 * middle of the path in some properties, and extract
2313 * the unit name (everything after the last '/').
2315 for (lp
= p
= namep
, ep
= namep
+ l
; p
< ep
; p
++) {
2322 *mem_start
= _ALIGN((unsigned long)lp
+ 1, 4);
2325 /* get it again for debugging */
2326 path
= prom_scratch
;
2327 memset(path
, 0, PROM_SCRATCH_SIZE
);
2328 call_prom("package-to-path", 3, 1, node
, path
, PROM_SCRATCH_SIZE
-1);
2330 /* get and store all properties */
2332 sstart
= (char *)dt_string_start
;
2334 if (call_prom("nextprop", 3, 1, node
, prev_name
,
2339 if (strcmp(pname
, "name") == 0) {
2344 /* find string offset */
2345 soff
= dt_find_string(pname
);
2347 prom_printf("WARNING: Can't find string index for"
2348 " <%s>, node %s\n", pname
, path
);
2351 prev_name
= sstart
+ soff
;
2354 l
= call_prom("getproplen", 2, 1, node
, pname
);
2357 if (l
== PROM_ERROR
)
2360 /* push property head */
2361 dt_push_token(OF_DT_PROP
, mem_start
, mem_end
);
2362 dt_push_token(l
, mem_start
, mem_end
);
2363 dt_push_token(soff
, mem_start
, mem_end
);
2365 /* push property content */
2366 valp
= make_room(mem_start
, mem_end
, l
, 4);
2367 call_prom("getprop", 4, 1, node
, pname
, valp
, l
);
2368 *mem_start
= _ALIGN(*mem_start
, 4);
2370 if (!strcmp(pname
, "phandle"))
2374 /* Add a "linux,phandle" property if no "phandle" property already
2375 * existed (can happen with OPAL)
2378 soff
= dt_find_string("linux,phandle");
2380 prom_printf("WARNING: Can't find string index for"
2381 " <linux-phandle> node %s\n", path
);
2383 dt_push_token(OF_DT_PROP
, mem_start
, mem_end
);
2384 dt_push_token(4, mem_start
, mem_end
);
2385 dt_push_token(soff
, mem_start
, mem_end
);
2386 valp
= make_room(mem_start
, mem_end
, 4, 4);
2387 *(__be32
*)valp
= cpu_to_be32(node
);
2391 /* do all our children */
2392 child
= call_prom("child", 1, 1, node
);
2393 while (child
!= 0) {
2394 scan_dt_build_struct(child
, mem_start
, mem_end
);
2395 child
= call_prom("peer", 1, 1, child
);
2398 dt_push_token(OF_DT_END_NODE
, mem_start
, mem_end
);
2401 static void __init
flatten_device_tree(void)
2404 unsigned long mem_start
, mem_end
, room
;
2405 struct boot_param_header
*hdr
;
2410 * Check how much room we have between alloc top & bottom (+/- a
2411 * few pages), crop to 1MB, as this is our "chunk" size
2413 room
= alloc_top
- alloc_bottom
- 0x4000;
2414 if (room
> DEVTREE_CHUNK_SIZE
)
2415 room
= DEVTREE_CHUNK_SIZE
;
2416 prom_debug("starting device tree allocs at %x\n", alloc_bottom
);
2418 /* Now try to claim that */
2419 mem_start
= (unsigned long)alloc_up(room
, PAGE_SIZE
);
2421 prom_panic("Can't allocate initial device-tree chunk\n");
2422 mem_end
= mem_start
+ room
;
2424 /* Get root of tree */
2425 root
= call_prom("peer", 1, 1, (phandle
)0);
2426 if (root
== (phandle
)0)
2427 prom_panic ("couldn't get device tree root\n");
2429 /* Build header and make room for mem rsv map */
2430 mem_start
= _ALIGN(mem_start
, 4);
2431 hdr
= make_room(&mem_start
, &mem_end
,
2432 sizeof(struct boot_param_header
), 4);
2433 dt_header_start
= (unsigned long)hdr
;
2434 rsvmap
= make_room(&mem_start
, &mem_end
, sizeof(mem_reserve_map
), 8);
2436 /* Start of strings */
2437 mem_start
= PAGE_ALIGN(mem_start
);
2438 dt_string_start
= mem_start
;
2439 mem_start
+= 4; /* hole */
2441 /* Add "linux,phandle" in there, we'll need it */
2442 namep
= make_room(&mem_start
, &mem_end
, 16, 1);
2443 strcpy(namep
, "linux,phandle");
2444 mem_start
= (unsigned long)namep
+ strlen(namep
) + 1;
2446 /* Build string array */
