1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Procedures for interfacing to Open Firmware.
5 * Paul Mackerras August 1996.
6 * Copyright (C) 1996-2005 Paul Mackerras.
8 * Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
9 * {engebret|bergner}@us.ibm.com
14 /* we cannot use FORTIFY as it brings in new symbols */
18 #include <linux/kernel.h>
19 #include <linux/string.h>
20 #include <linux/init.h>
21 #include <linux/threads.h>
22 #include <linux/spinlock.h>
23 #include <linux/types.h>
24 #include <linux/pci.h>
25 #include <linux/proc_fs.h>
26 #include <linux/delay.h>
27 #include <linux/initrd.h>
28 #include <linux/bitops.h>
32 #include <asm/processor.h>
37 #include <asm/pgtable.h>
38 #include <asm/iommu.h>
39 #include <asm/btext.h>
40 #include <asm/sections.h>
41 #include <asm/machdep.h>
42 #include <asm/asm-prototypes.h>
43 #include <asm/ultravisor-api.h>
45 #include <linux/linux_logo.h>
47 /* All of prom_init bss lives here */
48 #define __prombss __section(.bss.prominit)
51 * Eventually bump that one up
53 #define DEVTREE_CHUNK_SIZE 0x100000
56 * This is the size of the local memory reserve map that gets copied
57 * into the boot params passed to the kernel. That size is totally
58 * flexible as the kernel just reads the list until it encounters an
59 * entry with size 0, so it can be changed without breaking binary
62 #define MEM_RESERVE_MAP_SIZE 8
65 * prom_init() is called very early on, before the kernel text
66 * and data have been mapped to KERNELBASE. At this point the code
67 * is running at whatever address it has been loaded at.
68 * On ppc32 we compile with -mrelocatable, which means that references
69 * to extern and static variables get relocated automatically.
70 * ppc64 objects are always relocatable, we just need to relocate the
73 * Because OF may have mapped I/O devices into the area starting at
74 * KERNELBASE, particularly on CHRP machines, we can't safely call
75 * OF once the kernel has been mapped to KERNELBASE. Therefore all
76 * OF calls must be done within prom_init().
78 * ADDR is used in calls to call_prom. The 4th and following
79 * arguments to call_prom should be 32-bit values.
80 * On ppc64, 64 bit values are truncated to 32 bits (and
81 * fortunately don't get interpreted as two arguments).
83 #define ADDR(x) (u32)(unsigned long)(x)
86 #define OF_WORKAROUNDS 0
88 #define OF_WORKAROUNDS of_workarounds
89 static int of_workarounds __prombss;
92 #define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
93 #define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
95 #define PROM_BUG() do { \
96 prom_printf("kernel BUG at %s line 0x%x!\n", \
97 __FILE__, __LINE__); \
102 #define prom_debug(x...) prom_printf(x)
104 #define prom_debug(x...) do { } while (0)
108 typedef u32 prom_arg_t;
126 struct mem_map_entry {
131 typedef __be32 cell_t;
133 extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
134 unsigned long r6, unsigned long r7, unsigned long r8,
138 extern int enter_prom(struct prom_args *args, unsigned long entry);
140 static inline int enter_prom(struct prom_args *args, unsigned long entry)
142 return ((int (*)(struct prom_args *))entry)(args);
146 extern void copy_and_flush(unsigned long dest, unsigned long src,
147 unsigned long size, unsigned long offset);
150 static struct prom_t __prombss prom;
152 static unsigned long __prombss prom_entry;
154 static char __prombss of_stdout_device[256];
155 static char __prombss prom_scratch[256];
157 static unsigned long __prombss dt_header_start;
158 static unsigned long __prombss dt_struct_start, dt_struct_end;
159 static unsigned long __prombss dt_string_start, dt_string_end;
161 static unsigned long __prombss prom_initrd_start, prom_initrd_end;
164 static int __prombss prom_iommu_force_on;
165 static int __prombss prom_iommu_off;
166 static unsigned long __prombss prom_tce_alloc_start;
167 static unsigned long __prombss prom_tce_alloc_end;
170 #ifdef CONFIG_PPC_PSERIES
171 static bool __prombss prom_radix_disable;
172 static bool __prombss prom_xive_disable;
175 #ifdef CONFIG_PPC_SVM
176 static bool __prombss prom_svm_enable;
