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>
40 #include <asm/iommu.h>
41 #include <asm/btext.h>
42 #include <asm/sections.h>
43 #include <asm/machdep.h>
45 #include <asm/asm-prototypes.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);
397 static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
401 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
403 * Old OF requires we claim physical and virtual separately
404 * and then map explicitly (assuming virtual mode)
409 ret = call_prom_ret("call-method", 5, 2, &result,
410 ADDR("claim"), prom.memory,
412 if (ret != 0 || result == -1)
414 ret = call_prom_ret("call-method", 5, 2, &result,
415 ADDR("claim"), prom.mmumap,
418 call_prom("call-method", 4, 1, ADDR("release"),
419 prom.memory, size, virt);
422 /* the 0x12 is M (coherence) + PP == read/write */
423 call_prom("call-method", 6, 1,
424 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
427 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
431 static void __init __attribute__((noreturn)) prom_panic(const char *reason)
434 /* Do not call exit because it clears the screen on pmac
435 * it also causes some sort of double-fault on early pmacs */
436 if (of_platform == PLATFORM_POWERMAC)
439 /* ToDo: should put up an SRC here on pSeries */
440 call_prom("exit", 0, 0);
442 for (;;) /* should never get here */
447 static int __init prom_next_node(phandle *nodep)
451 if ((node = *nodep) != 0
452 && (*nodep = call_prom("child", 1, 1, node)) != 0)
454 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
457 if ((node = call_prom("parent", 1, 1, node)) == 0)
459 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
464 static inline int prom_getprop(phandle node, const char *pname,
465 void *value, size_t valuelen)
467 return call_prom("getprop", 4, 1, node, ADDR(pname),
468 (u32)(unsigned long) value, (u32) valuelen);
471 static inline int prom_getproplen(phandle node, const char *pname)
473 return call_prom("getproplen", 2, 1, node, ADDR(pname));
476 static void add_string(char **str, const char *q)
486 static char *tohex(unsigned int x)
488 static char digits[] = "0123456789abcdef";
489 static char result[9];
496 result[i] = digits[x & 0xf];
498 } while (x != 0 && i > 0);
502 static int __init prom_setprop(phandle node, const char *nodename,
503 const char *pname, void *value, size_t valuelen)
507 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
508 return call_prom("setprop", 4, 1, node, ADDR(pname),
509 (u32)(unsigned long) value, (u32) valuelen);
511 /* gah... setprop doesn't work on longtrail, have to use interpret */
513 add_string(&p, "dev");
514 add_string(&p, nodename);
515 add_string(&p, tohex((u32)(unsigned long) value));
516 add_string(&p, tohex(valuelen));
517 add_string(&p, tohex(ADDR(pname)));
518 add_string(&p, tohex(strlen(pname)));
519 add_string(&p, "property");
521 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
524 /* We can't use the standard versions because of relocation headaches. */
525 #define isxdigit(c) (('0' <= (c) && (c) <= '9') \
526 || ('a' <= (c) && (c) <= 'f') \
527 || ('A' <= (c) && (c) <= 'F'))
529 #define isdigit(c) ('0' <= (c) && (c) <= '9')
530 #define islower(c) ('a' <= (c) && (c) <= 'z')
531 #define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
533 static unsigned long prom_strtoul(const char *cp, const char **endp)
535 unsigned long result = 0, base = 10, value;
540 if (toupper(*cp) == 'X') {
546 while (isxdigit(*cp) &&
547 (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
548 result = result * base + value;
558 static unsigned long prom_memparse(const char *ptr, const char **retptr)
560 unsigned long ret = prom_strtoul(ptr, retptr);
564 * We can't use a switch here because GCC *may* generate a
565 * jump table which won't work, because we're not running at
566 * the address we're linked at.
568 if ('G' == **retptr || 'g' == **retptr)
571 if ('M' == **retptr || 'm' == **retptr)
574 if ('K' == **retptr || 'k' == **retptr)
586 * Early parsing of the command line passed to the kernel, used for
587 * "mem=x" and the options that affect the iommu
589 static void __init early_cmdline_parse(void)
596 prom_cmd_line[0] = 0;
598 if ((long)prom.chosen > 0)
599 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
600 #ifdef CONFIG_CMDLINE
601 if (l <= 0 || p[0] == '\0') /* dbl check */
602 strlcpy(prom_cmd_line,
603 CONFIG_CMDLINE, sizeof(prom_cmd_line));
604 #endif /* CONFIG_CMDLINE */
605 prom_printf("command line: %s\n", prom_cmd_line);
608 opt = strstr(prom_cmd_line, "iommu=");
610 prom_printf("iommu opt is: %s\n", opt);
612 while (*opt && *opt == ' ')
614 if (!strncmp(opt, "off", 3))
616 else if (!strncmp(opt, "force", 5))
617 prom_iommu_force_on = 1;
620 opt = strstr(prom_cmd_line, "mem=");
623 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
625 /* Align to 16 MB == size of ppc64 large page */
626 prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
631 #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
633 * The architecture vector has an array of PVR mask/value pairs,
634 * followed by # option vectors - 1, followed by the option vectors.
636 * See prom.h for the definition of the bits specified in the
637 * architecture vector.
640 /* Firmware expects the value to be n - 1, where n is the # of vectors */
641 #define NUM_VECTORS(n) ((n) - 1)
644 * Firmware expects 1 + n - 2, where n is the length of the option vector in
645 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
647 #define VECTOR_LENGTH(n) (1 + (n) - 2)
649 struct option_vector1 {
654 struct option_vector2 {
668 struct option_vector3 {
673 struct option_vector4 {
678 struct option_vector5 {
690 u8 platform_facilities;
696 struct option_vector6 {
702 struct ibm_arch_vec {
703 struct { u32 mask, val; } pvrs[10];
708 struct option_vector1 vec1;
711 struct option_vector2 vec2;
714 struct option_vector3 vec3;
717 struct option_vector4 vec4;
720 struct option_vector5 vec5;
723 struct option_vector6 vec6;
726 struct ibm_arch_vec __cacheline_aligned ibm_architecture_vec = {
729 .mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
730 .val = cpu_to_be32(0x003a0000),
733 .mask = cpu_to_be32(0xffff0000), /* POWER6 */
734 .val = cpu_to_be32(0x003e0000),
737 .mask = cpu_to_be32(0xffff0000), /* POWER7 */
738 .val = cpu_to_be32(0x003f0000),
741 .mask = cpu_to_be32(0xffff0000), /* POWER8E */
742 .val = cpu_to_be32(0x004b0000),
745 .mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
746 .val = cpu_to_be32(0x004c0000),
749 .mask = cpu_to_be32(0xffff0000), /* POWER8 */
750 .val = cpu_to_be32(0x004d0000),
753 .mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
754 .val = cpu_to_be32(0x0f000004),
757 .mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
758 .val = cpu_to_be32(0x0f000003),
761 .mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
762 .val = cpu_to_be32(0x0f000002),
765 .mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
766 .val = cpu_to_be32(0x0f000001),
770 .num_vectors = NUM_VECTORS(6),
772 .vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
775 .arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
776 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
779 .vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
780 /* option vector 2: Open Firmware options supported */
782 .byte1 = OV2_REAL_MODE,
784 .real_base = cpu_to_be32(0xffffffff),
785 .real_size = cpu_to_be32(0xffffffff),
786 .virt_base = cpu_to_be32(0xffffffff),
787 .virt_size = cpu_to_be32(0xffffffff),
788 .load_base = cpu_to_be32(0xffffffff),
789 .min_rma = cpu_to_be32(256), /* 256MB min RMA */
790 .min_load = cpu_to_be32(0xffffffff), /* full client load */
791 .min_rma_percent = 0, /* min RMA percentage of total RAM */
792 .max_pft_size = 48, /* max log_2(hash table size) */
795 .vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
796 /* option vector 3: processor options supported */
798 .byte1 = 0, /* don't ignore, don't halt */
799 .byte2 = OV3_FP | OV3_VMX | OV3_DFP,
802 .vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
803 /* option vector 4: IBM PAPR implementation */
805 .byte1 = 0, /* don't halt */
806 .min_vp_cap = OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
809 .vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
810 /* option vector 5: PAPR/OF options */
812 .byte1 = 0, /* don't ignore, don't halt */
813 .byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
814 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
815 #ifdef CONFIG_PCI_MSI
816 /* PCIe/MSI support. Without MSI full PCIe is not supported */
823 #ifdef CONFIG_PPC_SMLPAR
824 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
828 .associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
830 .micro_checkpoint = 0,
832 .max_cpus = cpu_to_be32(NR_CPUS), /* number of cores supported */
835 .platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
841 /* option vector 6: IBM PAPR hints */
842 .vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
846 .os_name = OV6_LINUX,
850 /* Old method - ELF header with PT_NOTE sections only works on BE */
851 #ifdef __BIG_ENDIAN__
852 static struct fake_elf {
859 char name[8]; /* "PowerPC" */
873 char name[24]; /* "IBM,RPA-Client-Config" */
887 .e_ident = { 0x7f, 'E', 'L', 'F',
888 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
889 .e_type = ET_EXEC, /* yeah right */
891 .e_version = EV_CURRENT,
892 .e_phoff = offsetof(struct fake_elf, phdr),
893 .e_phentsize = sizeof(Elf32_Phdr),
899 .p_offset = offsetof(struct fake_elf, chrpnote),
900 .p_filesz = sizeof(struct chrpnote)
903 .p_offset = offsetof(struct fake_elf, rpanote),
904 .p_filesz = sizeof(struct rpanote)
908 .namesz = sizeof("PowerPC"),
909 .descsz = sizeof(struct chrpdesc),
913 .real_mode = ~0U, /* ~0 means "don't care" */
922 .namesz = sizeof("IBM,RPA-Client-Config"),
923 .descsz = sizeof(struct rpadesc),
925 .name = "IBM,RPA-Client-Config",
928 .min_rmo_size = 64, /* in megabytes */
929 .min_rmo_percent = 0,
930 .max_pft_size = 48, /* 2^48 bytes max PFT size */
937 #endif /* __BIG_ENDIAN__ */
939 static int __init prom_count_smt_threads(void)
945 /* Pick up th first CPU node we can find */
946 for (node = 0; prom_next_node(&node); ) {
948 prom_getprop(node, "device_type", type, sizeof(type));
950 if (strcmp(type, "cpu"))
953 * There is an entry for each smt thread, each entry being
954 * 4 bytes long. All cpus should have the same number of
955 * smt threads, so return after finding the first.
