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.
18 /* we cannot use FORTIFY as it brings in new symbols */
22 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/threads.h>
26 #include <linux/spinlock.h>
27 #include <linux/types.h>
28 #include <linux/pci.h>
29 #include <linux/proc_fs.h>
30 #include <linux/delay.h>
31 #include <linux/initrd.h>
32 #include <linux/bitops.h>
36 #include <asm/processor.h>
41 #include <asm/pgtable.h>
42 #include <asm/iommu.h>
43 #include <asm/btext.h>
44 #include <asm/sections.h>
45 #include <asm/machdep.h>
47 #include <asm/asm-prototypes.h>
49 #include <linux/linux_logo.h>
52 * Eventually bump that one up
54 #define DEVTREE_CHUNK_SIZE 0x100000
57 * This is the size of the local memory reserve map that gets copied
58 * into the boot params passed to the kernel. That size is totally
59 * flexible as the kernel just reads the list until it encounters an
60 * entry with size 0, so it can be changed without breaking binary
63 #define MEM_RESERVE_MAP_SIZE 8
66 * prom_init() is called very early on, before the kernel text
67 * and data have been mapped to KERNELBASE. At this point the code
68 * is running at whatever address it has been loaded at.
69 * On ppc32 we compile with -mrelocatable, which means that references
70 * to extern and static variables get relocated automatically.
71 * ppc64 objects are always relocatable, we just need to relocate the
74 * Because OF may have mapped I/O devices into the area starting at
75 * KERNELBASE, particularly on CHRP machines, we can't safely call
76 * OF once the kernel has been mapped to KERNELBASE. Therefore all
77 * OF calls must be done within prom_init().
79 * ADDR is used in calls to call_prom. The 4th and following
80 * arguments to call_prom should be 32-bit values.
81 * On ppc64, 64 bit values are truncated to 32 bits (and
82 * fortunately don't get interpreted as two arguments).
84 #define ADDR(x) (u32)(unsigned long)(x)
87 #define OF_WORKAROUNDS 0
89 #define OF_WORKAROUNDS of_workarounds
90 static int of_workarounds;
93 #define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
94 #define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
96 #define PROM_BUG() do { \
97 prom_printf("kernel BUG at %s line 0x%x!\n", \
98 __FILE__, __LINE__); \
99 __asm__ __volatile__(".long " BUG_ILLEGAL_INSTR); \
103 #define prom_debug(x...) prom_printf(x)
105 #define prom_debug(x...) do { } while (0)
109 typedef u32 prom_arg_t;
127 struct mem_map_entry {
132 typedef __be32 cell_t;
134 extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
135 unsigned long r6, unsigned long r7, unsigned long r8,
139 extern int enter_prom(struct prom_args *args, unsigned long entry);
141 static inline int enter_prom(struct prom_args *args, unsigned long entry)
143 return ((int (*)(struct prom_args *))entry)(args);
147 extern void copy_and_flush(unsigned long dest, unsigned long src,
148 unsigned long size, unsigned long offset);
151 static struct prom_t __initdata prom;
153 static unsigned long prom_entry __initdata;
155 #define PROM_SCRATCH_SIZE 256
157 static char __initdata of_stdout_device[256];
158 static char __initdata prom_scratch[PROM_SCRATCH_SIZE];
160 static unsigned long __initdata dt_header_start;
161 static unsigned long __initdata dt_struct_start, dt_struct_end;
162 static unsigned long __initdata dt_string_start, dt_string_end;
164 static unsigned long __initdata prom_initrd_start, prom_initrd_end;
167 static int __initdata prom_iommu_force_on;
168 static int __initdata prom_iommu_off;
169 static unsigned long __initdata prom_tce_alloc_start;
170 static unsigned long __initdata prom_tce_alloc_end;
173 static bool prom_radix_disable __initdata = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
175 struct platform_support {
182 /* Platforms codes are now obsolete in the kernel. Now only used within this
183 * file and ultimately gone too. Feel free to change them if you need, they
184 * are not shared with anything outside of this file anymore
186 #define PLATFORM_PSERIES 0x0100
187 #define PLATFORM_PSERIES_LPAR 0x0101
188 #define PLATFORM_LPAR 0x0001
189 #define PLATFORM_POWERMAC 0x0400
190 #define PLATFORM_GENERIC 0x0500
191 #define PLATFORM_OPAL 0x0600
193 static int __initdata of_platform;
195 static char __initdata prom_cmd_line[COMMAND_LINE_SIZE];
197 static unsigned long __initdata prom_memory_limit;
199 static unsigned long __initdata alloc_top;
200 static unsigned long __initdata alloc_top_high;
201 static unsigned long __initdata alloc_bottom;
202 static unsigned long __initdata rmo_top;
203 static unsigned long __initdata ram_top;
205 static struct mem_map_entry __initdata mem_reserve_map[MEM_RESERVE_MAP_SIZE];
206 static int __initdata mem_reserve_cnt;
208 static cell_t __initdata regbuf[1024];
210 static bool rtas_has_query_cpu_stopped;
214 * Error results ... some OF calls will return "-1" on error, some
215 * will return 0, some will return either. To simplify, here are
216 * macros to use with any ihandle or phandle return value to check if
220 #define PROM_ERROR (-1u)
221 #define PHANDLE_VALID(p) ((p) != 0 && (p) != PROM_ERROR)
222 #define IHANDLE_VALID(i) ((i) != 0 && (i) != PROM_ERROR)
225 /* This is the one and *ONLY* place where we actually call open
229 static int __init call_prom(const char *service, int nargs, int nret, ...)
232 struct prom_args args;
235 args.service = cpu_to_be32(ADDR(service));
236 args.nargs = cpu_to_be32(nargs);
237 args.nret = cpu_to_be32(nret);
239 va_start(list, nret);
240 for (i = 0; i < nargs; i++)
241 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
244 for (i = 0; i < nret; i++)
245 args.args[nargs+i] = 0;
247 if (enter_prom(&args, prom_entry) < 0)
250 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
253 static int __init call_prom_ret(const char *service, int nargs, int nret,
254 prom_arg_t *rets, ...)
257 struct prom_args args;
260 args.service = cpu_to_be32(ADDR(service));
261 args.nargs = cpu_to_be32(nargs);
262 args.nret = cpu_to_be32(nret);
264 va_start(list, rets);
265 for (i = 0; i < nargs; i++)
266 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
269 for (i = 0; i < nret; i++)
270 args.args[nargs+i] = 0;
272 if (enter_prom(&args, prom_entry) < 0)
276 for (i = 1; i < nret; ++i)
277 rets[i-1] = be32_to_cpu(args.args[nargs+i]);
279 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
283 static void __init prom_print(const char *msg)
287 if (prom.stdout == 0)
290 for (p = msg; *p != 0; p = q) {
291 for (q = p; *q != 0 && *q != '\n'; ++q)
294 call_prom("write", 3, 1, prom.stdout, p, q - p);
298 call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
304 * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
305 * we do not need __udivdi3 or __umoddi3 on 32bits.
307 static void __init prom_print_hex(unsigned long val)
309 int i, nibbles = sizeof(val)*2;
310 char buf[sizeof(val)*2+1];
312 for (i = nibbles-1; i >= 0; i--) {
313 buf[i] = (val & 0xf) + '0';
315 buf[i] += ('a'-'0'-10);
319 call_prom("write", 3, 1, prom.stdout, buf, nibbles);
322 /* max number of decimal digits in an unsigned long */
324 static void __init prom_print_dec(unsigned long val)
327 char buf[UL_DIGITS+1];
329 for (i = UL_DIGITS-1; i >= 0; i--) {
330 buf[i] = (val % 10) + '0';
335 /* shift stuff down */
336 size = UL_DIGITS - i;
337 call_prom("write", 3, 1, prom.stdout, buf+i, size);
341 static void __init prom_printf(const char *format, ...)