2447 prom_printf("Building dt strings...\n");
2448 scan_dt_build_strings(root
, &mem_start
, &mem_end
);
2449 dt_string_end
= mem_start
;
2451 /* Build structure */
2452 mem_start
= PAGE_ALIGN(mem_start
);
2453 dt_struct_start
= mem_start
;
2454 prom_printf("Building dt structure...\n");
2455 scan_dt_build_struct(root
, &mem_start
, &mem_end
);
2456 dt_push_token(OF_DT_END
, &mem_start
, &mem_end
);
2457 dt_struct_end
= PAGE_ALIGN(mem_start
);
2460 hdr
->boot_cpuid_phys
= cpu_to_be32(prom
.cpu
);
2461 hdr
->magic
= cpu_to_be32(OF_DT_HEADER
);
2462 hdr
->totalsize
= cpu_to_be32(dt_struct_end
- dt_header_start
);
2463 hdr
->off_dt_struct
= cpu_to_be32(dt_struct_start
- dt_header_start
);
2464 hdr
->off_dt_strings
= cpu_to_be32(dt_string_start
- dt_header_start
);
2465 hdr
->dt_strings_size
= cpu_to_be32(dt_string_end
- dt_string_start
);
2466 hdr
->off_mem_rsvmap
= cpu_to_be32(((unsigned long)rsvmap
) - dt_header_start
);
2467 hdr
->version
= cpu_to_be32(OF_DT_VERSION
);
2468 /* Version 16 is not backward compatible */
2469 hdr
->last_comp_version
= cpu_to_be32(0x10);
2471 /* Copy the reserve map in */
2472 memcpy(rsvmap
, mem_reserve_map
, sizeof(mem_reserve_map
));
2477 prom_printf("reserved memory map:\n");
2478 for (i
= 0; i
< mem_reserve_cnt
; i
++)
2479 prom_printf(" %x - %x\n",
2480 be64_to_cpu(mem_reserve_map
[i
].base
),
2481 be64_to_cpu(mem_reserve_map
[i
].size
));
2484 /* Bump mem_reserve_cnt to cause further reservations to fail
2485 * since it's too late.
2487 mem_reserve_cnt
= MEM_RESERVE_MAP_SIZE
;
2489 prom_printf("Device tree strings 0x%x -> 0x%x\n",
2490 dt_string_start
, dt_string_end
);
2491 prom_printf("Device tree struct 0x%x -> 0x%x\n",
2492 dt_struct_start
, dt_struct_end
);
2495 #ifdef CONFIG_PPC_MAPLE
2496 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2497 * The values are bad, and it doesn't even have the right number of cells. */
2498 static void __init
fixup_device_tree_maple(void)
2501 u32 rloc
= 0x01002000; /* IO space; PCI device = 4 */
2505 name
= "/ht@0/isa@4";
2506 isa
= call_prom("finddevice", 1, 1, ADDR(name
));
2507 if (!PHANDLE_VALID(isa
)) {
2508 name
= "/ht@0/isa@6";
2509 isa
= call_prom("finddevice", 1, 1, ADDR(name
));
2510 rloc
= 0x01003000; /* IO space; PCI device = 6 */
2512 if (!PHANDLE_VALID(isa
))
2515 if (prom_getproplen(isa
, "ranges") != 12)
2517 if (prom_getprop(isa
, "ranges", isa_ranges
, sizeof(isa_ranges
))
2521 if (isa_ranges
[0] != 0x1 ||
2522 isa_ranges
[1] != 0xf4000000 ||
2523 isa_ranges
[2] != 0x00010000)
2526 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2528 isa_ranges
[0] = 0x1;
2529 isa_ranges
[1] = 0x0;
2530 isa_ranges
[2] = rloc
;
2531 isa_ranges
[3] = 0x0;
2532 isa_ranges
[4] = 0x0;
2533 isa_ranges
[5] = 0x00010000;
2534 prom_setprop(isa
, name
, "ranges",
2535 isa_ranges
, sizeof(isa_ranges
));
2538 #define CPC925_MC_START 0xf8000000
2539 #define CPC925_MC_LENGTH 0x1000000
2540 /* The values for memory-controller don't have right number of cells */
2541 static void __init
fixup_device_tree_maple_memory_controller(void)
2545 char *name
= "/hostbridge@f8000000";
2548 mc
= call_prom("finddevice", 1, 1, ADDR(name
));
2549 if (!PHANDLE_VALID(mc
))
2552 if (prom_getproplen(mc
, "reg") != 8)
2555 prom_getprop(prom
.root
, "#address-cells", &ac
, sizeof(ac
));
2556 prom_getprop(prom
.root
, "#size-cells", &sc
, sizeof(sc
));
2557 if ((ac
!= 2) || (sc
!= 2))
2560 if (prom_getprop(mc
, "reg", mc_reg
, sizeof(mc_reg
)) == PROM_ERROR
)
2563 if (mc_reg
[0] != CPC925_MC_START