957 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
958 if (plen == PROM_ERROR)
961 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
964 if (plen < 1 || plen > 64) {
965 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
966 (unsigned long)plen);
971 prom_debug("No threads found, assuming 1 per core\n");
978 static void __init prom_send_capabilities(void)
984 root = call_prom("open", 1, 1, ADDR("/"));
986 /* We need to tell the FW about the number of cores we support.
988 * To do that, we count the number of threads on the first core
989 * (we assume this is the same for all cores) and use it to
993 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
994 prom_printf("Max number of cores passed to firmware: %lu (NR_CPUS = %lu)\n",
997 ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
999 /* try calling the ibm,client-architecture-support method */
1000 prom_printf("Calling ibm,client-architecture-support...");
1001 if (call_prom_ret("call-method", 3, 2, &ret,
1002 ADDR("ibm,client-architecture-support"),
1004 ADDR(&ibm_architecture_vec)) == 0) {
1005 /* the call exists... */
1007 prom_printf("\nWARNING: ibm,client-architecture"
1008 "-support call FAILED!\n");
1009 call_prom("close", 1, 0, root);
1010 prom_printf(" done\n");
1013 call_prom("close", 1, 0, root);
1014 prom_printf(" not implemented\n");
1017 #ifdef __BIG_ENDIAN__
1021 /* no ibm,client-architecture-support call, try the old way */
1022 elfloader = call_prom("open", 1, 1,
1023 ADDR("/packages/elf-loader"));
1024 if (elfloader == 0) {
1025 prom_printf("couldn't open /packages/elf-loader\n");
1028 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1029 elfloader, ADDR(&fake_elf));
1030 call_prom("close", 1, 0, elfloader);
1032 #endif /* __BIG_ENDIAN__ */
1034 #endif /* #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV) */
1037 * Memory allocation strategy... our layout is normally:
1039 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
1040 * rare cases, initrd might end up being before the kernel though.
1041 * We assume this won't override the final kernel at 0, we have no
1042 * provision to handle that in this version, but it should hopefully
1045 * alloc_top is set to the top of RMO, eventually shrink down if the
1048 * alloc_bottom is set to the top of kernel/initrd
1050 * from there, allocations are done this way : rtas is allocated
1051 * topmost, and the device-tree is allocated from the bottom. We try
1052 * to grow the device-tree allocation as we progress. If we can't,
1053 * then we fail, we don't currently have a facility to restart
1054 * elsewhere, but that shouldn't be necessary.
1056 * Note that calls to reserve_mem have to be done explicitly, memory
1057 * allocated with either alloc_up or alloc_down isn't automatically
1063 * Allocates memory in the RMO upward from the kernel/initrd
1065 * When align is 0, this is a special case, it means to allocate in place
1066 * at the current location of alloc_bottom or fail (that is basically
1067 * extending the previous allocation). Used for the device-tree flattening
1069 static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1071 unsigned long base = alloc_bottom;
1072 unsigned long addr = 0;
1075 base = _ALIGN_UP(base, align);
1076 prom_debug("alloc_up(%x, %x)\n", size, align);
1078 prom_panic("alloc_up() called with mem not initialized\n");
1081 base = _ALIGN_UP(alloc_bottom, align);
1083 base = alloc_bottom;
1085 for(; (base + size) <= alloc_top;
1086 base = _ALIGN_UP(base + 0x100000, align)) {
1087 prom_debug(" trying: 0x%x\n\r", base);
1088 addr = (unsigned long)prom_claim(base, size, 0);
1089 if (addr != PROM_ERROR && addr != 0)
1097 alloc_bottom = addr + size;
1099 prom_debug(" -> %x\n", addr);
1100 prom_debug(" alloc_bottom : %x\n", alloc_bottom);
1101 prom_debug(" alloc_top : %x\n", alloc_top);
1102 prom_debug(" alloc_top_hi : %x\n", alloc_top_high);
1103 prom_debug(" rmo_top : %x\n", rmo_top);
1104 prom_debug(" ram_top : %x\n", ram_top);
1110 * Allocates memory downward, either from top of RMO, or if highmem
1111 * is set, from the top of RAM. Note that this one doesn't handle
1112 * failures. It does claim memory if highmem is not set.
1114 static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1117 unsigned long base, addr = 0;
1119 prom_debug("alloc_down(%x, %x, %s)\n", size, align,
1120 highmem ? "(high)" : "(low)");
1122 prom_panic("alloc_down() called with mem not initialized\n");
1125 /* Carve out storage for the TCE table. */
1126 addr = _ALIGN_DOWN(alloc_top_high - size, align);
1127 if (addr <= alloc_bottom)
1129 /* Will we bump into the RMO ? If yes, check out that we
1130 * didn't overlap existing allocations there, if we did,
1131 * we are dead, we must be the first in town !
1133 if (addr < rmo_top) {
1134 /* Good, we are first */
1135 if (alloc_top == rmo_top)
1136 alloc_top = rmo_top = addr;
1140 alloc_top_high = addr;
1144 base = _ALIGN_DOWN(alloc_top - size, align);
1145 for (; base > alloc_bottom;
1146 base = _ALIGN_DOWN(base - 0x100000, align)) {
1147 prom_debug(" trying: 0x%x\n\r", base);
1148 addr = (unsigned long)prom_claim(base, size, 0);
1149 if (addr != PROM_ERROR && addr != 0)
1158 prom_debug(" -> %x\n", addr);
1159 prom_debug(" alloc_bottom : %x\n", alloc_bottom);
1160 prom_debug(" alloc_top : %x\n", alloc_top);
1161 prom_debug(" alloc_top_hi : %x\n", alloc_top_high);
1162 prom_debug(" rmo_top : %x\n", rmo_top);
1163 prom_debug(" ram_top : %x\n", ram_top);
1169 * Parse a "reg" cell
1171 static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1174 unsigned long r = 0;
1176 /* Ignore more than 2 cells */
1177 while (s > sizeof(unsigned long) / 4) {
1181 r = be32_to_cpu(*p++);
1185 r |= be32_to_cpu(*(p++));
1193 * Very dumb function for adding to the memory reserve list, but
1194 * we don't need anything smarter at this point
1196 * XXX Eventually check for collisions. They should NEVER happen.