343 const char *p, *q, *s;
349 va_start(args, format);
350 for (p = format; *p != 0; p = q) {
351 for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
354 call_prom("write", 3, 1, prom.stdout, p, q - p);
359 call_prom("write", 3, 1, prom.stdout,
373 s = va_arg(args, const char *);
380 v = va_arg(args, unsigned int);
383 v = va_arg(args, unsigned long);
387 v = va_arg(args, unsigned long long);
396 v = va_arg(args, unsigned int);
399 v = va_arg(args, unsigned long);
403 v = va_arg(args, unsigned long long);
412 vs = va_arg(args, int);
415 vs = va_arg(args, long);
419 vs = va_arg(args, long long);
434 static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
438 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
440 * Old OF requires we claim physical and virtual separately
441 * and then map explicitly (assuming virtual mode)
446 ret = call_prom_ret("call-method", 5, 2, &result,
447 ADDR("claim"), prom.memory,
449 if (ret != 0 || result == -1)
451 ret = call_prom_ret("call-method", 5, 2, &result,
452 ADDR("claim"), prom.mmumap,
455 call_prom("call-method", 4, 1, ADDR("release"),
456 prom.memory, size, virt);
459 /* the 0x12 is M (coherence) + PP == read/write */
460 call_prom("call-method", 6, 1,
461 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
464 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
468 static void __init __attribute__((noreturn)) prom_panic(const char *reason)
471 /* Do not call exit because it clears the screen on pmac
472 * it also causes some sort of double-fault on early pmacs */
473 if (of_platform == PLATFORM_POWERMAC)
476 /* ToDo: should put up an SRC here on pSeries */
477 call_prom("exit", 0, 0);
479 for (;;) /* should never get here */
484 static int __init prom_next_node(phandle *nodep)
488 if ((node = *nodep) != 0
489 && (*nodep = call_prom("child", 1, 1, node)) != 0)
491 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
494 if ((node = call_prom("parent", 1, 1, node)) == 0)
496 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
501 static inline int prom_getprop(phandle node, const char *pname,
502 void *value, size_t valuelen)
504 return call_prom("getprop", 4, 1, node, ADDR(pname),
505 (u32)(unsigned long) value, (u32) valuelen);
508 static inline int prom_getproplen(phandle node, const char *pname)
510 return call_prom("getproplen", 2, 1, node, ADDR(pname));
513 static void add_string(char **str, const char *q)
523 static char *tohex(unsigned int x)
525 static char digits[] = "0123456789abcdef";
526 static char result[9];
533 result[i] = digits[x & 0xf];
535 } while (x != 0 && i > 0);
539 static int __init prom_setprop(phandle node, const char *nodename,
540 const char *pname, void *value, size_t valuelen)
544 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
545 return call_prom("setprop", 4, 1, node, ADDR(pname),
546 (u32)(unsigned long) value, (u32) valuelen);
548 /* gah... setprop doesn't work on longtrail, have to use interpret */
550 add_string(&p, "dev");
551 add_string(&p, nodename);
552 add_string(&p, tohex((u32)(unsigned long) value));
553 add_string(&p, tohex(valuelen));
554 add_string(&p, tohex(ADDR(pname)));
555 add_string(&p, tohex(strlen(pname)));
556 add_string(&p, "property");
558 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
561 /* We can't use the standard versions because of relocation headaches. */
562 #define isxdigit(c) (('0' <= (c) && (c) <= '9') \
563 || ('a' <= (c) && (c) <= 'f') \
564 || ('A' <= (c) && (c) <= 'F'))
566 #define isdigit(c) ('0' <= (c) && (c) <= '9')
567 #define islower(c) ('a' <= (c) && (c) <= 'z')
568 #define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
570 static unsigned long prom_strtoul(const char *cp, const char **endp)
572 unsigned long result = 0, base = 10, value;
577 if (toupper(*cp) == 'X') {
583 while (isxdigit(*cp) &&
584 (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
585 result = result * base + value;
595 static unsigned long prom_memparse(const char *ptr, const char **retptr)
597 unsigned long ret = prom_strtoul(ptr, retptr);
601 * We can't use a switch here because GCC *may* generate a
602 * jump table which won't work, because we're not running at
603 * the address we're linked at.
605 if ('G' == **retptr || 'g' == **retptr)
608 if ('M' == **retptr || 'm' == **retptr)
611 if ('K' == **retptr || 'k' == **retptr)
623 * Early parsing of the command line passed to the kernel, used for
624 * "mem=x" and the options that affect the iommu
626 static void __init early_cmdline_parse(void)
631 int l __maybe_unused = 0;
633 prom_cmd_line[0] = 0;
635 if ((long)prom.chosen > 0)
636 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
637 #ifdef CONFIG_CMDLINE
638 if (l <= 0 || p[0] == '\0') /* dbl check */
639 strlcpy(prom_cmd_line,
640 CONFIG_CMDLINE, sizeof(prom_cmd_line));
641 #endif /* CONFIG_CMDLINE */
642 prom_printf("command line: %s\n", prom_cmd_line);
645 opt = strstr(prom_cmd_line, "iommu=");
647 prom_printf("iommu opt is: %s\n", opt);
649 while (*opt && *opt == ' ')
651 if (!strncmp(opt, "off", 3))
653 else if (!strncmp(opt, "force", 5))
654 prom_iommu_force_on = 1;
657 opt = strstr(prom_cmd_line, "mem=");
660 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
662 /* Align to 16 MB == size of ppc64 large page */
663 prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
667 opt = strstr(prom_cmd_line, "disable_radix");
670 if (*opt && *opt == '=') {
673 if (kstrtobool(++opt, &val))
674 prom_radix_disable = false;
676 prom_radix_disable = val;
678 prom_radix_disable = true;
680 if (prom_radix_disable)
681 prom_debug("Radix disabled from cmdline\n");
684 #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
686 * The architecture vector has an array of PVR mask/value pairs,
687 * followed by # option vectors - 1, followed by the option vectors.
689 * See prom.h for the definition of the bits specified in the
690 * architecture vector.
693 /* Firmware expects the value to be n - 1, where n is the # of vectors */
694 #define NUM_VECTORS(n) ((n) - 1)
697 * Firmware expects 1 + n - 2, where n is the length of the option vector in
698 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
700 #define VECTOR_LENGTH(n) (1 + (n) - 2)
702 struct option_vector1 {
708 struct option_vector2 {
722 struct option_vector3 {
727 struct option_vector4 {
732 struct option_vector5 {
744 u8 platform_facilities;
755 struct option_vector6 {
761 struct ibm_arch_vec {
762 struct { u32 mask, val; } pvrs[12];
767 struct option_vector1 vec1;
770 struct option_vector2 vec2;
773 struct option_vector3 vec3;
776 struct option_vector4 vec4;
779 struct option_vector5 vec5;
782 struct option_vector6 vec6;
785 struct ibm_arch_vec __cacheline_aligned ibm_architecture_vec = {
788 .mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
789 .val = cpu_to_be32(0x003a0000),
792 .mask = cpu_to_be32(0xffff0000), /* POWER6 */
793 .val = cpu_to_be32(0x003e0000),
796 .mask = cpu_to_be32(0xffff0000), /* POWER7 */
797 .val = cpu_to_be32(0x003f0000),
800 .mask = cpu_to_be32(0xffff0000), /* POWER8E */
801 .val = cpu_to_be32(0x004b0000),
804 .mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
805 .val = cpu_to_be32(0x004c0000),
808 .mask = cpu_to_be32(0xffff0000), /* POWER8 */
809 .val = cpu_to_be32(0x004d0000),
812 .mask = cpu_to_be32(0xffff0000), /* POWER9 */
813 .val = cpu_to_be32(0x004e0000),
816 .mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
817 .val = cpu_to_be32(0x0f000005),
820 .mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
821 .val = cpu_to_be32(0x0f000004),
824 .mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
825 .val = cpu_to_be32(0x0f000003),
828 .mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
829 .val = cpu_to_be32(0x0f000002),
832 .mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
833 .val = cpu_to_be32(0x0f000001),
837 .num_vectors = NUM_VECTORS(6),
839 .vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
842 .arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
843 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
844 .arch_versions3 = OV1_PPC_3_00,
847 .vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
848 /* option vector 2: Open Firmware options supported */
850 .byte1 = OV2_REAL_MODE,
852 .real_base = cpu_to_be32(0xffffffff),
853 .real_size = cpu_to_be32(0xffffffff),
854 .virt_base = cpu_to_be32(0xffffffff),
855 .virt_size = cpu_to_be32(0xffffffff),
856 .load_base = cpu_to_be32(0xffffffff),
857 .min_rma = cpu_to_be32(512), /* 512MB min RMA */
858 .min_load = cpu_to_be32(0xffffffff), /* full client load */
859 .min_rma_percent = 0, /* min RMA percentage of total RAM */
860 .max_pft_size = 48, /* max log_2(hash table size) */
863 .vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
864 /* option vector 3: processor options supported */
866 .byte1 = 0, /* don't ignore, don't halt */
867 .byte2 = OV3_FP | OV3_VMX | OV3_DFP,
870 .vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
871 /* option vector 4: IBM PAPR implementation */
873 .byte1 = 0, /* don't halt */
874 .min_vp_cap = OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
877 .vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
878 /* option vector 5: PAPR/OF options */
880 .byte1 = 0, /* don't ignore, don't halt */
881 .byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
882 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
883 #ifdef CONFIG_PCI_MSI
884 /* PCIe/MSI support. Without MSI full PCIe is not supported */
891 #ifdef CONFIG_PPC_SMLPAR
892 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
896 .associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
897 .bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
898 .micro_checkpoint = 0,
900 .max_cpus = cpu_to_be32(NR_CPUS), /* number of cores supported */
903 .platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
907 .byte22 = OV5_FEAT(OV5_DRMEM_V2),
914 /* option vector 6: IBM PAPR hints */
915 .vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
919 .os_name = OV6_LINUX,
923 /* Old method - ELF header with PT_NOTE sections only works on BE */
924 #ifdef __BIG_ENDIAN__
925 static const struct fake_elf {
932 char name[8]; /* "PowerPC" */
946 char name[24]; /* "IBM,RPA-Client-Config" */
960 .e_ident = { 0x7f, 'E', 'L', 'F',
961 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
962 .e_type = ET_EXEC, /* yeah right */
964 .e_version = EV_CURRENT,
965 .e_phoff = offsetof(struct fake_elf, phdr),
966 .e_phentsize = sizeof(Elf32_Phdr),
972 .p_offset = offsetof(struct fake_elf, chrpnote),
973 .p_filesz = sizeof(struct chrpnote)
976 .p_offset = offsetof(struct fake_elf, rpanote),
977 .p_filesz = sizeof(struct rpanote)
981 .namesz = sizeof("PowerPC"),
982 .descsz = sizeof(struct chrpdesc),
986 .real_mode = ~0U, /* ~0 means "don't care" */
995 .namesz = sizeof("IBM,RPA-Client-Config"),
996 .descsz = sizeof(struct rpadesc),
998 .name = "IBM,RPA-Client-Config",
1001 .min_rmo_size = 64, /* in megabytes */
1002 .min_rmo_percent = 0,
1003 .max_pft_size = 48, /* 2^48 bytes max PFT size */
1010 #endif /* __BIG_ENDIAN__ */
1012 static int __init prom_count_smt_threads(void)
1018 /* Pick up th first CPU node we can find */
1019 for (node = 0; prom_next_node(&node); ) {
1021 prom_getprop(node, "device_type", type, sizeof(type));
1023 if (strcmp(type, "cpu"))
1026 * There is an entry for each smt thread, each entry being
1027 * 4 bytes long. All cpus should have the same number of
1028 * smt threads, so return after finding the first.