|| mc_reg
[1] != CPC925_MC_LENGTH
)
2566 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2569 mc_reg
[1] = CPC925_MC_START
;
2571 mc_reg
[3] = CPC925_MC_LENGTH
;
2572 prom_setprop(mc
, name
, "reg", mc_reg
, sizeof(mc_reg
));
2575 #define fixup_device_tree_maple()
2576 #define fixup_device_tree_maple_memory_controller()
2579 #ifdef CONFIG_PPC_CHRP
2581 * Pegasos and BriQ lacks the "ranges" property in the isa node
2582 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2583 * Pegasos has the IDE configured in legacy mode, but advertised as native
2585 static void __init
fixup_device_tree_chrp(void)
2589 u32 rloc
= 0x01006000; /* IO space; PCI device = 12 */
2593 name
= "/pci@80000000/isa@c";
2594 ph
= call_prom("finddevice", 1, 1, ADDR(name
));
2595 if (!PHANDLE_VALID(ph
)) {
2596 name
= "/pci@ff500000/isa@6";
2597 ph
= call_prom("finddevice", 1, 1, ADDR(name
));
2598 rloc
= 0x01003000; /* IO space; PCI device = 6 */
2600 if (PHANDLE_VALID(ph
)) {
2601 rc
= prom_getproplen(ph
, "ranges");
2602 if (rc
== 0 || rc
== PROM_ERROR
) {
2603 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2610 prop
[5] = 0x00010000;
2611 prom_setprop(ph
, name
, "ranges", prop
, sizeof(prop
));
2615 name
= "/pci@80000000/ide@C,1";
2616 ph
= call_prom("finddevice", 1, 1, ADDR(name
));
2617 if (PHANDLE_VALID(ph
)) {
2618 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2621 prom_setprop(ph
, name
, "interrupts", prop
, 2*sizeof(u32
));
2622 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2623 rc
= prom_getprop(ph
, "class-code", prop
, sizeof(u32
));
2624 if (rc
== sizeof(u32
)) {
2626 prom_setprop(ph
, name
, "class-code", prop
, sizeof(u32
));
2631 #define fixup_device_tree_chrp()
2634 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2635 static void __init
fixup_device_tree_pmac(void)
2637 phandle u3
, i2c
, mpic
;
2642 /* Some G5s have a missing interrupt definition, fix it up here */
2643 u3
= call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2644 if (!PHANDLE_VALID(u3
))
2646 i2c
= call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2647 if (!PHANDLE_VALID(i2c
))
2649 mpic
= call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2650 if (!PHANDLE_VALID(mpic
))
2653 /* check if proper rev of u3 */
2654 if (prom_getprop(u3
, "device-rev", &u3_rev
, sizeof(u3_rev
))
2657 if (u3_rev
< 0x35 || u3_rev
> 0x39)
2659 /* does it need fixup ? */
2660 if (prom_getproplen(i2c
, "interrupts") > 0)
2663 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2665 /* interrupt on this revision of u3 is number 0 and level */
2668 prom_setprop(i2c
, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2669 &interrupts
, sizeof(interrupts
));
2671 prom_setprop(i2c
, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2672 &parent
, sizeof(parent
));
2675 #define fixup_device_tree_pmac()
2678 #ifdef CONFIG_PPC_EFIKA
2680 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2681 * to talk to the phy. If the phy-handle property is missing, then this
2682 * function is called to add the appropriate nodes and link it to the
2685 static void __init
fixup_device_tree_efika_add_phy(void)
2691 /* Check if /builtin/ethernet exists - bail if it doesn't */
2692 node
= call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2693 if (!PHANDLE_VALID(node
))
2696 /* Check if the phy-handle property exists - bail if it does */
2697 rv
= prom_getprop(node
, "phy-handle", prop
, sizeof(prop
));
2702 * At this point the ethernet device doesn't have a phy described.