1197 * If problems seem to show up, it would be a good start to track
1200 static void __init reserve_mem(u64 base, u64 size)
1202 u64 top = base + size;
1203 unsigned long cnt = mem_reserve_cnt;
1208 /* We need to always keep one empty entry so that we
1209 * have our terminator with "size" set to 0 since we are
1210 * dumb and just copy this entire array to the boot params
1212 base = _ALIGN_DOWN(base, PAGE_SIZE);
1213 top = _ALIGN_UP(top, PAGE_SIZE);
1216 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1217 prom_panic("Memory reserve map exhausted !\n");
1218 mem_reserve_map[cnt].base = cpu_to_be64(base);
1219 mem_reserve_map[cnt].size = cpu_to_be64(size);
1220 mem_reserve_cnt = cnt + 1;
1224 * Initialize memory allocation mechanism, parse "memory" nodes and
1225 * obtain that way the top of memory and RMO to setup out local allocator
1227 static void __init prom_init_mem(void)
1230 char *path, type[64];
1237 * We iterate the memory nodes to find
1238 * 1) top of RMO (first node)
1241 val = cpu_to_be32(2);
1242 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1243 rac = be32_to_cpu(val);
1244 val = cpu_to_be32(1);
1245 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1246 rsc = be32_to_cpu(val);
1247 prom_debug("root_addr_cells: %x\n", rac);
1248 prom_debug("root_size_cells: %x\n", rsc);
1250 prom_debug("scanning memory:\n");
1251 path = prom_scratch;
1253 for (node = 0; prom_next_node(&node); ) {
1255 prom_getprop(node, "device_type", type, sizeof(type));
1259 * CHRP Longtrail machines have no device_type
1260 * on the memory node, so check the name instead...
1262 prom_getprop(node, "name", type, sizeof(type));
1264 if (strcmp(type, "memory"))
1267 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1268 if (plen > sizeof(regbuf)) {
1269 prom_printf("memory node too large for buffer !\n");
1270 plen = sizeof(regbuf);
1273 endp = p + (plen / sizeof(cell_t));
1276 memset(path, 0, PROM_SCRATCH_SIZE);
1277 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1278 prom_debug(" node %s :\n", path);
1279 #endif /* DEBUG_PROM */
1281 while ((endp - p) >= (rac + rsc)) {
1282 unsigned long base, size;
1284 base = prom_next_cell(rac, &p);
1285 size = prom_next_cell(rsc, &p);
1289 prom_debug(" %x %x\n", base, size);
1290 if (base == 0 && (of_platform & PLATFORM_LPAR))
1292 if ((base + size) > ram_top)
1293 ram_top = base + size;
1297 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1300 * If prom_memory_limit is set we reduce the upper limits *except* for
1301 * alloc_top_high. This must be the real top of RAM so we can put
1305 alloc_top_high = ram_top;
1307 if (prom_memory_limit) {
1308 if (prom_memory_limit <= alloc_bottom) {
1309 prom_printf("Ignoring mem=%x <= alloc_bottom.\n",
1311 prom_memory_limit = 0;
1312 } else if (prom_memory_limit >= ram_top) {
1313 prom_printf("Ignoring mem=%x >= ram_top.\n",
1315 prom_memory_limit = 0;
1317 ram_top = prom_memory_limit;
1318 rmo_top = min(rmo_top, prom_memory_limit);
1323 * Setup our top alloc point, that is top of RMO or top of
1324 * segment 0 when running non-LPAR.
1325 * Some RS64 machines have buggy firmware where claims up at
1326 * 1GB fail. Cap at 768MB as a workaround.
1327 * Since 768MB is plenty of room, and we need to cap to something
1328 * reasonable on 32-bit, cap at 768MB on all machines.
1332 rmo_top = min(0x30000000ul, rmo_top);
1333 alloc_top = rmo_top;
1334 alloc_top_high = ram_top;
1337 * Check if we have an initrd after the kernel but still inside
1338 * the RMO. If we do move our bottom point to after it.
1340 if (prom_initrd_start &&
1341 prom_initrd_start < rmo_top &&
1342 prom_initrd_end > alloc_bottom)
1343 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1345 prom_printf("memory layout at init:\n");
1346 prom_printf(" memory_limit : %x (16 MB aligned)\n", prom_memory_limit);
1347 prom_printf(" alloc_bottom : %x\n", alloc_bottom);
1348 prom_printf(" alloc_top : %x\n", alloc_top);
1349 prom_printf(" alloc_top_hi : %x\n", alloc_top_high);
1350 prom_printf(" rmo_top : %x\n", rmo_top);
1351 prom_printf(" ram_top : %x\n", ram_top);
1354 static void __init prom_close_stdin(void)
1359 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1360 stdin = be32_to_cpu(val);
1361 call_prom("close", 1, 0, stdin);
1365 #ifdef CONFIG_PPC_POWERNV
1367 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1368 static u64 __initdata prom_opal_base;
1369 static u64 __initdata prom_opal_entry;
1373 * Allocate room for and instantiate OPAL
1375 static void __init prom_instantiate_opal(void)
1380 u64 size = 0, align = 0x10000;
1384 prom_debug("prom_instantiate_opal: start...\n");
1386 opal_node = call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1387 prom_debug("opal_node: %x\n", opal_node);
1388 if (!PHANDLE_VALID(opal_node))
1392 prom_getprop(opal_node, "opal-runtime-size", &val64, sizeof(val64));
1393 size = be64_to_cpu(val64);
1397 prom_getprop(opal_node, "opal-runtime-alignment", &val64,sizeof(val64));
1398 align = be64_to_cpu(val64);
1400 base = alloc_down(size, align, 0);
1402 prom_printf("OPAL allocation failed !\n");
1406 opal_inst = call_prom("open", 1, 1, ADDR("/ibm,opal"));
1407 if (!IHANDLE_VALID(opal_inst)) {
1408 prom_printf("opening opal package failed (%x)\n", opal_inst);
1412 prom_printf("instantiating opal at 0x%x...", base);
1414 if (call_prom_ret("call-method", 4, 3, rets,
1415 ADDR("load-opal-runtime"),
1417 base >> 32, base & 0xffffffff) != 0
1418 || (rets[0] == 0 && rets[1] == 0)) {
1419 prom_printf(" failed\n");
1422 entry = (((u64)rets[0]) << 32) | rets[1];
1424 prom_printf(" done\n");
1426 reserve_mem(base, size);
1428 prom_debug("opal base = 0x%x\n", base);
1429 prom_debug("opal align = 0x%x\n", align);
1430 prom_debug("opal entry = 0x%x\n", entry);
1431 prom_debug("opal size = 0x%x\n", (long)size);
1433 prom_setprop(opal_node, "/ibm,opal", "opal-base-address",
1434 &base, sizeof(base));
1435 prom_setprop(opal_node, "/ibm,opal", "opal-entry-address",
1436 &entry, sizeof(entry));
1438 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1439 prom_opal_base = base;
1440 prom_opal_entry = entry;
1442 prom_debug("prom_instantiate_opal: end...\n");
1445 #endif /* CONFIG_PPC_POWERNV */
1448 * Allocate room for and instantiate RTAS
1450 static void __init prom_instantiate_rtas(void)
1454 u32 base, entry = 0;
1458 prom_debug("prom_instantiate_rtas: start...\n");
1460 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1461 prom_debug("rtas_node: %x\n", rtas_node);
1462 if (!PHANDLE_VALID(rtas_node))
1466 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1467 size = be32_to_cpu(val);
1471 base = alloc_down(size, PAGE_SIZE, 0);
1473 prom_panic("Could not allocate memory for RTAS\n");
1475 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1476 if (!IHANDLE_VALID(rtas_inst)) {
1477 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1481 prom_printf("instantiating rtas at 0x%x...", base);
1483 if (call_prom_ret("call-method", 3, 2, &entry,
1484 ADDR("instantiate-rtas"),
1485 rtas_inst, base) != 0
1487 prom_printf(" failed\n");
1490 prom_printf(" done\n");
1492 reserve_mem(base, size);
1494 val = cpu_to_be32(base);
1495 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1497 val = cpu_to_be32(entry);
1498 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1501 /* Check if it supports "query-cpu-stopped-state" */
1502 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1503 &val, sizeof(val)) != PROM_ERROR)
1504 rtas_has_query_cpu_stopped = true;
1506 prom_debug("rtas base = 0x%x\n", base);
1507 prom_debug("rtas entry = 0x%x\n", entry);
1508 prom_debug("rtas size = 0x%x\n", (long)size);
1510 prom_debug("prom_instantiate_rtas: end...\n");
1515 * Allocate room for and instantiate Stored Measurement Log (SML)
1517 static void __init prom_instantiate_sml(void)