1030 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1031 if (plen == PROM_ERROR)
1034 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1037 if (plen < 1 || plen > 64) {
1038 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1039 (unsigned long)plen);
1044 prom_debug("No threads found, assuming 1 per core\n");
1050 static void __init prom_parse_mmu_model(u8 val,
1051 struct platform_support *support)
1054 case OV5_FEAT(OV5_MMU_DYNAMIC):
1055 case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1056 prom_debug("MMU - either supported\n");
1057 support->radix_mmu = !prom_radix_disable;
1058 support->hash_mmu = true;
1060 case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1061 prom_debug("MMU - radix only\n");
1062 if (prom_radix_disable) {
1064 * If we __have__ to do radix, we're better off ignoring
1065 * the command line rather than not booting.
1067 prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1069 support->radix_mmu = true;
1071 case OV5_FEAT(OV5_MMU_HASH):
1072 prom_debug("MMU - hash only\n");
1073 support->hash_mmu = true;
1076 prom_debug("Unknown mmu support option: 0x%x\n", val);
1081 static void __init prom_parse_xive_model(u8 val,
1082 struct platform_support *support)
1085 case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1086 prom_debug("XIVE - either mode supported\n");
1087 support->xive = true;
1089 case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1090 prom_debug("XIVE - exploitation mode supported\n");
1091 support->xive = true;
1093 case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1094 prom_debug("XIVE - legacy mode supported\n");
1097 prom_debug("Unknown xive support option: 0x%x\n", val);
1102 static void __init prom_parse_platform_support(u8 index, u8 val,
1103 struct platform_support *support)
1106 case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1107 prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1109 case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1110 if (val & OV5_FEAT(OV5_RADIX_GTSE)) {
1111 prom_debug("Radix - GTSE supported\n");
1112 support->radix_gtse = true;
1115 case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1116 prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1122 static void __init prom_check_platform_support(void)
1124 struct platform_support supported = {
1127 .radix_gtse = false,
1130 int prop_len = prom_getproplen(prom.chosen,
1131 "ibm,arch-vec-5-platform-support");
1135 prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1137 if (prop_len > sizeof(vec))
1138 prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1140 prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support",
1142 for (i = 0; i < sizeof(vec); i += 2) {
1143 prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2
1146 prom_parse_platform_support(vec[i], vec[i + 1],
1151 if (supported.radix_mmu && supported.radix_gtse &&
1152 IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1153 /* Radix preferred - but we require GTSE for now */
1154 prom_debug("Asking for radix with GTSE\n");
1155 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1156 ibm_architecture_vec.vec5.radix_ext = OV5_FEAT(OV5_RADIX_GTSE);
1157 } else if (supported.hash_mmu) {
1158 /* Default to hash mmu (if we can) */
1159 prom_debug("Asking for hash\n");
1160 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1162 /* We're probably on a legacy hypervisor */
1163 prom_debug("Assuming legacy hash support\n");
1166 if (supported.xive) {
1167 prom_debug("Asking for XIVE\n");
1168 ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1172 static void __init prom_send_capabilities(void)
1178 /* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1179 prom_check_platform_support();
1181 root = call_prom("open", 1, 1, ADDR("/"));
1183 /* We need to tell the FW about the number of cores we support.
1185 * To do that, we count the number of threads on the first core
1186 * (we assume this is the same for all cores) and use it to
1190 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1191 prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1194 ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1196 /* try calling the ibm,client-architecture-support method */
1197 prom_printf("Calling ibm,client-architecture-support...");
1198 if (call_prom_ret("call-method", 3, 2, &ret,
1199 ADDR("ibm,client-architecture-support"),
1201 ADDR(&ibm_architecture_vec)) == 0) {
1202 /* the call exists... */
1204 prom_printf("\nWARNING: ibm,client-architecture"
1205 "-support call FAILED!\n");
1206 call_prom("close", 1, 0, root);
1207 prom_printf(" done\n");
1210 call_prom("close", 1, 0, root);
1211 prom_printf(" not implemented\n");
1214 #ifdef __BIG_ENDIAN__
1218 /* no ibm,client-architecture-support call, try the old way */
1219 elfloader = call_prom("open", 1, 1,
1220 ADDR("/packages/elf-loader"));
1221 if (elfloader == 0) {
1222 prom_printf("couldn't open /packages/elf-loader\n");
1225 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1226 elfloader, ADDR(&fake_elf));
1227 call_prom("close", 1, 0, elfloader);
1229 #endif /* __BIG_ENDIAN__ */
1231 #endif /* #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV) */
1234 * Memory allocation strategy... our layout is normally:
1236 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
1237 * rare cases, initrd might end up being before the kernel though.
1238 * We assume this won't override the final kernel at 0, we have no
1239 * provision to handle that in this version, but it should hopefully
1242 * alloc_top is set to the top of RMO, eventually shrink down if the
1245 * alloc_bottom is set to the top of kernel/initrd
1247 * from there, allocations are done this way : rtas is allocated
1248 * topmost, and the device-tree is allocated from the bottom. We try
1249 * to grow the device-tree allocation as we progress. If we can't,
1250 * then we fail, we don't currently have a facility to restart
1251 * elsewhere, but that shouldn't be necessary.
1253 * Note that calls to reserve_mem have to be done explicitly, memory
1254 * allocated with either alloc_up or alloc_down isn't automatically
1260 * Allocates memory in the RMO upward from the kernel/initrd
1262 * When align is 0, this is a special case, it means to allocate in place
1263 * at the current location of alloc_bottom or fail (that is basically
1264 * extending the previous allocation). Used for the device-tree flattening
1266 static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1268 unsigned long base = alloc_bottom;
1269 unsigned long addr = 0;
1272 base = _ALIGN_UP(base, align);
1273 prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1275 prom_panic("alloc_up() called with mem not initialized\n");
1278 base = _ALIGN_UP(alloc_bottom, align);
1280 base = alloc_bottom;
1282 for(; (base + size) <= alloc_top;
1283 base = _ALIGN_UP(base + 0x100000, align)) {
1284 prom_debug(" trying: 0x%lx\n\r", base);
1285 addr = (unsigned long)prom_claim(base, size, 0);
1286 if (addr != PROM_ERROR && addr != 0)
1294 alloc_bottom = addr + size;
1296 prom_debug(" -> %lx\n", addr);
1297 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1298 prom_debug(" alloc_top : %lx\n", alloc_top);
1299 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1300 prom_debug(" rmo_top : %lx\n", rmo_top);
1301 prom_debug(" ram_top : %lx\n", ram_top);
1307 * Allocates memory downward, either from top of RMO, or if highmem
1308 * is set, from the top of RAM. Note that this one doesn't handle
1309 * failures. It does claim memory if highmem is not set.
1311 static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1314 unsigned long base, addr = 0;
1316 prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1317 highmem ? "(high)" : "(low)");
1319 prom_panic("alloc_down() called with mem not initialized\n");
1322 /* Carve out storage for the TCE table. */
1323 addr = _ALIGN_DOWN(alloc_top_high - size, align);
1324 if (addr <= alloc_bottom)
1326 /* Will we bump into the RMO ? If yes, check out that we
1327 * didn't overlap existing allocations there, if we did,
1328 * we are dead, we must be the first in town !
1330 if (addr < rmo_top) {
1331 /* Good, we are first */
1332 if (alloc_top == rmo_top)
1333 alloc_top = rmo_top = addr;
1337 alloc_top_high = addr;
1341 base = _ALIGN_DOWN(alloc_top - size, align);
1342 for (; base > alloc_bottom;
1343 base = _ALIGN_DOWN(base - 0x100000, align)) {
1344 prom_debug(" trying: 0x%lx\n\r", base);
1345 addr = (unsigned long)prom_claim(base, size, 0);
1346 if (addr != PROM_ERROR && addr != 0)
1355 prom_debug(" -> %lx\n", addr);
1356 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1357 prom_debug(" alloc_top : %lx\n", alloc_top);
1358 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1359 prom_debug(" rmo_top : %lx\n", rmo_top);
1360 prom_debug(" ram_top : %lx\n", ram_top);
1366 * Parse a "reg" cell
1368 static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1371 unsigned long r = 0;
1373 /* Ignore more than 2 cells */
1374 while (s > sizeof(unsigned long) / 4) {
1378 r = be32_to_cpu(*p++);
1382 r |= be32_to_cpu(*(p++));
1390 * Very dumb function for adding to the memory reserve list, but
1391 * we don't need anything smarter at this point
1393 * XXX Eventually check for collisions. They should NEVER happen.
1394 * If problems seem to show up, it would be a good start to track
1397 static void __init reserve_mem(u64 base, u64 size)
1399 u64 top = base + size;
1400 unsigned long cnt = mem_reserve_cnt;
1405 /* We need to always keep one empty entry so that we
1406 * have our terminator with "size" set to 0 since we are
1407 * dumb and just copy this entire array to the boot params
1409 base = _ALIGN_DOWN(base, PAGE_SIZE);
1410 top = _ALIGN_UP(top, PAGE_SIZE);
1413 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1414 prom_panic("Memory reserve map exhausted !\n");
1415 mem_reserve_map[cnt].base = cpu_to_be64(base);
1416 mem_reserve_map[cnt].size = cpu_to_be64(size);
1417 mem_reserve_cnt = cnt + 1;
1421 * Initialize memory allocation mechanism, parse "memory" nodes and
1422 * obtain that way the top of memory and RMO to setup out local allocator
1424 static void __init prom_init_mem(void)
1437 * We iterate the memory nodes to find
1438 * 1) top of RMO (first node)
1441 val = cpu_to_be32(2);
1442 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1443 rac = be32_to_cpu(val);
1444 val = cpu_to_be32(1);
1445 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1446 rsc = be32_to_cpu(val);
1447 prom_debug("root_addr_cells: %x\n", rac);
1448 prom_debug("root_size_cells: %x\n", rsc);
1450 prom_debug("scanning memory:\n");
1452 path = prom_scratch;
1455 for (node = 0; prom_next_node(&node); ) {
1457 prom_getprop(node, "device_type", type, sizeof(type));
1461 * CHRP Longtrail machines have no device_type
1462 * on the memory node, so check the name instead...