179 struct platform_support {
186 /* Platforms codes are now obsolete in the kernel. Now only used within this
187 * file and ultimately gone too. Feel free to change them if you need, they
188 * are not shared with anything outside of this file anymore
190 #define PLATFORM_PSERIES 0x0100
191 #define PLATFORM_PSERIES_LPAR 0x0101
192 #define PLATFORM_LPAR 0x0001
193 #define PLATFORM_POWERMAC 0x0400
194 #define PLATFORM_GENERIC 0x0500
196 static int __prombss of_platform;
198 static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
200 static unsigned long __prombss prom_memory_limit;
202 static unsigned long __prombss alloc_top;
203 static unsigned long __prombss alloc_top_high;
204 static unsigned long __prombss alloc_bottom;
205 static unsigned long __prombss rmo_top;
206 static unsigned long __prombss ram_top;
208 static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
209 static int __prombss mem_reserve_cnt;
211 static cell_t __prombss regbuf[1024];
213 static bool __prombss rtas_has_query_cpu_stopped;
217 * Error results ... some OF calls will return "-1" on error, some
218 * will return 0, some will return either. To simplify, here are
219 * macros to use with any ihandle or phandle return value to check if
223 #define PROM_ERROR (-1u)
224 #define PHANDLE_VALID(p) ((p) != 0 && (p) != PROM_ERROR)
225 #define IHANDLE_VALID(i) ((i) != 0 && (i) != PROM_ERROR)
227 /* Copied from lib/string.c and lib/kstrtox.c */
229 static int __init prom_strcmp(const char *cs, const char *ct)
231 unsigned char c1, c2;
237 return c1 < c2 ? -1 : 1;
244 static char __init *prom_strcpy(char *dest, const char *src)
248 while ((*dest++ = *src++) != '\0')
253 static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
255 unsigned char c1, c2;
261 return c1 < c2 ? -1 : 1;
269 static size_t __init prom_strlen(const char *s)
273 for (sc = s; *sc != '\0'; ++sc)
278 static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
280 const unsigned char *su1, *su2;
283 for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
284 if ((res = *su1 - *su2) != 0)
289 static char __init *prom_strstr(const char *s1, const char *s2)
293 l2 = prom_strlen(s2);
296 l1 = prom_strlen(s1);
299 if (!prom_memcmp(s1, s2, l2))
306 static size_t __init prom_strlcpy(char *dest, const char *src, size_t size)
308 size_t ret = prom_strlen(src);
311 size_t len = (ret >= size) ? size - 1 : ret;
312 memcpy(dest, src, len);
318 #ifdef CONFIG_PPC_PSERIES
319 static int __init prom_strtobool(const char *s, bool *res)
357 /* This is the one and *ONLY* place where we actually call open
361 static int __init call_prom(const char *service, int nargs, int nret, ...)
364 struct prom_args args;
367 args.service = cpu_to_be32(ADDR(service));
368 args.nargs = cpu_to_be32(nargs);
369 args.nret = cpu_to_be32(nret);
371 va_start(list, nret);
372 for (i = 0; i < nargs; i++)
373 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
376 for (i = 0; i < nret; i++)
377 args.args[nargs+i] = 0;
379 if (enter_prom(&args, prom_entry) < 0)
382 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
385 static int __init call_prom_ret(const char *service, int nargs, int nret,
386 prom_arg_t *rets, ...)
389 struct prom_args args;
392 args.service = cpu_to_be32(ADDR(service));
393 args.nargs = cpu_to_be32(nargs);
394 args.nret = cpu_to_be32(nret);
396 va_start(list, rets);
397 for (i = 0; i < nargs; i++)
398 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
401 for (i = 0; i < nret; i++)
402 args.args[nargs+i] = 0;
404 if (enter_prom(&args, prom_entry) < 0)
408 for (i = 1; i < nret; ++i)
409 rets[i-1] = be32_to_cpu(args.args[nargs+i]);
411 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
415 static void __init prom_print(const char *msg)
419 if (prom.stdout == 0)
422 for (p = msg; *p != 0; p = q) {
423 for (q = p; *q != 0 && *q != '\n'; ++q)
426 call_prom("write", 3, 1, prom.stdout, p, q - p);
430 call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
436 * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
437 * we do not need __udivdi3 or __umoddi3 on 32bits.