2703 * Now we need to add the missing phy node and linkage
2706 /* Check for an MDIO bus node - if missing then create one */
2707 node
= call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2708 if (!PHANDLE_VALID(node
)) {
2709 prom_printf("Adding Ethernet MDIO node\n");
2710 call_prom("interpret", 1, 1,
2711 " s\" /builtin\" find-device"
2713 " 1 encode-int s\" #address-cells\" property"
2714 " 0 encode-int s\" #size-cells\" property"
2715 " s\" mdio\" device-name"
2716 " s\" fsl,mpc5200b-mdio\" encode-string"
2717 " s\" compatible\" property"
2718 " 0xf0003000 0x400 reg"
2720 " 0x5 encode-int encode+"
2721 " 0x3 encode-int encode+"
2722 " s\" interrupts\" property"
2726 /* Check for a PHY device node - if missing then create one and
2727 * give it's phandle to the ethernet node */
2728 node
= call_prom("finddevice", 1, 1,
2729 ADDR("/builtin/mdio/ethernet-phy"));
2730 if (!PHANDLE_VALID(node
)) {
2731 prom_printf("Adding Ethernet PHY node\n");
2732 call_prom("interpret", 1, 1,
2733 " s\" /builtin/mdio\" find-device"
2735 " s\" ethernet-phy\" device-name"
2736 " 0x10 encode-int s\" reg\" property"
2740 " s\" /builtin/ethernet\" find-device"
2742 " s\" phy-handle\" property"
2747 static void __init
fixup_device_tree_efika(void)
2749 int sound_irq
[3] = { 2, 2, 0 };
2750 int bcomm_irq
[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2751 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2752 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2753 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2758 /* Check if we're really running on a EFIKA */
2759 node
= call_prom("finddevice", 1, 1, ADDR("/"));
2760 if (!PHANDLE_VALID(node
))
2763 rv
= prom_getprop(node
, "model", prop
, sizeof(prop
));
2764 if (rv
== PROM_ERROR
)
2766 if (strcmp(prop
, "EFIKA5K2"))
2769 prom_printf("Applying EFIKA device tree fixups\n");
2771 /* Claiming to be 'chrp' is death */
2772 node
= call_prom("finddevice", 1, 1, ADDR("/"));
2773 rv
= prom_getprop(node
, "device_type", prop
, sizeof(prop
));
2774 if (rv
!= PROM_ERROR
&& (strcmp(prop
, "chrp") == 0))
2775 prom_setprop(node
, "/", "device_type", "efika", sizeof("efika"));
2777 /* CODEGEN,description is exposed in /proc/cpuinfo so
2779 rv
= prom_getprop(node
, "CODEGEN,description", prop
, sizeof(prop
));
2780 if (rv
!= PROM_ERROR
&& (strstr(prop
, "CHRP")))
2781 prom_setprop(node
, "/", "CODEGEN,description",
2782 "Efika 5200B PowerPC System",
2783 sizeof("Efika 5200B PowerPC System"));
2785 /* Fixup bestcomm interrupts property */
2786 node
= call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2787 if (PHANDLE_VALID(node
)) {
2788 len
= prom_getproplen(node
, "interrupts");
2790 prom_printf("Fixing bestcomm interrupts property\n");
2791 prom_setprop(node
, "/builtin/bestcom", "interrupts",
2792 bcomm_irq
, sizeof(bcomm_irq
));
2796 /* Fixup sound interrupts property */
2797 node
= call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2798 if (PHANDLE_VALID(node
)) {
2799 rv
= prom_getprop(node
, "interrupts", prop
, sizeof(prop
));
2800 if (rv
== PROM_ERROR
) {
2801 prom_printf("Adding sound interrupts property\n");
2802 prom_setprop(node
, "/builtin/sound", "interrupts",
2803 sound_irq
, sizeof(sound_irq
));
2807 /* Make sure ethernet phy-handle property exists */
2808 fixup_device_tree_efika_add_phy();
2811 #define fixup_device_tree_efika()
2814 static void __init
fixup_device_tree(void)
2816 fixup_device_tree_maple();
2817 fixup_device_tree_maple_memory_controller();
2818 fixup_device_tree_chrp();
2819 fixup_device_tree_pmac();
2820 fixup_device_tree_efika();
2823 static void __init
prom_find_boot_cpu(void)
2830 if (prom_getprop(prom
.chosen
, "cpu", &rval
, sizeof(rval
)) <= 0)
2832 prom_cpu
= be32_to_cpu(rval
);
2834 cpu_pkg