1519 phandle ibmvtpm_node;
1520 ihandle ibmvtpm_inst;
1521 u32 entry = 0, size = 0, succ = 0;
1525 prom_debug("prom_instantiate_sml: start...\n");
1527 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1528 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1529 if (!PHANDLE_VALID(ibmvtpm_node))
1532 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1533 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1534 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1538 if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1539 &val, sizeof(val)) != PROM_ERROR) {
1540 if (call_prom_ret("call-method", 2, 2, &succ,
1541 ADDR("reformat-sml-to-efi-alignment"),
1542 ibmvtpm_inst) != 0 || succ == 0) {
1543 prom_printf("Reformat SML to EFI alignment failed\n");
1547 if (call_prom_ret("call-method", 2, 2, &size,
1548 ADDR("sml-get-allocated-size"),
1549 ibmvtpm_inst) != 0 || size == 0) {
1550 prom_printf("SML get allocated size failed\n");
1554 if (call_prom_ret("call-method", 2, 2, &size,
1555 ADDR("sml-get-handover-size"),
1556 ibmvtpm_inst) != 0 || size == 0) {
1557 prom_printf("SML get handover size failed\n");
1562 base = alloc_down(size, PAGE_SIZE, 0);
1564 prom_panic("Could not allocate memory for sml\n");
1566 prom_printf("instantiating sml at 0x%x...", base);
1568 memset((void *)base, 0, size);
1570 if (call_prom_ret("call-method", 4, 2, &entry,
1571 ADDR("sml-handover"),
1572 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1573 prom_printf("SML handover failed\n");
1576 prom_printf(" done\n");
1578 reserve_mem(base, size);
1580 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1581 &base, sizeof(base));
1582 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1583 &size, sizeof(size));
1585 prom_debug("sml base = 0x%x\n", base);
1586 prom_debug("sml size = 0x%x\n", (long)size);
1588 prom_debug("prom_instantiate_sml: end...\n");
1592 * Allocate room for and initialize TCE tables
1594 #ifdef __BIG_ENDIAN__
1595 static void __init prom_initialize_tce_table(void)
1599 char compatible[64], type[64], model[64];
1600 char *path = prom_scratch;
1602 u32 minalign, minsize;
1603 u64 tce_entry, *tce_entryp;
1604 u64 local_alloc_top, local_alloc_bottom;
1610 prom_debug("starting prom_initialize_tce_table\n");
1612 /* Cache current top of allocs so we reserve a single block */
1613 local_alloc_top = alloc_top_high;
1614 local_alloc_bottom = local_alloc_top;
1616 /* Search all nodes looking for PHBs. */
1617 for (node = 0; prom_next_node(&node); ) {
1621 prom_getprop(node, "compatible",
1622 compatible, sizeof(compatible));
1623 prom_getprop(node, "device_type", type, sizeof(type));
1624 prom_getprop(node, "model", model, sizeof(model));
1626 if ((type[0] == 0) || (strstr(type, "pci") == NULL))
1629 /* Keep the old logic intact to avoid regression. */
1630 if (compatible[0] != 0) {
1631 if ((strstr(compatible, "python") == NULL) &&
1632 (strstr(compatible, "Speedwagon") == NULL) &&
1633 (strstr(compatible, "Winnipeg") == NULL))
1635 } else if (model[0] != 0) {
1636 if ((strstr(model, "ython") == NULL) &&
1637 (strstr(model, "peedwagon") == NULL) &&
1638 (strstr(model, "innipeg") == NULL))
1642 if (prom_getprop(node, "tce-table-minalign", &minalign,
1643 sizeof(minalign)) == PROM_ERROR)
1645 if (prom_getprop(node, "tce-table-minsize", &minsize,
1646 sizeof(minsize)) == PROM_ERROR)
1647 minsize = 4UL << 20;
1650 * Even though we read what OF wants, we just set the table
1651 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
1652 * By doing this, we avoid the pitfalls of trying to DMA to
1653 * MMIO space and the DMA alias hole.
1655 * On POWER4, firmware sets the TCE region by assuming
1656 * each TCE table is 8MB. Using this memory for anything
1657 * else will impact performance, so we always allocate 8MB.
1660 if (pvr_version_is(PVR_POWER4) || pvr_version_is(PVR_POWER4p))
1661 minsize = 8UL << 20;
1663 minsize = 4UL << 20;
1665 /* Align to the greater of the align or size */
1666 align = max(minalign, minsize);
1667 base = alloc_down(minsize, align, 1);
1669 prom_panic("ERROR, cannot find space for TCE table.\n");
1670 if (base < local_alloc_bottom)
1671 local_alloc_bottom = base;
1673 /* It seems OF doesn't null-terminate the path :-( */
1674 memset(path, 0, PROM_SCRATCH_SIZE);
1675 /* Call OF to setup the TCE hardware */
1676 if (call_prom("package-to-path", 3, 1, node,
1677 path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1678 prom_printf("package-to-path failed\n");
1681 /* Save away the TCE table attributes for later use. */
1682 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1683 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1685 prom_debug("TCE table: %s\n", path);
1686 prom_debug("\tnode = 0x%x\n", node);
1687 prom_debug("\tbase = 0x%x\n", base);
1688 prom_debug("\tsize = 0x%x\n", minsize);
1690 /* Initialize the table to have a one-to-one mapping
1691 * over the allocated size.
1693 tce_entryp = (u64 *)base;
1694 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1695 tce_entry = (i << PAGE_SHIFT);
1697 *tce_entryp = tce_entry;
1700 prom_printf("opening PHB %s", path);
1701 phb_node = call_prom("open", 1, 1, path);
1703 prom_printf("... failed\n");
1705 prom_printf("... done\n");
1707 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1708 phb_node, -1, minsize,
1709 (u32) base, (u32) (base >> 32));
1710 call_prom("close", 1, 0, phb_node);
1713 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1715 /* These are only really needed if there is a memory limit in
1716 * effect, but we don't know so export them always. */
1717 prom_tce_alloc_start = local_alloc_bottom;
1718 prom_tce_alloc_end = local_alloc_top;
1720 /* Flag the first invalid entry */
1721 prom_debug("ending prom_initialize_tce_table\n");
1723 #endif /* __BIG_ENDIAN__ */
1724 #endif /* CONFIG_PPC64 */
1727 * With CHRP SMP we need to use the OF to start the other processors.
1728 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1729 * so we have to put the processors into a holding pattern controlled
1730 * by the kernel (not OF) before we destroy the OF.
1732 * This uses a chunk of low memory, puts some holding pattern
1733 * code there and sends the other processors off to there until
1734 * smp_boot_cpus tells them to do something. The holding pattern
1735 * checks that address until its cpu # is there, when it is that
1736 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
1737 * of setting those values.
1739 * We also use physical address 0x4 here to tell when a cpu
1740 * is in its holding pattern code.
1745 * We want to reference the copy of __secondary_hold_* in the
1746 * 0 - 0x100 address range
1748 #define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
1750 static void __init prom_hold_cpus(void)
1755 unsigned long *spinloop
1756 = (void *) LOW_ADDR(__secondary_hold_spinloop);
1757 unsigned long *acknowledge
1758 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
1759 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1762 * On pseries, if RTAS supports "query-cpu-stopped-state",
1763 * we skip this stage, the CPUs will be started by the
1764 * kernel using RTAS.
1766 if ((of_platform == PLATFORM_PSERIES ||
1767 of_platform == PLATFORM_PSERIES_LPAR) &&
1768 rtas_has_query_cpu_stopped) {
1769 prom_printf("prom_hold_cpus: skipped\n");
1773 prom_debug("prom_hold_cpus: start...\n");
1774 prom_debug(" 1) spinloop = 0x%x\n", (unsigned long)spinloop);
1775 prom_debug(" 1) *spinloop = 0x%x\n", *spinloop);
1776 prom_debug(" 1) acknowledge = 0x%x\n",
1777 (unsigned long)acknowledge);
1778 prom_debug(" 1) *acknowledge = 0x%x\n", *acknowledge);
1779 prom_debug(" 1) secondary_hold = 0x%x\n", secondary_hold);
1781 /* Set the common spinloop variable, so all of the secondary cpus
1782 * will block when they are awakened from their OF spinloop.