1464 prom_getprop(node, "name", type, sizeof(type));
1466 if (strcmp(type, "memory"))
1469 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1470 if (plen > sizeof(regbuf)) {
1471 prom_printf("memory node too large for buffer !\n");
1472 plen = sizeof(regbuf);
1475 endp = p + (plen / sizeof(cell_t));
1478 memset(path, 0, PROM_SCRATCH_SIZE);
1479 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1480 prom_debug(" node %s :\n", path);
1481 #endif /* DEBUG_PROM */
1483 while ((endp - p) >= (rac + rsc)) {
1484 unsigned long base, size;
1486 base = prom_next_cell(rac, &p);
1487 size = prom_next_cell(rsc, &p);
1491 prom_debug(" %lx %lx\n", base, size);
1492 if (base == 0 && (of_platform & PLATFORM_LPAR))
1494 if ((base + size) > ram_top)
1495 ram_top = base + size;
1499 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1502 * If prom_memory_limit is set we reduce the upper limits *except* for
1503 * alloc_top_high. This must be the real top of RAM so we can put
1507 alloc_top_high = ram_top;
1509 if (prom_memory_limit) {
1510 if (prom_memory_limit <= alloc_bottom) {
1511 prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1513 prom_memory_limit = 0;
1514 } else if (prom_memory_limit >= ram_top) {
1515 prom_printf("Ignoring mem=%lx >= ram_top.\n",
1517 prom_memory_limit = 0;
1519 ram_top = prom_memory_limit;
1520 rmo_top = min(rmo_top, prom_memory_limit);
1525 * Setup our top alloc point, that is top of RMO or top of
1526 * segment 0 when running non-LPAR.
1527 * Some RS64 machines have buggy firmware where claims up at
1528 * 1GB fail. Cap at 768MB as a workaround.
1529 * Since 768MB is plenty of room, and we need to cap to something
1530 * reasonable on 32-bit, cap at 768MB on all machines.
1534 rmo_top = min(0x30000000ul, rmo_top);
1535 alloc_top = rmo_top;
1536 alloc_top_high = ram_top;
1539 * Check if we have an initrd after the kernel but still inside
1540 * the RMO. If we do move our bottom point to after it.
1542 if (prom_initrd_start &&
1543 prom_initrd_start < rmo_top &&
1544 prom_initrd_end > alloc_bottom)
1545 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1547 prom_printf("memory layout at init:\n");
1548 prom_printf(" memory_limit : %lx (16 MB aligned)\n",
1550 prom_printf(" alloc_bottom : %lx\n", alloc_bottom);
1551 prom_printf(" alloc_top : %lx\n", alloc_top);
1552 prom_printf(" alloc_top_hi : %lx\n", alloc_top_high);
1553 prom_printf(" rmo_top : %lx\n", rmo_top);
1554 prom_printf(" ram_top : %lx\n", ram_top);
1557 static void __init prom_close_stdin(void)
1562 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1563 stdin = be32_to_cpu(val);
1564 call_prom("close", 1, 0, stdin);
1568 #ifdef CONFIG_PPC_POWERNV
1570 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1571 static u64 __initdata prom_opal_base;
1572 static u64 __initdata prom_opal_entry;
1576 * Allocate room for and instantiate OPAL
1578 static void __init prom_instantiate_opal(void)
1583 u64 size = 0, align = 0x10000;
1587 prom_debug("prom_instantiate_opal: start...\n");
1589 opal_node = call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1590 prom_debug("opal_node: %x\n", opal_node);
1591 if (!PHANDLE_VALID(opal_node))
1595 prom_getprop(opal_node, "opal-runtime-size", &val64, sizeof(val64));
1596 size = be64_to_cpu(val64);
1600 prom_getprop(opal_node, "opal-runtime-alignment", &val64,sizeof(val64));
1601 align = be64_to_cpu(val64);
1603 base = alloc_down(size, align, 0);
1605 prom_printf("OPAL allocation failed !\n");
1609 opal_inst = call_prom("open", 1, 1, ADDR("/ibm,opal"));
1610 if (!IHANDLE_VALID(opal_inst)) {
1611 prom_printf("opening opal package failed (%x)\n", opal_inst);
1615 prom_printf("instantiating opal at 0x%llx...", base);
1617 if (call_prom_ret("call-method", 4, 3, rets,
1618 ADDR("load-opal-runtime"),
1620 base >> 32, base & 0xffffffff) != 0
1621 || (rets[0] == 0 && rets[1] == 0)) {
1622 prom_printf(" failed\n");
1625 entry = (((u64)rets[0]) << 32) | rets[1];
1627 prom_printf(" done\n");
1629 reserve_mem(base, size);
1631 prom_debug("opal base = 0x%llx\n", base);
1632 prom_debug("opal align = 0x%llx\n", align);
1633 prom_debug("opal entry = 0x%llx\n", entry);
1634 prom_debug("opal size = 0x%llx\n", size);
1636 prom_setprop(opal_node, "/ibm,opal", "opal-base-address",
1637 &base, sizeof(base));
1638 prom_setprop(opal_node, "/ibm,opal", "opal-entry-address",
1639 &entry, sizeof(entry));
1641 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1642 prom_opal_base = base;
1643 prom_opal_entry = entry;
1645 prom_debug("prom_instantiate_opal: end...\n");
1648 #endif /* CONFIG_PPC_POWERNV */
1651 * Allocate room for and instantiate RTAS
1653 static void __init prom_instantiate_rtas(void)
1657 u32 base, entry = 0;
1661 prom_debug("prom_instantiate_rtas: start...\n");
1663 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1664 prom_debug("rtas_node: %x\n", rtas_node);
1665 if (!PHANDLE_VALID(rtas_node))
1669 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1670 size = be32_to_cpu(val);
1674 base = alloc_down(size, PAGE_SIZE, 0);
1676 prom_panic("Could not allocate memory for RTAS\n");
1678 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1679 if (!IHANDLE_VALID(rtas_inst)) {
1680 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1684 prom_printf("instantiating rtas at 0x%x...", base);
1686 if (call_prom_ret("call-method", 3, 2, &entry,
1687 ADDR("instantiate-rtas"),
1688 rtas_inst, base) != 0
1690 prom_printf(" failed\n");
1693 prom_printf(" done\n");
1695 reserve_mem(base, size);
1697 val = cpu_to_be32(base);
1698 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1700 val = cpu_to_be32(entry);
1701 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1704 /* Check if it supports "query-cpu-stopped-state" */
1705 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1706 &val, sizeof(val)) != PROM_ERROR)
1707 rtas_has_query_cpu_stopped = true;
1709 prom_debug("rtas base = 0x%x\n", base);
1710 prom_debug("rtas entry = 0x%x\n", entry);
1711 prom_debug("rtas size = 0x%x\n", size);
1713 prom_debug("prom_instantiate_rtas: end...\n");
1718 * Allocate room for and instantiate Stored Measurement Log (SML)
1720 static void __init prom_instantiate_sml(void)
1722 phandle ibmvtpm_node;
1723 ihandle ibmvtpm_inst;
1724 u32 entry = 0, size = 0, succ = 0;
1728 prom_debug("prom_instantiate_sml: start...\n");
1730 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1731 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1732 if (!PHANDLE_VALID(ibmvtpm_node))
1735 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1736 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1737 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1741 if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1742 &val, sizeof(val)) != PROM_ERROR) {
1743 if (call_prom_ret("call-method", 2, 2, &succ,
1744 ADDR("reformat-sml-to-efi-alignment"),
1745 ibmvtpm_inst) != 0 || succ == 0) {
1746 prom_printf("Reformat SML to EFI alignment failed\n");
1750 if (call_prom_ret("call-method", 2, 2, &size,
1751 ADDR("sml-get-allocated-size"),
1752 ibmvtpm_inst) != 0 || size == 0) {
1753 prom_printf("SML get allocated size failed\n");
1757 if (call_prom_ret("call-method", 2, 2, &size,
1758 ADDR("sml-get-handover-size"),
1759 ibmvtpm_inst) != 0 || size == 0) {
1760 prom_printf("SML get handover size failed\n");
1765 base = alloc_down(size, PAGE_SIZE, 0);
1767 prom_panic("Could not allocate memory for sml\n");
1769 prom_printf("instantiating sml at 0x%llx...", base);
1771 memset((void *)base, 0, size);
1773 if (call_prom_ret("call-method", 4, 2, &entry,
1774 ADDR("sml-handover"),
1775 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1776 prom_printf("SML handover failed\n");
1779 prom_printf(" done\n");
1781 reserve_mem(base, size);
1783 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1784 &base, sizeof(base));
1785 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1786 &size, sizeof(size));
1788 prom_debug("sml base = 0x%llx\n", base);
1789 prom_debug("sml size = 0x%x\n", size);
1791 prom_debug("prom_instantiate_sml: end...\n");
1795 * Allocate room for and initialize TCE tables
1797 #ifdef __BIG_ENDIAN__
1798 static void __init prom_initialize_tce_table(void)
1802 char compatible[64], type[64], model[64];
1803 char *path = prom_scratch;
1805 u32 minalign, minsize;
1806 u64 tce_entry, *tce_entryp;
1807 u64 local_alloc_top, local_alloc_bottom;
1813 prom_debug("starting prom_initialize_tce_table\n");
1815 /* Cache current top of allocs so we reserve a single block */
1816 local_alloc_top = alloc_top_high;
1817 local_alloc_bottom = local_alloc_top;
1819 /* Search all nodes looking for PHBs. */
1820 for (node = 0; prom_next_node(&node); ) {
1824 prom_getprop(node, "compatible",
1825 compatible, sizeof(compatible));
1826 prom_getprop(node, "device_type", type, sizeof(type));
1827 prom_getprop(node, "model", model, sizeof(model));
1829 if ((type[0] == 0) || (strstr(type, "pci") == NULL))
1832 /* Keep the old logic intact to avoid regression. */
1833 if (compatible[0] != 0) {
1834 if ((strstr(compatible, "python") == NULL) &&
1835 (strstr(compatible, "Speedwagon") == NULL) &&
1836 (strstr(compatible, "Winnipeg") == NULL))
1838 } else if (model[0] != 0) {
1839 if ((strstr(model, "ython") == NULL) &&
1840 (strstr(model, "peedwagon") == NULL) &&
1841 (strstr(model, "innipeg") == NULL))
1845 if (prom_getprop(node, "tce-table-minalign", &minalign,
1846 sizeof(minalign)) == PROM_ERROR)
1848 if (prom_getprop(node, "tce-table-minsize", &minsize,
1849 sizeof(minsize)) == PROM_ERROR)
1850 minsize = 4UL << 20;
1853 * Even though we read what OF wants, we just set the table
1854 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
1855 * By doing this, we avoid the pitfalls of trying to DMA to
1856 * MMIO space and the DMA alias hole.