= call_prom("instance-to-package", 1, 1, prom_cpu
);
2836 prom_getprop(cpu_pkg
, "reg", &rval
, sizeof(rval
));
2837 prom
.cpu
= be32_to_cpu(rval
);
2839 prom_debug("Booting CPU hw index = %lu\n", prom
.cpu
);
2842 static void __init
prom_check_initrd(unsigned long r3
, unsigned long r4
)
2844 #ifdef CONFIG_BLK_DEV_INITRD
2845 if (r3
&& r4
&& r4
!= 0xdeadbeef) {
2848 prom_initrd_start
= is_kernel_addr(r3
) ? __pa(r3
) : r3
;
2849 prom_initrd_end
= prom_initrd_start
+ r4
;
2851 val
= cpu_to_be64(prom_initrd_start
);
2852 prom_setprop(prom
.chosen
, "/chosen", "linux,initrd-start",
2854 val
= cpu_to_be64(prom_initrd_end
);
2855 prom_setprop(prom
.chosen
, "/chosen", "linux,initrd-end",
2858 reserve_mem(prom_initrd_start
,
2859 prom_initrd_end
- prom_initrd_start
);
2861 prom_debug("initrd_start=0x%x\n", prom_initrd_start
);
2862 prom_debug("initrd_end=0x%x\n", prom_initrd_end
);
2864 #endif /* CONFIG_BLK_DEV_INITRD */
2868 #ifdef CONFIG_RELOCATABLE
2869 static void reloc_toc(void)
2873 static void unreloc_toc(void)
2877 static void __reloc_toc(unsigned long offset
, unsigned long nr_entries
)
2880 unsigned long *toc_entry
;
2882 /* Get the start of the TOC by using r2 directly. */
2883 asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry
));
2885 for (i
= 0; i
< nr_entries
; i
++) {
2886 *toc_entry
= *toc_entry
+ offset
;
2891 static void reloc_toc(void)
2893 unsigned long offset
= reloc_offset();
2894 unsigned long nr_entries
=
2895 (__prom_init_toc_end
- __prom_init_toc_start
) / sizeof(long);
2897 __reloc_toc(offset
, nr_entries
);
2902 static void unreloc_toc(void)
2904 unsigned long offset
= reloc_offset();
2905 unsigned long nr_entries
=
2906 (__prom_init_toc_end
- __prom_init_toc_start
) / sizeof(long);
2910 __reloc_toc(-offset
, nr_entries
);
2916 * We enter here early on, when the Open Firmware prom is still
2917 * handling exceptions and the MMU hash table for us.
2920 unsigned long __init
prom_init(unsigned long r3
, unsigned long r4
,
2922 unsigned long r6
, unsigned long r7
,
2923 unsigned long kbase
)
2928 unsigned long offset
= reloc_offset();
2935 * First zero the BSS
2937 memset(&__bss_start
, 0, __bss_stop
- __bss_start
);
2940 * Init interface to Open Firmware, get some node references,
2943 prom_init_client_services(pp
);
2946 * See if this OF is old enough that we need to do explicit maps
2947 * and other workarounds
2952 * Init prom stdout device
2956 prom_printf("Preparing to boot %s", linux_banner
);
2959 * Get default machine type. At this point, we do not differentiate
2960 * between pSeries SMP and pSeries LPAR
2962 of_platform
= prom_find_machine_type();
2963 prom_printf("Detected machine type: %x\n", of_platform
);
2965 #ifndef CONFIG_NONSTATIC_KERNEL
2966 /* Bail if this is a kdump kernel. */
2967 if (PHYSICAL_START
> 0)
2968 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2972 * Check for an initrd
2974 prom_check_initrd(r3
, r4
);
2976 #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
2978 * On pSeries, inform the firmware about our capabilities
2980 if (of_platform
== PLATFORM_PSERIES
||
2981 of_platform
== PLATFORM_PSERIES_LPAR
)
2982 prom_send_capabilities();
2986 * Copy the CPU hold code
2988 if (of_platform
!= PLATFORM_POWERMAC
)
2989 copy_and_flush(0, kbase
, 0x100, 0);
2992 * Do early parsing of command line
2994 early_cmdline_parse();
2997 * Initialize memory management within prom_init
3002 * Determine which cpu is actually running right _now_
3004 prom_find_boot_cpu();
3007 * Initialize display devices
3009 prom_check_displays();
3011 #if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3013 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3014 * that uses the allocator, we need to make sure we get the top of memory
3015 * available for us here...