1783 * This must occur for both SMP and non SMP kernels, since OF will
1784 * be trashed when we move the kernel.
1789 for (node = 0; prom_next_node(&node); ) {
1790 unsigned int cpu_no;
1794 prom_getprop(node, "device_type", type, sizeof(type));
1795 if (strcmp(type, "cpu") != 0)
1798 /* Skip non-configured cpus. */
1799 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1800 if (strcmp(type, "okay") != 0)
1803 reg = cpu_to_be32(-1); /* make sparse happy */
1804 prom_getprop(node, "reg", ®, sizeof(reg));
1805 cpu_no = be32_to_cpu(reg);
1807 prom_debug("cpu hw idx = %lu\n", cpu_no);
1809 /* Init the acknowledge var which will be reset by
1810 * the secondary cpu when it awakens from its OF
1813 *acknowledge = (unsigned long)-1;
1815 if (cpu_no != prom.cpu) {
1816 /* Primary Thread of non-boot cpu or any thread */
1817 prom_printf("starting cpu hw idx %lu... ", cpu_no);
1818 call_prom("start-cpu", 3, 0, node,
1819 secondary_hold, cpu_no);
1821 for (i = 0; (i < 100000000) &&
1822 (*acknowledge == ((unsigned long)-1)); i++ )
1825 if (*acknowledge == cpu_no)
1826 prom_printf("done\n");
1828 prom_printf("failed: %x\n", *acknowledge);
1832 prom_printf("boot cpu hw idx %lu\n", cpu_no);
1833 #endif /* CONFIG_SMP */
1836 prom_debug("prom_hold_cpus: end...\n");
1840 static void __init prom_init_client_services(unsigned long pp)
1842 /* Get a handle to the prom entry point before anything else */
1845 /* get a handle for the stdout device */
1846 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
1847 if (!PHANDLE_VALID(prom.chosen))
1848 prom_panic("cannot find chosen"); /* msg won't be printed :( */
1850 /* get device tree root */
1851 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
1852 if (!PHANDLE_VALID(prom.root))
1853 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1860 * For really old powermacs, we need to map things we claim.
1861 * For that, we need the ihandle of the mmu.
1862 * Also, on the longtrail, we need to work around other bugs.
1864 static void __init prom_find_mmu(void)
1869 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
1870 if (!PHANDLE_VALID(oprom))
1872 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
1874 version[sizeof(version) - 1] = 0;
1875 /* XXX might need to add other versions here */
1876 if (strcmp(version, "Open Firmware, 1.0.5") == 0)
1877 of_workarounds = OF_WA_CLAIM;
1878 else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
1879 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
1880 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1883 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
1884 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
1885 sizeof(prom.mmumap));
1886 prom.mmumap = be32_to_cpu(prom.mmumap);
1887 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
1888 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
1891 #define prom_find_mmu()
1894 static void __init prom_init_stdout(void)
1896 char *path = of_stdout_device;
1898 phandle stdout_node;
1901 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
1902 prom_panic("cannot find stdout");
1904 prom.stdout = be32_to_cpu(val);
1906 /* Get the full OF pathname of the stdout device */
1907 memset(path, 0, 256);
1908 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
1909 prom_printf("OF stdout device is: %s\n", of_stdout_device);
1910 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
1911 path, strlen(path) + 1);
1913 /* instance-to-package fails on PA-Semi */
1914 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
1915 if (stdout_node != PROM_ERROR) {
1916 val = cpu_to_be32(stdout_node);
1917 prom_setprop(prom.chosen, "/chosen", "linux,stdout-package",
1920 /* If it's a display, note it */
1921 memset(type, 0, sizeof(type));
1922 prom_getprop(stdout_node, "device_type", type, sizeof(type));
1923 if (strcmp(type, "display") == 0)
1924 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
1928 static int __init prom_find_machine_type(void)
1937 /* Look for a PowerMac or a Cell */
1938 len = prom_getprop(prom.root, "compatible",
1939 compat, sizeof(compat)-1);
1943 char *p = &compat[i];
1947 if (strstr(p, "Power Macintosh") ||
1948 strstr(p, "MacRISC"))
1949 return PLATFORM_POWERMAC;
1951 /* We must make sure we don't detect the IBM Cell
1952 * blades as pSeries due to some firmware issues,
1955 if (strstr(p, "IBM,CBEA") ||
1956 strstr(p, "IBM,CPBW-1.0"))
1957 return PLATFORM_GENERIC;
1958 #endif /* CONFIG_PPC64 */
1963 /* Try to detect OPAL */
1964 if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
1965 return PLATFORM_OPAL;
1967 /* Try to figure out if it's an IBM pSeries or any other
1968 * PAPR compliant platform. We assume it is if :
1969 * - /device_type is "chrp" (please, do NOT use that for future
1973 len = prom_getprop(prom.root, "device_type",
1974 compat, sizeof(compat)-1);
1976 return PLATFORM_GENERIC;
1977 if (strcmp(compat, "chrp"))
1978 return PLATFORM_GENERIC;
1980 /* Default to pSeries. We need to know if we are running LPAR */
1981 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1982 if (!PHANDLE_VALID(rtas))
1983 return PLATFORM_GENERIC;
1984 x = prom_getproplen(rtas, "ibm,hypertas-functions");
1985 if (x != PROM_ERROR) {
1986 prom_debug("Hypertas detected, assuming LPAR !\n");
1987 return PLATFORM_PSERIES_LPAR;
1989 return PLATFORM_PSERIES;
1991 return PLATFORM_GENERIC;
1995 static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
1997 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2001 * If we have a display that we don't know how to drive,
2002 * we will want to try to execute OF's open method for it
2003 * later. However, OF will probably fall over if we do that
2004 * we've taken over the MMU.
2005 * So we check whether we will need to open the display,
2006 * and if so, open it now.