1858 minsize = 4UL << 20;
1860 /* Align to the greater of the align or size */
1861 align = max(minalign, minsize);
1862 base = alloc_down(minsize, align, 1);
1864 prom_panic("ERROR, cannot find space for TCE table.\n");
1865 if (base < local_alloc_bottom)
1866 local_alloc_bottom = base;
1868 /* It seems OF doesn't null-terminate the path :-( */
1869 memset(path, 0, PROM_SCRATCH_SIZE);
1870 /* Call OF to setup the TCE hardware */
1871 if (call_prom("package-to-path", 3, 1, node,
1872 path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1873 prom_printf("package-to-path failed\n");
1876 /* Save away the TCE table attributes for later use. */
1877 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1878 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1880 prom_debug("TCE table: %s\n", path);
1881 prom_debug("\tnode = 0x%x\n", node);
1882 prom_debug("\tbase = 0x%llx\n", base);
1883 prom_debug("\tsize = 0x%x\n", minsize);
1885 /* Initialize the table to have a one-to-one mapping
1886 * over the allocated size.
1888 tce_entryp = (u64 *)base;
1889 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1890 tce_entry = (i << PAGE_SHIFT);
1892 *tce_entryp = tce_entry;
1895 prom_printf("opening PHB %s", path);
1896 phb_node = call_prom("open", 1, 1, path);
1898 prom_printf("... failed\n");
1900 prom_printf("... done\n");
1902 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1903 phb_node, -1, minsize,
1904 (u32) base, (u32) (base >> 32));
1905 call_prom("close", 1, 0, phb_node);
1908 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1910 /* These are only really needed if there is a memory limit in
1911 * effect, but we don't know so export them always. */
1912 prom_tce_alloc_start = local_alloc_bottom;
1913 prom_tce_alloc_end = local_alloc_top;
1915 /* Flag the first invalid entry */
1916 prom_debug("ending prom_initialize_tce_table\n");
1918 #endif /* __BIG_ENDIAN__ */
1919 #endif /* CONFIG_PPC64 */
1922 * With CHRP SMP we need to use the OF to start the other processors.
1923 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1924 * so we have to put the processors into a holding pattern controlled
1925 * by the kernel (not OF) before we destroy the OF.
1927 * This uses a chunk of low memory, puts some holding pattern
1928 * code there and sends the other processors off to there until
1929 * smp_boot_cpus tells them to do something. The holding pattern
1930 * checks that address until its cpu # is there, when it is that
1931 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
1932 * of setting those values.
1934 * We also use physical address 0x4 here to tell when a cpu
1935 * is in its holding pattern code.
1940 * We want to reference the copy of __secondary_hold_* in the
1941 * 0 - 0x100 address range
1943 #define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
1945 static void __init prom_hold_cpus(void)
1950 unsigned long *spinloop
1951 = (void *) LOW_ADDR(__secondary_hold_spinloop);
1952 unsigned long *acknowledge
1953 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
1954 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1957 * On pseries, if RTAS supports "query-cpu-stopped-state",
1958 * we skip this stage, the CPUs will be started by the
1959 * kernel using RTAS.
1961 if ((of_platform == PLATFORM_PSERIES ||
1962 of_platform == PLATFORM_PSERIES_LPAR) &&
1963 rtas_has_query_cpu_stopped) {
1964 prom_printf("prom_hold_cpus: skipped\n");
1968 prom_debug("prom_hold_cpus: start...\n");
1969 prom_debug(" 1) spinloop = 0x%lx\n", (unsigned long)spinloop);
1970 prom_debug(" 1) *spinloop = 0x%lx\n", *spinloop);
1971 prom_debug(" 1) acknowledge = 0x%lx\n",
1972 (unsigned long)acknowledge);
1973 prom_debug(" 1) *acknowledge = 0x%lx\n", *acknowledge);
1974 prom_debug(" 1) secondary_hold = 0x%lx\n", secondary_hold);
1976 /* Set the common spinloop variable, so all of the secondary cpus
1977 * will block when they are awakened from their OF spinloop.
1978 * This must occur for both SMP and non SMP kernels, since OF will
1979 * be trashed when we move the kernel.
1984 for (node = 0; prom_next_node(&node); ) {
1985 unsigned int cpu_no;
1989 prom_getprop(node, "device_type", type, sizeof(type));
1990 if (strcmp(type, "cpu") != 0)
1993 /* Skip non-configured cpus. */
1994 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1995 if (strcmp(type, "okay") != 0)
1998 reg = cpu_to_be32(-1); /* make sparse happy */
1999 prom_getprop(node, "reg", ®, sizeof(reg));
2000 cpu_no = be32_to_cpu(reg);
2002 prom_debug("cpu hw idx = %u\n", cpu_no);
2004 /* Init the acknowledge var which will be reset by
2005 * the secondary cpu when it awakens from its OF
2008 *acknowledge = (unsigned long)-1;
2010 if (cpu_no != prom.cpu) {
2011 /* Primary Thread of non-boot cpu or any thread */
2012 prom_printf("starting cpu hw idx %u... ", cpu_no);
2013 call_prom("start-cpu", 3, 0, node,
2014 secondary_hold, cpu_no);
2016 for (i = 0; (i < 100000000) &&
2017 (*acknowledge == ((unsigned long)-1)); i++ )
2020 if (*acknowledge == cpu_no)
2021 prom_printf("done\n");
2023 prom_printf("failed: %lx\n", *acknowledge);
2027 prom_printf("boot cpu hw idx %u\n", cpu_no);
2028 #endif /* CONFIG_SMP */
2031 prom_debug("prom_hold_cpus: end...\n");
2035 static void __init prom_init_client_services(unsigned long pp)
2037 /* Get a handle to the prom entry point before anything else */
2040 /* get a handle for the stdout device */
2041 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2042 if (!PHANDLE_VALID(prom.chosen))
2043 prom_panic("cannot find chosen"); /* msg won't be printed :( */
2045 /* get device tree root */
2046 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2047 if (!PHANDLE_VALID(prom.root))
2048 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2055 * For really old powermacs, we need to map things we claim.
2056 * For that, we need the ihandle of the mmu.
2057 * Also, on the longtrail, we need to work around other bugs.