3017 if (of_platform
== PLATFORM_PSERIES
)
3018 prom_initialize_tce_table();
3022 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3023 * have a usable RTAS implementation.
3025 if (of_platform
!= PLATFORM_POWERMAC
&&
3026 of_platform
!= PLATFORM_OPAL
)
3027 prom_instantiate_rtas();
3029 #ifdef CONFIG_PPC_POWERNV
3030 #ifdef __BIG_ENDIAN__
3031 /* Detect HAL and try instanciating it & doing takeover */
3032 if (of_platform
== PLATFORM_PSERIES_LPAR
) {
3034 if (of_platform
== PLATFORM_OPAL
) {
3035 prom_opal_hold_cpus();
3036 prom_opal_takeover();
3039 #endif /* __BIG_ENDIAN__ */
3040 if (of_platform
== PLATFORM_OPAL
)
3041 prom_instantiate_opal();
3042 #endif /* CONFIG_PPC_POWERNV */
3045 /* instantiate sml */
3046 prom_instantiate_sml();
3050 * On non-powermacs, put all CPUs in spin-loops.
3052 * PowerMacs use a different mechanism to spin CPUs
3054 * (This must be done after instanciating RTAS)
3056 if (of_platform
!= PLATFORM_POWERMAC
&&
3057 of_platform
!= PLATFORM_OPAL
)
3061 * Fill in some infos for use by the kernel later on
3063 if (prom_memory_limit
) {
3064 __be64 val
= cpu_to_be64(prom_memory_limit
);
3065 prom_setprop(prom
.chosen
, "/chosen", "linux,memory-limit",
3070 prom_setprop(prom
.chosen
, "/chosen", "linux,iommu-off",
3073 if (prom_iommu_force_on
)
3074 prom_setprop(prom
.chosen
, "/chosen", "linux,iommu-force-on",
3077 if (prom_tce_alloc_start
) {
3078 prom_setprop(prom
.chosen
, "/chosen", "linux,tce-alloc-start",
3079 &prom_tce_alloc_start
,
3080 sizeof(prom_tce_alloc_start
));
3081 prom_setprop(prom
.chosen
, "/chosen", "linux,tce-alloc-end",
3082 &prom_tce_alloc_end
,
3083 sizeof(prom_tce_alloc_end
));
3088 * Fixup any known bugs in the device-tree
3090 fixup_device_tree();
3093 * Now finally create the flattened device-tree
3095 prom_printf("copying OF device tree...\n");
3096 flatten_device_tree();
3099 * in case stdin is USB and still active on IBM machines...
3100 * Unfortunately quiesce crashes on some powermacs if we have
3101 * closed stdin already (in particular the powerbook 101). It
3102 * appears that the OPAL version of OFW doesn't like it either.
3104 if (of_platform
!= PLATFORM_POWERMAC
&&
3105 of_platform
!= PLATFORM_OPAL
)
3109 * Call OF "quiesce" method to shut down pending DMA's from
3112 prom_printf("Calling quiesce...\n");
3113 call_prom("quiesce", 0, 0);
3116 * And finally, call the kernel passing it the flattened device
3117 * tree and NULL as r5, thus triggering the new entry point which
3118 * is common to us and kexec
3120 hdr
= dt_header_start
;
3122 /* Don't print anything after quiesce under OPAL, it crashes OFW */
3123 if (of_platform
!= PLATFORM_OPAL
) {
3124 prom_printf("returning from prom_init\n");
3125 prom_debug("->dt_header_start=0x%x\n", hdr
);
3129 reloc_got2(-offset
);
3134 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
3135 /* OPAL early debug gets the OPAL base & entry in r8 and r9 */
3136 __start(hdr
, kbase
, 0, 0, 0,
3137 prom_opal_base
, prom_opal_entry
);
3139 __start(hdr
, kbase
, 0, 0, 0, 0, 0);