439 static void __init prom_print_hex(unsigned long val)
441 int i, nibbles = sizeof(val)*2;
442 char buf[sizeof(val)*2+1];
444 for (i = nibbles-1; i >= 0; i--) {
445 buf[i] = (val & 0xf) + '0';
447 buf[i] += ('a'-'0'-10);
451 call_prom("write", 3, 1, prom.stdout, buf, nibbles);
454 /* max number of decimal digits in an unsigned long */
456 static void __init prom_print_dec(unsigned long val)
459 char buf[UL_DIGITS+1];
461 for (i = UL_DIGITS-1; i >= 0; i--) {
462 buf[i] = (val % 10) + '0';
467 /* shift stuff down */
468 size = UL_DIGITS - i;
469 call_prom("write", 3, 1, prom.stdout, buf+i, size);
473 static void __init prom_printf(const char *format, ...)
475 const char *p, *q, *s;
481 va_start(args, format);
482 for (p = format; *p != 0; p = q) {
483 for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
486 call_prom("write", 3, 1, prom.stdout, p, q - p);
491 call_prom("write", 3, 1, prom.stdout,
505 s = va_arg(args, const char *);
512 v = va_arg(args, unsigned int);
515 v = va_arg(args, unsigned long);
519 v = va_arg(args, unsigned long long);
528 v = va_arg(args, unsigned int);
531 v = va_arg(args, unsigned long);
535 v = va_arg(args, unsigned long long);
544 vs = va_arg(args, int);
547 vs = va_arg(args, long);
551 vs = va_arg(args, long long);
566 static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
570 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
572 * Old OF requires we claim physical and virtual separately
573 * and then map explicitly (assuming virtual mode)
578 ret = call_prom_ret("call-method", 5, 2, &result,
579 ADDR("claim"), prom.memory,
581 if (ret != 0 || result == -1)
583 ret = call_prom_ret("call-method", 5, 2, &result,
584 ADDR("claim"), prom.mmumap,
587 call_prom("call-method", 4, 1, ADDR("release"),
588 prom.memory, size, virt);
591 /* the 0x12 is M (coherence) + PP == read/write */
592 call_prom("call-method", 6, 1,
593 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
596 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
600 static void __init __attribute__((noreturn)) prom_panic(const char *reason)
603 /* Do not call exit because it clears the screen on pmac
604 * it also causes some sort of double-fault on early pmacs */
605 if (of_platform == PLATFORM_POWERMAC)
608 /* ToDo: should put up an SRC here on pSeries */
609 call_prom("exit", 0, 0);
611 for (;;) /* should never get here */
616 static int __init prom_next_node(phandle *nodep)
620 if ((node = *nodep) != 0
621 && (*nodep = call_prom("child", 1, 1, node)) != 0)
623 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
626 if ((node = call_prom("parent", 1, 1, node)) == 0)
628 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
633 static inline int __init prom_getprop(phandle node, const char *pname,
634 void *value, size_t valuelen)
636 return call_prom("getprop", 4, 1, node, ADDR(pname),
637 (u32)(unsigned long) value, (u32) valuelen);
640 static inline int __init prom_getproplen(phandle node, const char *pname)
642 return call_prom("getproplen", 2, 1, node, ADDR(pname));
645 static void add_string(char **str, const char *q)
655 static char *tohex(unsigned int x)
657 static const char digits[] __initconst = "0123456789abcdef";
658 static char result[9] __prombss;
665 result[i] = digits[x & 0xf];
667 } while (x != 0 && i > 0);
671 static int __init prom_setprop(phandle node, const char *nodename,
672 const char *pname, void *value, size_t valuelen)
676 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