2059 static void __init prom_find_mmu(void)
2064 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2065 if (!PHANDLE_VALID(oprom))
2067 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2069 version[sizeof(version) - 1] = 0;
2070 /* XXX might need to add other versions here */
2071 if (strcmp(version, "Open Firmware, 1.0.5") == 0)
2072 of_workarounds = OF_WA_CLAIM;
2073 else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
2074 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2075 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2078 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2079 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2080 sizeof(prom.mmumap));
2081 prom.mmumap = be32_to_cpu(prom.mmumap);
2082 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2083 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
2086 #define prom_find_mmu()
2089 static void __init prom_init_stdout(void)
2091 char *path = of_stdout_device;
2093 phandle stdout_node;
2096 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2097 prom_panic("cannot find stdout");
2099 prom.stdout = be32_to_cpu(val);
2101 /* Get the full OF pathname of the stdout device */
2102 memset(path, 0, 256);
2103 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2104 prom_printf("OF stdout device is: %s\n", of_stdout_device);
2105 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2106 path, strlen(path) + 1);
2108 /* instance-to-package fails on PA-Semi */
2109 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2110 if (stdout_node != PROM_ERROR) {
2111 val = cpu_to_be32(stdout_node);
2113 /* If it's a display, note it */
2114 memset(type, 0, sizeof(type));
2115 prom_getprop(stdout_node, "device_type", type, sizeof(type));
2116 if (strcmp(type, "display") == 0)
2117 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2121 static int __init prom_find_machine_type(void)
2130 /* Look for a PowerMac or a Cell */
2131 len = prom_getprop(prom.root, "compatible",
2132 compat, sizeof(compat)-1);
2136 char *p = &compat[i];
2140 if (strstr(p, "Power Macintosh") ||
2141 strstr(p, "MacRISC"))
2142 return PLATFORM_POWERMAC;
2144 /* We must make sure we don't detect the IBM Cell
2145 * blades as pSeries due to some firmware issues,
2148 if (strstr(p, "IBM,CBEA") ||
2149 strstr(p, "IBM,CPBW-1.0"))
2150 return PLATFORM_GENERIC;
2151 #endif /* CONFIG_PPC64 */
2156 /* Try to detect OPAL */
2157 if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
2158 return PLATFORM_OPAL;
2160 /* Try to figure out if it's an IBM pSeries or any other
2161 * PAPR compliant platform. We assume it is if :
2162 * - /device_type is "chrp" (please, do NOT use that for future
2166 len = prom_getprop(prom.root, "device_type",
2167 compat, sizeof(compat)-1);
2169 return PLATFORM_GENERIC;
2170 if (strcmp(compat, "chrp"))
2171 return PLATFORM_GENERIC;
2173 /* Default to pSeries. We need to know if we are running LPAR */
2174 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2175 if (!PHANDLE_VALID(rtas))
2176 return PLATFORM_GENERIC;
2177 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2178 if (x != PROM_ERROR) {
2179 prom_debug("Hypertas detected, assuming LPAR !\n");
2180 return PLATFORM_PSERIES_LPAR;
2182 return PLATFORM_PSERIES;
2184 return PLATFORM_GENERIC;
2188 static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2190 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2194 * If we have a display that we don't know how to drive,
2195 * we will want to try to execute OF's open method for it
2196 * later. However, OF will probably fall over if we do that
2197 * we've taken over the MMU.
2198 * So we check whether we will need to open the display,
2199 * and if so, open it now.
2201 static void __init prom_check_displays(void)
2203 char type[16], *path;
2208 static unsigned char default_colors[] = {
2226 const unsigned char *clut;
2228 prom_debug("Looking for displays\n");
2229 for (node = 0; prom_next_node(&node); ) {
2230 memset(type, 0, sizeof(type));
2231 prom_getprop(node, "device_type", type, sizeof(type));
2232 if (strcmp(type, "display") != 0)
2235 /* It seems OF doesn't null-terminate the path :-( */
2236 path = prom_scratch;
2237 memset(path, 0, PROM_SCRATCH_SIZE);
2240 * leave some room at the end of the path for appending extra
2243 if (call_prom("package-to-path", 3, 1, node, path,
2244 PROM_SCRATCH_SIZE-10) == PROM_ERROR)
2246 prom_printf("found display : %s, opening... ", path);
2248 ih = call_prom("open", 1, 1, path);
2250 prom_printf("failed\n");
2255 prom_printf("done\n");
2256 prom_setprop(node, path, "linux,opened", NULL, 0);
2258 /* Setup a usable color table when the appropriate
2259 * method is available. Should update this to set-colors */
2260 clut = default_colors;
2261 for (i = 0; i < 16; i++, clut += 3)
2262 if (prom_set_color(ih, i, clut[0], clut[1],
2266 #ifdef CONFIG_LOGO_LINUX_CLUT224
2267 clut = PTRRELOC(logo_linux_clut224.clut);
2268 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2269 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2272 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
2274 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2275 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2277 u32 width, height, pitch, addr;
2279 prom_printf("Setting btext !\n");
2280 prom_getprop(node, "width", &width, 4);
2281 prom_getprop(node, "height", &height, 4);
2282 prom_getprop(node, "linebytes", &pitch, 4);
2283 prom_getprop(node, "address", &addr, 4);
2284 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2285 width, height, pitch, addr);
2286 btext_setup_display(width, height, 8, pitch, addr);
2288 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2293 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2294 static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2295 unsigned long needed, unsigned long align)
2299 *mem_start = _ALIGN(*mem_start, align);
2300 while ((*mem_start + needed) > *mem_end) {
2301 unsigned long room, chunk;
2303 prom_debug("Chunk exhausted, claiming more at %lx...\n",
2305 room = alloc_top - alloc_bottom;
2306 if (room > DEVTREE_CHUNK_SIZE)
2307 room = DEVTREE_CHUNK_SIZE;
2308 if (room < PAGE_SIZE)
2309 prom_panic("No memory for flatten_device_tree "
2311 chunk = alloc_up(room, 0);
2313 prom_panic("No memory for flatten_device_tree "
2314 "(claim failed)\n");
2315 *mem_end = chunk + room;
2318 ret = (void *)*mem_start;
2319 *mem_start += needed;
2324 #define dt_push_token(token, mem_start, mem_end) do { \
2325 void *room = make_room(mem_start, mem_end, 4, 4); \
2326 *(__be32 *)room = cpu_to_be32(token); \
2329 static unsigned long __init dt_find_string(char *str)
2333 s = os = (char *)dt_string_start;
2335 while (s < (char *)dt_string_end) {
2336 if (strcmp(s, str) == 0)
2344 * The Open Firmware 1275 specification states properties must be 31 bytes or
2345 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2347 #define MAX_PROPERTY_NAME 64
2349 static void __init scan_dt_build_strings(phandle node,
2350 unsigned long *mem_start,
2351 unsigned long *mem_end)
2353 char *prev_name, *namep, *sstart;
2357 sstart = (char *)dt_string_start;
2359 /* get and store all property names */
2362 /* 64 is max len of name including nul. */
2363 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2364 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2365 /* No more nodes: unwind alloc */
2366 *mem_start = (unsigned long)namep;
2371 if (strcmp(namep, "name") == 0) {
2372 *mem_start = (unsigned long)namep;
2376 /* get/create string entry */
2377 soff = dt_find_string(namep);
2379 *mem_start = (unsigned long)namep;
2380 namep = sstart + soff;
2382 /* Trim off some if we can */
2383 *mem_start = (unsigned long)namep + strlen(namep) + 1;
2384 dt_string_end = *mem_start;
2389 /* do all our children */
2390 child = call_prom("child", 1, 1, node);
2391 while (child != 0) {
2392 scan_dt_build_strings(child, mem_start, mem_end);
2393 child = call_prom("peer", 1, 1, child);
2397 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2398 unsigned long *mem_end)
2401 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2403 unsigned char *valp;
2404 static char pname[MAX_PROPERTY_NAME];
2405 int l, room, has_phandle = 0;
2407 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2409 /* get the node's full name */
2410 namep = (char *)*mem_start;
2411 room = *mem_end - *mem_start;
2414 l = call_prom("package-to-path", 3, 1, node, namep, room);
2416 /* Didn't fit? Get more room. */
2418 if (l >= *mem_end - *mem_start)
2419 namep = make_room(mem_start, mem_end, l+1, 1);
2420 call_prom("package-to-path", 3, 1, node, namep, l);
2424 /* Fixup an Apple bug where they have bogus \0 chars in the
2425 * middle of the path in some properties, and extract
2426 * the unit name (everything after the last '/').
2428 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2435 *mem_start = _ALIGN((unsigned long)lp + 1, 4);
2438 /* get it again for debugging */
2439 path = prom_scratch;
2440 memset(path, 0, PROM_SCRATCH_SIZE);
2441 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
2443 /* get and store all properties */
2445 sstart = (char *)dt_string_start;
2447 if (call_prom("nextprop", 3, 1, node, prev_name,
2452 if (strcmp(pname, "name") == 0) {
2457 /* find string offset */
2458 soff = dt_find_string(pname);
2460 prom_printf("WARNING: Can't find string index for"
2461 " <%s>, node %s\n", pname, path);
2464 prev_name = sstart + soff;
2467 l = call_prom("getproplen", 2, 1, node, pname);
2470 if (l == PROM_ERROR)
2473 /* push property head */
2474 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2475 dt_push_token(l, mem_start, mem_end);
2476 dt_push_token(soff, mem_start, mem_end);
2478 /* push property content */
2479 valp = make_room(mem_start, mem_end, l, 4);
2480 call_prom("getprop", 4, 1, node, pname, valp, l);
2481 *mem_start = _ALIGN(*mem_start, 4);
2483 if (!strcmp(pname, "phandle"))
2487 /* Add a "linux,phandle" property if no "phandle" property already
2488 * existed (can happen with OPAL)
2491 soff = dt_find_string("linux,phandle");
2493 prom_printf("WARNING: Can't find string index for"
2494 " <linux-phandle> node %s\n", path);
2496 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2497 dt_push_token(4, mem_start, mem_end);
2498 dt_push_token(soff, mem_start, mem_end);
2499 valp = make_room(mem_start, mem_end, 4, 4);
2500 *(__be32 *)valp = cpu_to_be32(node);
2504 /* do all our children */
2505 child = call_prom("child", 1, 1, node);
2506 while (child != 0) {
2507 scan_dt_build_struct(child, mem_start, mem_end);
2508 child = call_prom("peer", 1, 1, child);
2511 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2514 static void __init flatten_device_tree(void)
2517 unsigned long mem_start, mem_end, room;
2518 struct boot_param_header *hdr;
2523 * Check how much room we have between alloc top & bottom (+/- a
2524 * few pages), crop to 1MB, as this is our "chunk" size
2526 room = alloc_top - alloc_bottom - 0x4000;
2527 if (room > DEVTREE_CHUNK_SIZE)
2528 room = DEVTREE_CHUNK_SIZE;
2529 prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2531 /* Now try to claim that */
2532 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2534 prom_panic("Can't allocate initial device-tree chunk\n");
2535 mem_end = mem_start + room;
2537 /* Get root of tree */
2538 root = call_prom("peer", 1, 1, (phandle)0);
2539 if (root == (phandle)0)
2540 prom_panic ("couldn't get device tree root\n");
2542 /* Build header and make room for mem rsv map */
2543 mem_start = _ALIGN(mem_start, 4);
2544 hdr = make_room(&mem_start, &mem_end,
2545 sizeof(struct boot_param_header), 4);
2546 dt_header_start = (unsigned long)hdr;
2547 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2549 /* Start of strings */
2550 mem_start = PAGE_ALIGN(mem_start);
2551 dt_string_start = mem_start;
2552 mem_start += 4; /* hole */
2554 /* Add "linux,phandle" in there, we'll need it */
2555 namep = make_room(&mem_start, &mem_end, 16, 1);
2556 strcpy(namep, "linux,phandle");
2557 mem_start = (unsigned long)namep + strlen(namep) + 1;
2559 /* Build string array */
2560 prom_printf("Building dt strings...\n");
2561 scan_dt_build_strings(root, &mem_start, &mem_end);
2562 dt_string_end = mem_start;
2564 /* Build structure */
2565 mem_start = PAGE_ALIGN(mem_start);
2566 dt_struct_start = mem_start;
2567 prom_printf("Building dt structure...\n");
2568 scan_dt_build_struct(root, &mem_start, &mem_end);
2569 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2570 dt_struct_end = PAGE_ALIGN(mem_start);
2573 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2574 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2575 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2576 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2577 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2578 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2579 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2580 hdr->version = cpu_to_be32(OF_DT_VERSION);
2581 /* Version 16 is not backward compatible */
2582 hdr->last_comp_version = cpu_to_be32(0x10);
2584 /* Copy the reserve map in */
2585 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2590 prom_printf("reserved memory map:\n");
2591 for (i = 0; i < mem_reserve_cnt; i++)
2592 prom_printf(" %llx - %llx\n",
2593 be64_to_cpu(mem_reserve_map[i].base),
2594 be64_to_cpu(mem_reserve_map[i].size));
2597 /* Bump mem_reserve_cnt to cause further reservations to fail
2598 * since it's too late.