2008 static void __init prom_check_displays(void)
2010 char type[16], *path;
2015 static unsigned char default_colors[] = {
2033 const unsigned char *clut;
2035 prom_debug("Looking for displays\n");
2036 for (node = 0; prom_next_node(&node); ) {
2037 memset(type, 0, sizeof(type));
2038 prom_getprop(node, "device_type", type, sizeof(type));
2039 if (strcmp(type, "display") != 0)
2042 /* It seems OF doesn't null-terminate the path :-( */
2043 path = prom_scratch;
2044 memset(path, 0, PROM_SCRATCH_SIZE);
2047 * leave some room at the end of the path for appending extra
2050 if (call_prom("package-to-path", 3, 1, node, path,
2051 PROM_SCRATCH_SIZE-10) == PROM_ERROR)
2053 prom_printf("found display : %s, opening... ", path);
2055 ih = call_prom("open", 1, 1, path);
2057 prom_printf("failed\n");
2062 prom_printf("done\n");
2063 prom_setprop(node, path, "linux,opened", NULL, 0);
2065 /* Setup a usable color table when the appropriate
2066 * method is available. Should update this to set-colors */
2067 clut = default_colors;
2068 for (i = 0; i < 16; i++, clut += 3)
2069 if (prom_set_color(ih, i, clut[0], clut[1],
2073 #ifdef CONFIG_LOGO_LINUX_CLUT224
2074 clut = PTRRELOC(logo_linux_clut224.clut);
2075 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2076 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2079 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
2081 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2082 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2084 u32 width, height, pitch, addr;
2086 prom_printf("Setting btext !\n");
2087 prom_getprop(node, "width", &width, 4);
2088 prom_getprop(node, "height", &height, 4);
2089 prom_getprop(node, "linebytes", &pitch, 4);
2090 prom_getprop(node, "address", &addr, 4);
2091 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2092 width, height, pitch, addr);
2093 btext_setup_display(width, height, 8, pitch, addr);
2095 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2100 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2101 static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2102 unsigned long needed, unsigned long align)
2106 *mem_start = _ALIGN(*mem_start, align);
2107 while ((*mem_start + needed) > *mem_end) {
2108 unsigned long room, chunk;
2110 prom_debug("Chunk exhausted, claiming more at %x...\n",
2112 room = alloc_top - alloc_bottom;
2113 if (room > DEVTREE_CHUNK_SIZE)
2114 room = DEVTREE_CHUNK_SIZE;
2115 if (room < PAGE_SIZE)
2116 prom_panic("No memory for flatten_device_tree "
2118 chunk = alloc_up(room, 0);
2120 prom_panic("No memory for flatten_device_tree "
2121 "(claim failed)\n");
2122 *mem_end = chunk + room;
2125 ret = (void *)*mem_start;
2126 *mem_start += needed;
2131 #define dt_push_token(token, mem_start, mem_end) do { \
2132 void *room = make_room(mem_start, mem_end, 4, 4); \
2133 *(__be32 *)room = cpu_to_be32(token); \
2136 static unsigned long __init dt_find_string(char *str)
2140 s = os = (char *)dt_string_start;
2142 while (s < (char *)dt_string_end) {
2143 if (strcmp(s, str) == 0)
2151 * The Open Firmware 1275 specification states properties must be 31 bytes or
2152 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2154 #define MAX_PROPERTY_NAME 64
2156 static void __init scan_dt_build_strings(phandle node,
2157 unsigned long *mem_start,
2158 unsigned long *mem_end)
2160 char *prev_name, *namep, *sstart;
2164 sstart = (char *)dt_string_start;
2166 /* get and store all property names */
2169 /* 64 is max len of name including nul. */
2170 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2171 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2172 /* No more nodes: unwind alloc */
2173 *mem_start = (unsigned long)namep;
2178 if (strcmp(namep, "name") == 0) {
2179 *mem_start = (unsigned long)namep;
2183 /* get/create string entry */
2184 soff = dt_find_string(namep);
2186 *mem_start = (unsigned long)namep;
2187 namep = sstart + soff;
2189 /* Trim off some if we can */
2190 *mem_start = (unsigned long)namep + strlen(namep) + 1;
2191 dt_string_end = *mem_start;
2196 /* do all our children */
2197 child = call_prom("child", 1, 1, node);
2198 while (child != 0) {
2199 scan_dt_build_strings(child, mem_start, mem_end);
2200 child = call_prom("peer", 1, 1, child);
2204 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2205 unsigned long *mem_end)
2208 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2210 unsigned char *valp;
2211 static char pname[MAX_PROPERTY_NAME];
2212 int l, room, has_phandle = 0;
2214 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2216 /* get the node's full name */
2217 namep = (char *)*mem_start;
2218 room = *mem_end - *mem_start;
2221 l = call_prom("package-to-path", 3, 1, node, namep, room);
2223 /* Didn't fit? Get more room. */
2225 if (l >= *mem_end - *mem_start)
2226 namep = make_room(mem_start, mem_end, l+1, 1);
2227 call_prom("package-to-path", 3, 1, node, namep, l);
2231 /* Fixup an Apple bug where they have bogus \0 chars in the
2232 * middle of the path in some properties, and extract
2233 * the unit name (everything after the last '/').
2235 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2242 *mem_start = _ALIGN((unsigned long)lp + 1, 4);
2245 /* get it again for debugging */
2246 path = prom_scratch;
2247 memset(path, 0, PROM_SCRATCH_SIZE);
2248 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
2250 /* get and store all properties */
2252 sstart = (char *)dt_string_start;
2254 if (call_prom("nextprop", 3, 1, node, prev_name,
2259 if (strcmp(pname, "name") == 0) {
2264 /* find string offset */
2265 soff = dt_find_string(pname);
2267 prom_printf("WARNING: Can't find string index for"
2268 " <%s>, node %s\n", pname, path);
2271 prev_name = sstart + soff;
2274 l = call_prom("getproplen", 2, 1, node, pname);
2277 if (l == PROM_ERROR)
2280 /* push property head */
2281 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2282 dt_push_token(l, mem_start, mem_end);
2283 dt_push_token(soff, mem_start, mem_end);
2285 /* push property content */
2286 valp = make_room(mem_start, mem_end, l, 4);
2287 call_prom("getprop", 4, 1, node, pname, valp, l);
2288 *mem_start = _ALIGN(*mem_start, 4);
2290 if (!strcmp(pname, "phandle"))
2294 /* Add a "linux,phandle" property if no "phandle" property already
2295 * existed (can happen with OPAL)
2298 soff = dt_find_string("linux,phandle");
2300 prom_printf("WARNING: Can't find string index for"
2301 " <linux-phandle> node %s\n", path);
2303 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2304 dt_push_token(4, mem_start, mem_end);
2305 dt_push_token(soff, mem_start, mem_end);
2306 valp = make_room(mem_start, mem_end, 4, 4);
2307 *(__be32 *)valp = cpu_to_be32(node);
2311 /* do all our children */
2312 child = call_prom("child", 1, 1, node);
2313 while (child != 0) {
2314 scan_dt_build_struct(child, mem_start, mem_end);
2315 child = call_prom("peer", 1, 1, child);
2318 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2321 static void __init flatten_device_tree(void)
2324 unsigned long mem_start, mem_end, room;
2325 struct boot_param_header *hdr;
2330 * Check how much room we have between alloc top & bottom (+/- a
2331 * few pages), crop to 1MB, as this is our "chunk" size
2333 room = alloc_top - alloc_bottom - 0x4000;
2334 if (room > DEVTREE_CHUNK_SIZE)
2335 room = DEVTREE_CHUNK_SIZE;
2336 prom_debug("starting device tree allocs at %x\n", alloc_bottom);
2338 /* Now try to claim that */
2339 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2341 prom_panic("Can't allocate initial device-tree chunk\n");
2342 mem_end = mem_start + room;
2344 /* Get root of tree */
2345 root = call_prom("peer", 1, 1, (phandle)0);
2346 if (root == (phandle)0)
2347 prom_panic ("couldn't get device tree root\n");
2349 /* Build header and make room for mem rsv map */
2350 mem_start = _ALIGN(mem_start, 4);
2351 hdr = make_room(&mem_start, &mem_end,
2352 sizeof(struct boot_param_header), 4);
2353 dt_header_start = (unsigned long)hdr;
2354 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2356 /* Start of strings */
2357 mem_start = PAGE_ALIGN(mem_start);
2358 dt_string_start = mem_start;
2359 mem_start += 4; /* hole */
2361 /* Add "linux,phandle" in there, we'll need it */
2362 namep = make_room(&mem_start, &mem_end, 16, 1);
2363 strcpy(namep, "linux,phandle");
2364 mem_start = (unsigned long)namep + strlen(namep) + 1;
2366 /* Build string array */
2367 prom_printf("Building dt strings...\n");
2368 scan_dt_build_strings(root, &mem_start, &mem_end);
2369 dt_string_end = mem_start;
2371 /* Build structure */
2372 mem_start = PAGE_ALIGN(mem_start);
2373 dt_struct_start = mem_start;
2374 prom_printf("Building dt structure...\n");
2375 scan_dt_build_struct(root, &mem_start, &mem_end);
2376 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2377 dt_struct_end = PAGE_ALIGN(mem_start);
2380 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2381 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2382 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2383 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2384 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2385 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2386 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2387 hdr->version = cpu_to_be32(OF_DT_VERSION);
2388 /* Version 16 is not backward compatible */
2389 hdr->last_comp_version = cpu_to_be32(0x10);
2391 /* Copy the reserve map in */
2392 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2397 prom_printf("reserved memory map:\n");
2398 for (i = 0; i < mem_reserve_cnt; i++)
2399 prom_printf(" %x - %x\n",
2400 be64_to_cpu(mem_reserve_map[i].base),
2401 be64_to_cpu(mem_reserve_map[i].size));
2404 /* Bump mem_reserve_cnt to cause further reservations to fail
2405 * since it's too late.