677 return call_prom("setprop", 4, 1, node, ADDR(pname),
678 (u32)(unsigned long) value, (u32) valuelen);
680 /* gah... setprop doesn't work on longtrail, have to use interpret */
682 add_string(&p, "dev");
683 add_string(&p, nodename);
684 add_string(&p, tohex((u32)(unsigned long) value));
685 add_string(&p, tohex(valuelen));
686 add_string(&p, tohex(ADDR(pname)));
687 add_string(&p, tohex(prom_strlen(pname)));
688 add_string(&p, "property");
690 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
693 /* We can't use the standard versions because of relocation headaches. */
694 #define isxdigit(c) (('0' <= (c) && (c) <= '9') \
695 || ('a' <= (c) && (c) <= 'f') \
696 || ('A' <= (c) && (c) <= 'F'))
698 #define isdigit(c) ('0' <= (c) && (c) <= '9')
699 #define islower(c) ('a' <= (c) && (c) <= 'z')
700 #define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
702 static unsigned long prom_strtoul(const char *cp, const char **endp)
704 unsigned long result = 0, base = 10, value;
709 if (toupper(*cp) == 'X') {
715 while (isxdigit(*cp) &&
716 (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
717 result = result * base + value;
727 static unsigned long prom_memparse(const char *ptr, const char **retptr)
729 unsigned long ret = prom_strtoul(ptr, retptr);
733 * We can't use a switch here because GCC *may* generate a
734 * jump table which won't work, because we're not running at
735 * the address we're linked at.
737 if ('G' == **retptr || 'g' == **retptr)
740 if ('M' == **retptr || 'm' == **retptr)
743 if ('K' == **retptr || 'k' == **retptr)
755 * Early parsing of the command line passed to the kernel, used for
756 * "mem=x" and the options that affect the iommu
758 static void __init early_cmdline_parse(void)
765 prom_cmd_line[0] = 0;
767 if ((long)prom.chosen > 0)
768 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
769 if (IS_ENABLED(CONFIG_CMDLINE_BOOL) && (l <= 0 || p[0] == '\0')) /* dbl check */
770 prom_strlcpy(prom_cmd_line, CONFIG_CMDLINE, sizeof(prom_cmd_line));
771 prom_printf("command line: %s\n", prom_cmd_line);
774 opt = prom_strstr(prom_cmd_line, "iommu=");
776 prom_printf("iommu opt is: %s\n", opt);
778 while (*opt && *opt == ' ')
780 if (!prom_strncmp(opt, "off", 3))
782 else if (!prom_strncmp(opt, "force", 5))
783 prom_iommu_force_on = 1;
786 opt = prom_strstr(prom_cmd_line, "mem=");
789 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
791 /* Align to 16 MB == size of ppc64 large page */
792 prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
796 #ifdef CONFIG_PPC_PSERIES
797 prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
798 opt = prom_strstr(prom_cmd_line, "disable_radix");
801 if (*opt && *opt == '=') {
804 if (prom_strtobool(++opt, &val))
805 prom_radix_disable = false;
807 prom_radix_disable = val;
809 prom_radix_disable = true;
811 if (prom_radix_disable)
812 prom_debug("Radix disabled from cmdline\n");
814 opt = prom_strstr(prom_cmd_line, "xive=off");
816 prom_xive_disable = true;
817 prom_debug("XIVE disabled from cmdline\n");
819 #endif /* CONFIG_PPC_PSERIES */
821 #ifdef CONFIG_PPC_SVM
822 opt = prom_strstr(prom_cmd_line, "svm=");
826 opt += sizeof("svm=") - 1;
827 if (!prom_strtobool(opt, &val))
828 prom_svm_enable = val;
830 #endif /* CONFIG_PPC_SVM */
833 #ifdef CONFIG_PPC_PSERIES
835 * The architecture vector has an array of PVR mask/value pairs,
836 * followed by # option vectors - 1, followed by the option vectors.
838 * See prom.h for the definition of the bits specified in the
839 * architecture vector.