2600 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2602 prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2603 dt_string_start, dt_string_end);
2604 prom_printf("Device tree struct 0x%lx -> 0x%lx\n",
2605 dt_struct_start, dt_struct_end);
2608 #ifdef CONFIG_PPC_MAPLE
2609 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2610 * The values are bad, and it doesn't even have the right number of cells. */
2611 static void __init fixup_device_tree_maple(void)
2614 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2618 name = "/ht@0/isa@4";
2619 isa = call_prom("finddevice", 1, 1, ADDR(name));
2620 if (!PHANDLE_VALID(isa)) {
2621 name = "/ht@0/isa@6";
2622 isa = call_prom("finddevice", 1, 1, ADDR(name));
2623 rloc = 0x01003000; /* IO space; PCI device = 6 */
2625 if (!PHANDLE_VALID(isa))
2628 if (prom_getproplen(isa, "ranges") != 12)
2630 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2634 if (isa_ranges[0] != 0x1 ||
2635 isa_ranges[1] != 0xf4000000 ||
2636 isa_ranges[2] != 0x00010000)
2639 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2641 isa_ranges[0] = 0x1;
2642 isa_ranges[1] = 0x0;
2643 isa_ranges[2] = rloc;
2644 isa_ranges[3] = 0x0;
2645 isa_ranges[4] = 0x0;
2646 isa_ranges[5] = 0x00010000;
2647 prom_setprop(isa, name, "ranges",
2648 isa_ranges, sizeof(isa_ranges));
2651 #define CPC925_MC_START 0xf8000000
2652 #define CPC925_MC_LENGTH 0x1000000
2653 /* The values for memory-controller don't have right number of cells */
2654 static void __init fixup_device_tree_maple_memory_controller(void)
2658 char *name = "/hostbridge@f8000000";
2661 mc = call_prom("finddevice", 1, 1, ADDR(name));
2662 if (!PHANDLE_VALID(mc))
2665 if (prom_getproplen(mc, "reg") != 8)
2668 prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2669 prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2670 if ((ac != 2) || (sc != 2))
2673 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2676 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2679 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2682 mc_reg[1] = CPC925_MC_START;
2684 mc_reg[3] = CPC925_MC_LENGTH;
2685 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2688 #define fixup_device_tree_maple()
2689 #define fixup_device_tree_maple_memory_controller()
2692 #ifdef CONFIG_PPC_CHRP
2694 * Pegasos and BriQ lacks the "ranges" property in the isa node
2695 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2696 * Pegasos has the IDE configured in legacy mode, but advertised as native
2698 static void __init fixup_device_tree_chrp(void)
2702 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2706 name = "/pci@80000000/isa@c";
2707 ph = call_prom("finddevice", 1, 1, ADDR(name));
2708 if (!PHANDLE_VALID(ph)) {
2709 name = "/pci@ff500000/isa@6";
2710 ph = call_prom("finddevice", 1, 1, ADDR(name));
2711 rloc = 0x01003000; /* IO space; PCI device = 6 */
2713 if (PHANDLE_VALID(ph)) {
2714 rc = prom_getproplen(ph, "ranges");
2715 if (rc == 0 || rc == PROM_ERROR) {
2716 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2723 prop[5] = 0x00010000;
2724 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2728 name = "/pci@80000000/ide@C,1";
2729 ph = call_prom("finddevice", 1, 1, ADDR(name));
2730 if (PHANDLE_VALID(ph)) {
2731 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2734 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2735 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2736 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2737 if (rc == sizeof(u32)) {
2739 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2744 #define fixup_device_tree_chrp()
2747 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2748 static void __init fixup_device_tree_pmac(void)
2750 phandle u3, i2c, mpic;
2755 /* Some G5s have a missing interrupt definition, fix it up here */
2756 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2757 if (!PHANDLE_VALID(u3))
2759 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2760 if (!PHANDLE_VALID(i2c))
2762 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2763 if (!PHANDLE_VALID(mpic))
2766 /* check if proper rev of u3 */
2767 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2770 if (u3_rev < 0x35 || u3_rev > 0x39)
2772 /* does it need fixup ? */
2773 if (prom_getproplen(i2c, "interrupts") > 0)
2776 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2778 /* interrupt on this revision of u3 is number 0 and level */
2781 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2782 &interrupts, sizeof(interrupts));
2784 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2785 &parent, sizeof(parent));
2788 #define fixup_device_tree_pmac()
2791 #ifdef CONFIG_PPC_EFIKA
2793 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2794 * to talk to the phy. If the phy-handle property is missing, then this
2795 * function is called to add the appropriate nodes and link it to the
2798 static void __init fixup_device_tree_efika_add_phy(void)
2804 /* Check if /builtin/ethernet exists - bail if it doesn't */
2805 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2806 if (!PHANDLE_VALID(node))
2809 /* Check if the phy-handle property exists - bail if it does */
2810 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2815 * At this point the ethernet device doesn't have a phy described.
2816 * Now we need to add the missing phy node and linkage
2819 /* Check for an MDIO bus node - if missing then create one */
2820 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2821 if (!PHANDLE_VALID(node)) {
2822 prom_printf("Adding Ethernet MDIO node\n");
2823 call_prom("interpret", 1, 1,
2824 " s\" /builtin\" find-device"
2826 " 1 encode-int s\" #address-cells\" property"
2827 " 0 encode-int s\" #size-cells\" property"
2828 " s\" mdio\" device-name"
2829 " s\" fsl,mpc5200b-mdio\" encode-string"
2830 " s\" compatible\" property"
2831 " 0xf0003000 0x400 reg"
2833 " 0x5 encode-int encode+"
2834 " 0x3 encode-int encode+"
2835 " s\" interrupts\" property"
2839 /* Check for a PHY device node - if missing then create one and
2840 * give it's phandle to the ethernet node */
2841 node = call_prom("finddevice", 1, 1,
2842 ADDR("/builtin/mdio/ethernet-phy"));
2843 if (!PHANDLE_VALID(node)) {
2844 prom_printf("Adding Ethernet PHY node\n");
2845 call_prom("interpret", 1, 1,
2846 " s\" /builtin/mdio\" find-device"
2848 " s\" ethernet-phy\" device-name"
2849 " 0x10 encode-int s\" reg\" property"
2853 " s\" /builtin/ethernet\" find-device"
2855 " s\" phy-handle\" property"
2860 static void __init fixup_device_tree_efika(void)
2862 int sound_irq[3] = { 2, 2, 0 };
2863 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2864 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2865 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2866 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2871 /* Check if we're really running on a EFIKA */
2872 node = call_prom("finddevice", 1, 1, ADDR("/"));
2873 if (!PHANDLE_VALID(node))
2876 rv = prom_getprop(node, "model", prop, sizeof(prop));
2877 if (rv == PROM_ERROR)
2879 if (strcmp(prop, "EFIKA5K2"))
2882 prom_printf("Applying EFIKA device tree fixups\n");
2884 /* Claiming to be 'chrp' is death */
2885 node = call_prom("finddevice", 1, 1, ADDR("/"));
2886 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2887 if (rv != PROM_ERROR && (strcmp(prop, "chrp") == 0))
2888 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2890 /* CODEGEN,description is exposed in /proc/cpuinfo so
2892 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2893 if (rv != PROM_ERROR && (strstr(prop, "CHRP")))
2894 prom_setprop(node, "/", "CODEGEN,description",
2895 "Efika 5200B PowerPC System",
2896 sizeof("Efika 5200B PowerPC System"));
2898 /* Fixup bestcomm interrupts property */
2899 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2900 if (PHANDLE_VALID(node)) {
2901 len = prom_getproplen(node, "interrupts");
2903 prom_printf("Fixing bestcomm interrupts property\n");
2904 prom_setprop(node, "/builtin/bestcom", "interrupts",
2905 bcomm_irq, sizeof(bcomm_irq));
2909 /* Fixup sound interrupts property */
2910 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2911 if (PHANDLE_VALID(node)) {
2912 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2913 if (rv == PROM_ERROR) {
2914 prom_printf("Adding sound interrupts property\n");
2915 prom_setprop(node, "/builtin/sound", "interrupts",
2916 sound_irq, sizeof(sound_irq));
2920 /* Make sure ethernet phy-handle property exists */
2921 fixup_device_tree_efika_add_phy();
2924 #define fixup_device_tree_efika()
2927 #ifdef CONFIG_PPC_PASEMI_NEMO
2929 * CFE supplied on Nemo is broken in several ways, biggest
2930 * problem is that it reassigns ISA interrupts to unused mpic ints.