2407 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2409 prom_printf("Device tree strings 0x%x -> 0x%x\n",
2410 dt_string_start, dt_string_end);
2411 prom_printf("Device tree struct 0x%x -> 0x%x\n",
2412 dt_struct_start, dt_struct_end);
2415 #ifdef CONFIG_PPC_MAPLE
2416 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2417 * The values are bad, and it doesn't even have the right number of cells. */
2418 static void __init fixup_device_tree_maple(void)
2421 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2425 name = "/ht@0/isa@4";
2426 isa = call_prom("finddevice", 1, 1, ADDR(name));
2427 if (!PHANDLE_VALID(isa)) {
2428 name = "/ht@0/isa@6";
2429 isa = call_prom("finddevice", 1, 1, ADDR(name));
2430 rloc = 0x01003000; /* IO space; PCI device = 6 */
2432 if (!PHANDLE_VALID(isa))
2435 if (prom_getproplen(isa, "ranges") != 12)
2437 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2441 if (isa_ranges[0] != 0x1 ||
2442 isa_ranges[1] != 0xf4000000 ||
2443 isa_ranges[2] != 0x00010000)
2446 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2448 isa_ranges[0] = 0x1;
2449 isa_ranges[1] = 0x0;
2450 isa_ranges[2] = rloc;
2451 isa_ranges[3] = 0x0;
2452 isa_ranges[4] = 0x0;
2453 isa_ranges[5] = 0x00010000;
2454 prom_setprop(isa, name, "ranges",
2455 isa_ranges, sizeof(isa_ranges));
2458 #define CPC925_MC_START 0xf8000000
2459 #define CPC925_MC_LENGTH 0x1000000
2460 /* The values for memory-controller don't have right number of cells */
2461 static void __init fixup_device_tree_maple_memory_controller(void)
2465 char *name = "/hostbridge@f8000000";
2468 mc = call_prom("finddevice", 1, 1, ADDR(name));
2469 if (!PHANDLE_VALID(mc))
2472 if (prom_getproplen(mc, "reg") != 8)
2475 prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2476 prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2477 if ((ac != 2) || (sc != 2))
2480 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2483 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2486 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2489 mc_reg[1] = CPC925_MC_START;
2491 mc_reg[3] = CPC925_MC_LENGTH;
2492 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2495 #define fixup_device_tree_maple()
2496 #define fixup_device_tree_maple_memory_controller()
2499 #ifdef CONFIG_PPC_CHRP
2501 * Pegasos and BriQ lacks the "ranges" property in the isa node
2502 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2503 * Pegasos has the IDE configured in legacy mode, but advertised as native
2505 static void __init fixup_device_tree_chrp(void)
2509 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2513 name = "/pci@80000000/isa@c";
2514 ph = call_prom("finddevice", 1, 1, ADDR(name));
2515 if (!PHANDLE_VALID(ph)) {
2516 name = "/pci@ff500000/isa@6";
2517 ph = call_prom("finddevice", 1, 1, ADDR(name));
2518 rloc = 0x01003000; /* IO space; PCI device = 6 */
2520 if (PHANDLE_VALID(ph)) {
2521 rc = prom_getproplen(ph, "ranges");
2522 if (rc == 0 || rc == PROM_ERROR) {
2523 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2530 prop[5] = 0x00010000;
2531 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2535 name = "/pci@80000000/ide@C,1";
2536 ph = call_prom("finddevice", 1, 1, ADDR(name));
2537 if (PHANDLE_VALID(ph)) {
2538 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2541 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2542 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2543 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2544 if (rc == sizeof(u32)) {
2546 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2551 #define fixup_device_tree_chrp()
2554 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2555 static void __init fixup_device_tree_pmac(void)
2557 phandle u3, i2c, mpic;
2562 /* Some G5s have a missing interrupt definition, fix it up here */
2563 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2564 if (!PHANDLE_VALID(u3))
2566 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2567 if (!PHANDLE_VALID(i2c))
2569 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2570 if (!PHANDLE_VALID(mpic))
2573 /* check if proper rev of u3 */
2574 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2577 if (u3_rev < 0x35 || u3_rev > 0x39)
2579 /* does it need fixup ? */
2580 if (prom_getproplen(i2c, "interrupts") > 0)
2583 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2585 /* interrupt on this revision of u3 is number 0 and level */
2588 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2589 &interrupts, sizeof(interrupts));
2591 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2592 &parent, sizeof(parent));
2595 #define fixup_device_tree_pmac()
2598 #ifdef CONFIG_PPC_EFIKA
2600 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2601 * to talk to the phy. If the phy-handle property is missing, then this
2602 * function is called to add the appropriate nodes and link it to the
2605 static void __init fixup_device_tree_efika_add_phy(void)
2611 /* Check if /builtin/ethernet exists - bail if it doesn't */
2612 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2613 if (!PHANDLE_VALID(node))
2616 /* Check if the phy-handle property exists - bail if it does */
2617 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2622 * At this point the ethernet device doesn't have a phy described.
2623 * Now we need to add the missing phy node and linkage
2626 /* Check for an MDIO bus node - if missing then create one */
2627 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2628 if (!PHANDLE_VALID(node)) {
2629 prom_printf("Adding Ethernet MDIO node\n");
2630 call_prom("interpret", 1, 1,
2631 " s\" /builtin\" find-device"
2633 " 1 encode-int s\" #address-cells\" property"
2634 " 0 encode-int s\" #size-cells\" property"
2635 " s\" mdio\" device-name"
2636 " s\" fsl,mpc5200b-mdio\" encode-string"
2637 " s\" compatible\" property"
2638 " 0xf0003000 0x400 reg"
2640 " 0x5 encode-int encode+"
2641 " 0x3 encode-int encode+"
2642 " s\" interrupts\" property"
2646 /* Check for a PHY device node - if missing then create one and
2647 * give it's phandle to the ethernet node */
2648 node = call_prom("finddevice", 1, 1,
2649 ADDR("/builtin/mdio/ethernet-phy"));
2650 if (!PHANDLE_VALID(node)) {
2651 prom_printf("Adding Ethernet PHY node\n");
2652 call_prom("interpret", 1, 1,
2653 " s\" /builtin/mdio\" find-device"
2655 " s\" ethernet-phy\" device-name"
2656 " 0x10 encode-int s\" reg\" property"
2660 " s\" /builtin/ethernet\" find-device"
2662 " s\" phy-handle\" property"
2667 static void __init fixup_device_tree_efika(void)
2669 int sound_irq[3] = { 2, 2, 0 };
2670 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2671 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2672 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2673 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2678 /* Check if we're really running on a EFIKA */
2679 node = call_prom("finddevice", 1, 1, ADDR("/"));
2680 if (!PHANDLE_VALID(node))
2683 rv = prom_getprop(node, "model", prop, sizeof(prop));
2684 if (rv == PROM_ERROR)
2686 if (strcmp(prop, "EFIKA5K2"))
2689 prom_printf("Applying EFIKA device tree fixups\n");
2691 /* Claiming to be 'chrp' is death */
2692 node = call_prom("finddevice", 1, 1, ADDR("/"));
2693 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2694 if (rv != PROM_ERROR && (strcmp(prop, "chrp") == 0))
2695 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2697 /* CODEGEN,description is exposed in /proc/cpuinfo so
2699 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2700 if (rv != PROM_ERROR && (strstr(prop, "CHRP")))
2701 prom_setprop(node, "/", "CODEGEN,description",
2702 "Efika 5200B PowerPC System",
2703 sizeof("Efika 5200B PowerPC System"));
2705 /* Fixup bestcomm interrupts property */
2706 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2707 if (PHANDLE_VALID(node)) {
2708 len = prom_getproplen(node, "interrupts");
2710 prom_printf("Fixing bestcomm interrupts property\n");
2711 prom_setprop(node, "/builtin/bestcom", "interrupts",
2712 bcomm_irq, sizeof(bcomm_irq));
2716 /* Fixup sound interrupts property */
2717 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2718 if (PHANDLE_VALID(node)) {
2719 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2720 if (rv == PROM_ERROR) {
2721 prom_printf("Adding sound interrupts property\n");
2722 prom_setprop(node, "/builtin/sound", "interrupts",
2723 sound_irq, sizeof(sound_irq));
2727 /* Make sure ethernet phy-handle property exists */
2728 fixup_device_tree_efika_add_phy();
2731 #define fixup_device_tree_efika()
2734 #ifdef CONFIG_PPC_PASEMI_NEMO
2736 * CFE supplied on Nemo is broken in several ways, biggest
2737 * problem is that it reassigns ISA interrupts to unused mpic ints.