842 /* Firmware expects the value to be n - 1, where n is the # of vectors */
843 #define NUM_VECTORS(n) ((n) - 1)
846 * Firmware expects 1 + n - 2, where n is the length of the option vector in
847 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
849 #define VECTOR_LENGTH(n) (1 + (n) - 2)
851 struct option_vector1 {
857 struct option_vector2 {
871 struct option_vector3 {
876 struct option_vector4 {
881 struct option_vector5 {
893 u8 platform_facilities;
904 struct option_vector6 {
910 struct ibm_arch_vec {
911 struct { u32 mask, val; } pvrs[12];
916 struct option_vector1 vec1;
919 struct option_vector2 vec2;
922 struct option_vector3 vec3;
925 struct option_vector4 vec4;
928 struct option_vector5 vec5;
931 struct option_vector6 vec6;
934 static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
937 .mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
938 .val = cpu_to_be32(0x003a0000),
941 .mask = cpu_to_be32(0xffff0000), /* POWER6 */
942 .val = cpu_to_be32(0x003e0000),
945 .mask = cpu_to_be32(0xffff0000), /* POWER7 */
946 .val = cpu_to_be32(0x003f0000),
949 .mask = cpu_to_be32(0xffff0000), /* POWER8E */
950 .val = cpu_to_be32(0x004b0000),
953 .mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
954 .val = cpu_to_be32(0x004c0000),
957 .mask = cpu_to_be32(0xffff0000), /* POWER8 */
958 .val = cpu_to_be32(0x004d0000),
961 .mask = cpu_to_be32(0xffff0000), /* POWER9 */
962 .val = cpu_to_be32(0x004e0000),
965 .mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
966 .val = cpu_to_be32(0x0f000005),
969 .mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
970 .val = cpu_to_be32(0x0f000004),
973 .mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
974 .val = cpu_to_be32(0x0f000003),
977 .mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
978 .val = cpu_to_be32(0x0f000002),
981 .mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
982 .val = cpu_to_be32(0x0f000001),
986 .num_vectors = NUM_VECTORS(6),
988 .vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
991 .arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
992 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
993 .arch_versions3 = OV1_PPC_3_00,
996 .vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
997 /* option vector 2: Open Firmware options supported */
999 .byte1 = OV2_REAL_MODE,
1001 .real_base = cpu_to_be32(0xffffffff),
1002 .real_size = cpu_to_be32(0xffffffff),
1003 .virt_base = cpu_to_be32(0xffffffff),
1004 .virt_size = cpu_to_be32(0xffffffff),
1005 .load_base = cpu_to_be32(0xffffffff),
1006 .min_rma = cpu_to_be32(512), /* 512MB min RMA */
1007 .min_load = cpu_to_be32(0xffffffff), /* full client load */
1008 .min_rma_percent = 0, /* min RMA percentage of total RAM */
1009 .max_pft_size = 48, /* max log_2(hash table size) */
1012 .vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1013 /* option vector 3: processor options supported */
1015 .byte1 = 0, /* don't ignore, don't halt */
1016 .byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1019 .vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1020 /* option vector 4: IBM PAPR implementation */
1022 .byte1 = 0, /* don't halt */
1023 .min_vp_cap = OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
1026 .vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1027 /* option vector 5: PAPR/OF options */
1029 .byte1 = 0, /* don't ignore, don't halt */
1030 .byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1031 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1032 #ifdef CONFIG_PCI_MSI
1033 /* PCIe/MSI support. Without MSI full PCIe is not supported */
1040 #ifdef CONFIG_PPC_SMLPAR
1041 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1045 .associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
1046 .bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1047 .micro_checkpoint = 0,
1049 .max_cpus = cpu_to_be32(NR_CPUS), /* number of cores supported */
1052 .platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1056 .byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1063 /* option vector 6: IBM PAPR hints */
1064 .vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1067 .secondary_pteg = 0,
1068 .os_name = OV6_LINUX,
1072 static struct ibm_arch_vec __prombss ibm_architecture_vec ____cacheline_aligned;
1074 /* Old method - ELF header with PT_NOTE sections only works on BE */
1075 #ifdef __BIG_ENDIAN__
1076 static const struct fake_elf {
1083 char name[8]; /* "PowerPC" */