2931 * Add an interrupt-controller property for the io-bridge to use
2932 * and correct the ints so we can attach them to an irq_domain
2934 static void __init fixup_device_tree_pasemi(void)
2936 u32 interrupts[2], parent, rval, val = 0;
2937 char *name, *pci_name;
2940 /* Find the root pci node */
2941 name = "/pxp@0,e0000000";
2942 iob = call_prom("finddevice", 1, 1, ADDR(name));
2943 if (!PHANDLE_VALID(iob))
2946 /* check if interrupt-controller node set yet */
2947 if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
2950 prom_printf("adding interrupt-controller property for SB600...\n");
2952 prom_setprop(iob, name, "interrupt-controller", &val, 0);
2954 pci_name = "/pxp@0,e0000000/pci@11";
2955 node = call_prom("finddevice", 1, 1, ADDR(pci_name));
2958 for( ; prom_next_node(&node); ) {
2959 /* scan each node for one with an interrupt */
2960 if (!PHANDLE_VALID(node))
2963 rval = prom_getproplen(node, "interrupts");
2964 if (rval == 0 || rval == PROM_ERROR)
2967 prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
2968 if ((interrupts[0] < 212) || (interrupts[0] > 222))
2971 /* found a node, update both interrupts and interrupt-parent */
2972 if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
2973 interrupts[0] -= 203;
2974 if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
2975 interrupts[0] -= 213;
2976 if (interrupts[0] == 221)
2978 if (interrupts[0] == 222)
2981 prom_setprop(node, pci_name, "interrupts", interrupts,
2982 sizeof(interrupts));
2983 prom_setprop(node, pci_name, "interrupt-parent", &parent,
2988 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
2989 * so that generic isa-bridge code can add the SB600 and its on-board
2992 name = "/pxp@0,e0000000/io-bridge@0";
2993 iob = call_prom("finddevice", 1, 1, ADDR(name));
2994 if (!PHANDLE_VALID(iob))
2997 /* device_type is already set, just change it. */
2999 prom_printf("Changing device_type of SB600 node...\n");
3001 prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3003 #else /* !CONFIG_PPC_PASEMI_NEMO */
3004 static inline void fixup_device_tree_pasemi(void) { }
3007 static void __init fixup_device_tree(void)
3009 fixup_device_tree_maple();
3010 fixup_device_tree_maple_memory_controller();
3011 fixup_device_tree_chrp();
3012 fixup_device_tree_pmac();
3013 fixup_device_tree_efika();
3014 fixup_device_tree_pasemi();
3017 static void __init prom_find_boot_cpu(void)
3024 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3026 prom_cpu = be32_to_cpu(rval);
3028 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3030 if (!PHANDLE_VALID(cpu_pkg))
3033 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3034 prom.cpu = be32_to_cpu(rval);
3036 prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3039 static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3041 #ifdef CONFIG_BLK_DEV_INITRD
3042 if (r3 && r4 && r4 != 0xdeadbeef) {
3045 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3046 prom_initrd_end = prom_initrd_start + r4;
3048 val = cpu_to_be64(prom_initrd_start);
3049 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3051 val = cpu_to_be64(prom_initrd_end);
3052 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3055 reserve_mem(prom_initrd_start,
3056 prom_initrd_end - prom_initrd_start);
3058 prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3059 prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3061 #endif /* CONFIG_BLK_DEV_INITRD */
3065 #ifdef CONFIG_RELOCATABLE
3066 static void reloc_toc(void)
3070 static void unreloc_toc(void)
3074 static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
3077 unsigned long *toc_entry;
3079 /* Get the start of the TOC by using r2 directly. */
3080 asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
3082 for (i = 0; i < nr_entries; i++) {
3083 *toc_entry = *toc_entry + offset;
3088 static void reloc_toc(void)
3090 unsigned long offset = reloc_offset();
3091 unsigned long nr_entries =
3092 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3094 __reloc_toc(offset, nr_entries);
3099 static void unreloc_toc(void)
3101 unsigned long offset = reloc_offset();
3102 unsigned long nr_entries =
3103 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3107 __reloc_toc(-offset, nr_entries);
3113 * We enter here early on, when the Open Firmware prom is still
3114 * handling exceptions and the MMU hash table for us.
3117 unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3119 unsigned long r6, unsigned long r7,
3120 unsigned long kbase)
3125 unsigned long offset = reloc_offset();
3132 * First zero the BSS
3134 memset(&__bss_start, 0, __bss_stop - __bss_start);
3137 * Init interface to Open Firmware, get some node references,
3140 prom_init_client_services(pp);
3143 * See if this OF is old enough that we need to do explicit maps
3144 * and other workarounds
3149 * Init prom stdout device
3153 prom_printf("Preparing to boot %s", linux_banner);
3156 * Get default machine type. At this point, we do not differentiate
3157 * between pSeries SMP and pSeries LPAR
3159 of_platform = prom_find_machine_type();
3160 prom_printf("Detected machine type: %x\n", of_platform);
3162 #ifndef CONFIG_NONSTATIC_KERNEL
3163 /* Bail if this is a kdump kernel. */
3164 if (PHYSICAL_START > 0)
3165 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3169 * Check for an initrd
3171 prom_check_initrd(r3, r4);
3174 * Do early parsing of command line
3176 early_cmdline_parse();
3178 #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
3180 * On pSeries, inform the firmware about our capabilities
3182 if (of_platform == PLATFORM_PSERIES ||
3183 of_platform == PLATFORM_PSERIES_LPAR)
3184 prom_send_capabilities();
3188 * Copy the CPU hold code
3190 if (of_platform != PLATFORM_POWERMAC)
3191 copy_and_flush(0, kbase, 0x100, 0);
3194 * Initialize memory management within prom_init
3199 * Determine which cpu is actually running right _now_
3201 prom_find_boot_cpu();
3204 * Initialize display devices
3206 prom_check_displays();
3208 #if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3210 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3211 * that uses the allocator, we need to make sure we get the top of memory
3212 * available for us here...
3214 if (of_platform == PLATFORM_PSERIES)
3215 prom_initialize_tce_table();
3219 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3220 * have a usable RTAS implementation.
3222 if (of_platform != PLATFORM_POWERMAC &&
3223 of_platform != PLATFORM_OPAL)
3224 prom_instantiate_rtas();
3226 #ifdef CONFIG_PPC_POWERNV
3227 if (of_platform == PLATFORM_OPAL)
3228 prom_instantiate_opal();
3229 #endif /* CONFIG_PPC_POWERNV */
3232 /* instantiate sml */
3233 prom_instantiate_sml();
3237 * On non-powermacs, put all CPUs in spin-loops.
3239 * PowerMacs use a different mechanism to spin CPUs
3241 * (This must be done after instanciating RTAS)
3243 if (of_platform != PLATFORM_POWERMAC &&
3244 of_platform != PLATFORM_OPAL)
3248 * Fill in some infos for use by the kernel later on
3250 if (prom_memory_limit) {
3251 __be64 val = cpu_to_be64(prom_memory_limit);
3252 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3257 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3260 if (prom_iommu_force_on)
3261 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3264 if (prom_tce_alloc_start) {
3265 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3266 &prom_tce_alloc_start,
3267 sizeof(prom_tce_alloc_start));
3268 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3269 &prom_tce_alloc_end,
3270 sizeof(prom_tce_alloc_end));
3275 * Fixup any known bugs in the device-tree
3277 fixup_device_tree();
3280 * Now finally create the flattened device-tree
3282 prom_printf("copying OF device tree...\n");
3283 flatten_device_tree();
3286 * in case stdin is USB and still active on IBM machines...
3287 * Unfortunately quiesce crashes on some powermacs if we have
3288 * closed stdin already (in particular the powerbook 101). It
3289 * appears that the OPAL version of OFW doesn't like it either.
3291 if (of_platform != PLATFORM_POWERMAC &&
3292 of_platform != PLATFORM_OPAL)
3296 * Call OF "quiesce" method to shut down pending DMA's from
3299 prom_printf("Quiescing Open Firmware ...\n");
3300 call_prom("quiesce", 0, 0);
3303 * And finally, call the kernel passing it the flattened device
3304 * tree and NULL as r5, thus triggering the new entry point which
3305 * is common to us and kexec
3307 hdr = dt_header_start;
3309 /* Don't print anything after quiesce under OPAL, it crashes OFW */
3310 if (of_platform != PLATFORM_OPAL) {
3311 prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3312 prom_debug("->dt_header_start=0x%lx\n", hdr);
3316 reloc_got2(-offset);
3321 #ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
3322 /* OPAL early debug gets the OPAL base & entry in r8 and r9 */
3323 __start(hdr, kbase, 0, 0, 0,
3324 prom_opal_base, prom_opal_entry);
3326 __start(hdr, kbase, 0, 0, 0, 0, 0);