2738 * Add an interrupt-controller property for the io-bridge to use
2739 * and correct the ints so we can attach them to an irq_domain
2741 static void __init fixup_device_tree_pasemi(void)
2743 u32 interrupts[2], parent, rval, val = 0;
2744 char *name, *pci_name;
2747 /* Find the root pci node */
2748 name = "/pxp@0,e0000000";
2749 iob = call_prom("finddevice", 1, 1, ADDR(name));
2750 if (!PHANDLE_VALID(iob))
2753 /* check if interrupt-controller node set yet */
2754 if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
2757 prom_printf("adding interrupt-controller property for SB600...\n");
2759 prom_setprop(iob, name, "interrupt-controller", &val, 0);
2761 pci_name = "/pxp@0,e0000000/pci@11";
2762 node = call_prom("finddevice", 1, 1, ADDR(pci_name));
2765 for( ; prom_next_node(&node); ) {
2766 /* scan each node for one with an interrupt */
2767 if (!PHANDLE_VALID(node))
2770 rval = prom_getproplen(node, "interrupts");
2771 if (rval == 0 || rval == PROM_ERROR)
2774 prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
2775 if ((interrupts[0] < 212) || (interrupts[0] > 222))
2778 /* found a node, update both interrupts and interrupt-parent */
2779 if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
2780 interrupts[0] -= 203;
2781 if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
2782 interrupts[0] -= 213;
2783 if (interrupts[0] == 221)
2785 if (interrupts[0] == 222)
2788 prom_setprop(node, pci_name, "interrupts", interrupts,
2789 sizeof(interrupts));
2790 prom_setprop(node, pci_name, "interrupt-parent", &parent,
2795 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
2796 * so that generic isa-bridge code can add the SB600 and its on-board
2799 name = "/pxp@0,e0000000/io-bridge@0";
2800 iob = call_prom("finddevice", 1, 1, ADDR(name));
2801 if (!PHANDLE_VALID(iob))
2804 /* device_type is already set, just change it. */
2806 prom_printf("Changing device_type of SB600 node...\n");
2808 prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
2810 #else /* !CONFIG_PPC_PASEMI_NEMO */
2811 static inline void fixup_device_tree_pasemi(void) { }
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();
2821 fixup_device_tree_pasemi();
2824 static void __init prom_find_boot_cpu(void)
2831 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
2833 prom_cpu = be32_to_cpu(rval);
2835 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
2837 if (!PHANDLE_VALID(cpu_pkg))
2840 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
2841 prom.cpu = be32_to_cpu(rval);
2843 prom_debug("Booting CPU hw index = %lu\n", prom.cpu);
2846 static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
2848 #ifdef CONFIG_BLK_DEV_INITRD
2849 if (r3 && r4 && r4 != 0xdeadbeef) {
2852 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
2853 prom_initrd_end = prom_initrd_start + r4;
2855 val = cpu_to_be64(prom_initrd_start);
2856 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
2858 val = cpu_to_be64(prom_initrd_end);
2859 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
2862 reserve_mem(prom_initrd_start,
2863 prom_initrd_end - prom_initrd_start);
2865 prom_debug("initrd_start=0x%x\n", prom_initrd_start);
2866 prom_debug("initrd_end=0x%x\n", prom_initrd_end);
2868 #endif /* CONFIG_BLK_DEV_INITRD */
2872 #ifdef CONFIG_RELOCATABLE
2873 static void reloc_toc(void)
2877 static void unreloc_toc(void)
2881 static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
2884 unsigned long *toc_entry;
2886 /* Get the start of the TOC by using r2 directly. */
2887 asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
2889 for (i = 0; i < nr_entries; i++) {
2890 *toc_entry = *toc_entry + offset;
2895 static void reloc_toc(void)
2897 unsigned long offset = reloc_offset();
2898 unsigned long nr_entries =
2899 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2901 __reloc_toc(offset, nr_entries);
2906 static void unreloc_toc(void)
2908 unsigned long offset = reloc_offset();
2909 unsigned long nr_entries =
2910 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2914 __reloc_toc(-offset, nr_entries);
2920 * We enter here early on, when the Open Firmware prom is still
2921 * handling exceptions and the MMU hash table for us.
2924 unsigned long __init prom_init(unsigned long r3, unsigned long r4,
2926 unsigned long r6, unsigned long r7,
2927 unsigned long kbase)
2932 unsigned long offset = reloc_offset();
2939 * First zero the BSS
2941 memset(&__bss_start, 0, __bss_stop - __bss_start);
2944 * Init interface to Open Firmware, get some node references,
2947 prom_init_client_services(pp);
2950 * See if this OF is old enough that we need to do explicit maps
2951 * and other workarounds
2956 * Init prom stdout device
2960 prom_printf("Preparing to boot %s", linux_banner);
2963 * Get default machine type. At this point, we do not differentiate
2964 * between pSeries SMP and pSeries LPAR
2966 of_platform = prom_find_machine_type();
2967 prom_printf("Detected machine type: %x\n", of_platform);
2969 #ifndef CONFIG_NONSTATIC_KERNEL
2970 /* Bail if this is a kdump kernel. */
2971 if (PHYSICAL_START > 0)
2972 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2976 * Check for an initrd
2978 prom_check_initrd(r3, r4);
2980 #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
2982 * On pSeries, inform the firmware about our capabilities
2984 if (of_platform == PLATFORM_PSERIES ||
2985 of_platform == PLATFORM_PSERIES_LPAR)
2986 prom_send_capabilities();
2990 * Copy the CPU hold code
2992 if (of_platform != PLATFORM_POWERMAC)
2993 copy_and_flush(0, kbase, 0x100, 0);
2996 * Do early parsing of command line
2998 early_cmdline_parse();
3001 * Initialize memory management within prom_init
3006 * Determine which cpu is actually running right _now_
3008 prom_find_boot_cpu();
3011 * Initialize display devices
3013 prom_check_displays();
3015 #if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3017 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3018 * that uses the allocator, we need to make sure we get the top of memory
3019 * available for us here...
3021 if (of_platform == PLATFORM_PSERIES)
3022 prom_initialize_tce_table();
3026 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3027 * have a usable RTAS implementation.
3029 if (of_platform != PLATFORM_POWERMAC &&
3030 of_platform != PLATFORM_OPAL)
3031 prom_instantiate_rtas();
3033 #ifdef CONFIG_PPC_POWERNV
3034 if (of_platform == PLATFORM_OPAL)
3035 prom_instantiate_opal();
3036 #endif /* CONFIG_PPC_POWERNV */
3039 /* instantiate sml */
3040 prom_instantiate_sml();
3044 * On non-powermacs, put all CPUs in spin-loops.
3046 * PowerMacs use a different mechanism to spin CPUs
3048 * (This must be done after instanciating RTAS)
3050 if (of_platform != PLATFORM_POWERMAC &&
3051 of_platform != PLATFORM_OPAL)
3055 * Fill in some infos for use by the kernel later on
3057 if (prom_memory_limit) {
3058 __be64 val = cpu_to_be64(prom_memory_limit);
3059 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3064 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3067 if (prom_iommu_force_on)
3068 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3071 if (prom_tce_alloc_start) {
3072 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3073 &prom_tce_alloc_start,
3074 sizeof(prom_tce_alloc_start));
3075 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3076 &prom_tce_alloc_end,
3077 sizeof(prom_tce_alloc_end));
3082 * Fixup any known bugs in the device-tree
3084 fixup_device_tree();
3087 * Now finally create the flattened device-tree
3089 prom_printf("copying OF device tree...\n");
3090 flatten_device_tree();
3093 * in case stdin is USB and still active on IBM machines...
3094 * Unfortunately quiesce crashes on some powermacs if we have
3095 * closed stdin already (in particular the powerbook 101). It
3096 * appears that the OPAL version of OFW doesn't like it either.
3098 if (of_platform != PLATFORM_POWERMAC &&
3099 of_platform != PLATFORM_OPAL)
3103 * Call OF "quiesce" method to shut down pending DMA's from
3106 prom_printf("Quiescing Open Firmware ...\n");
3107 call_prom("quiesce", 0, 0);
3110 * And finally, call the kernel passing it the flattened device
3111 * tree and NULL as r5, thus triggering the new entry point which
3112 * is common to us and kexec
3114 hdr = dt_header_start;
3116 /* Don't print anything after quiesce under OPAL, it crashes OFW */
3117 if (of_platform != PLATFORM_OPAL) {
3118 prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3119 prom_debug("->dt_header_start=0x%x\n", hdr);
3123 reloc_got2(-offset);
3128 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
3129 /* OPAL early debug gets the OPAL base & entry in r8 and r9 */
3130 __start(hdr, kbase, 0, 0, 0,
3131 prom_opal_base, prom_opal_entry);
3133 __start(hdr, kbase, 0, 0, 0, 0, 0);