1097 char name[24]; /* "IBM,RPA-Client-Config" */
1101 u32 min_rmo_percent;
1109 } fake_elf __initconst = {
1111 .e_ident = { 0x7f, 'E', 'L', 'F',
1112 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1113 .e_type = ET_EXEC, /* yeah right */
1114 .e_machine = EM_PPC,
1115 .e_version = EV_CURRENT,
1116 .e_phoff = offsetof(struct fake_elf, phdr),
1117 .e_phentsize = sizeof(Elf32_Phdr),
1123 .p_offset = offsetof(struct fake_elf, chrpnote),
1124 .p_filesz = sizeof(struct chrpnote)
1127 .p_offset = offsetof(struct fake_elf, rpanote),
1128 .p_filesz = sizeof(struct rpanote)
1132 .namesz = sizeof("PowerPC"),
1133 .descsz = sizeof(struct chrpdesc),
1137 .real_mode = ~0U, /* ~0 means "don't care" */
1146 .namesz = sizeof("IBM,RPA-Client-Config"),
1147 .descsz = sizeof(struct rpadesc),
1149 .name = "IBM,RPA-Client-Config",
1152 .min_rmo_size = 64, /* in megabytes */
1153 .min_rmo_percent = 0,
1154 .max_pft_size = 48, /* 2^48 bytes max PFT size */
1161 #endif /* __BIG_ENDIAN__ */
1163 static int __init prom_count_smt_threads(void)
1169 /* Pick up th first CPU node we can find */
1170 for (node = 0; prom_next_node(&node); ) {
1172 prom_getprop(node, "device_type", type, sizeof(type));
1174 if (prom_strcmp(type, "cpu"))
1177 * There is an entry for each smt thread, each entry being
1178 * 4 bytes long. All cpus should have the same number of
1179 * smt threads, so return after finding the first.
1181 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1182 if (plen == PROM_ERROR)
1185 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1188 if (plen < 1 || plen > 64) {
1189 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1190 (unsigned long)plen);
1195 prom_debug("No threads found, assuming 1 per core\n");
1201 static void __init prom_parse_mmu_model(u8 val,
1202 struct platform_support *support)
1205 case OV5_FEAT(OV5_MMU_DYNAMIC):
1206 case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1207 prom_debug("MMU - either supported\n");
1208 support->radix_mmu = !prom_radix_disable;
1209 support->hash_mmu = true;
1211 case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1212 prom_debug("MMU - radix only\n");
1213 if (prom_radix_disable) {
1215 * If we __have__ to do radix, we're better off ignoring
1216 * the command line rather than not booting.
1218 prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1220 support->radix_mmu = true;
1222 case OV5_FEAT(OV5_MMU_HASH):
1223 prom_debug("MMU - hash only\n");
1224 support->hash_mmu = true;
1227 prom_debug("Unknown mmu support option: 0x%x\n", val);
1232 static void __init prom_parse_xive_model(u8 val,
1233 struct platform_support *support)
1236 case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1237 prom_debug("XIVE - either mode supported\n");
1238 support->xive = !prom_xive_disable;
1240 case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1241 prom_debug("XIVE - exploitation mode supported\n");
1242 if (prom_xive_disable) {
1244 * If we __have__ to do XIVE, we're better off ignoring
1245 * the command line rather than not booting.
1247 prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1249 support->xive = true;
1251 case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1252 prom_debug("XIVE - legacy mode supported\n");
1255 prom_debug("Unknown xive support option: 0x%x\n", val);
1260 static void __init prom_parse_platform_support(u8 index, u8 val,
1261 struct platform_support *support)
1264 case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1265 prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1267 case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1268 if (val & OV5_FEAT(OV5_RADIX_GTSE)) {
1269 prom_debug("Radix - GTSE supported\n");
1270 support->radix_gtse = true;
1273 case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1274 prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1280 static void __init prom_check_platform_support(void)
1282 struct platform_support supported = {
1285 .radix_gtse = false,
1288 int prop_len = prom_getproplen(prom.chosen,
1289 "ibm,arch-vec-5-platform-support");
1292 * First copy the architecture vec template
1294 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1295 * by __memcpy() when KASAN is active
1297 memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1298 sizeof(ibm_architecture_vec));
1303 prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1305 if (prop_len > sizeof(vec))
1306 prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1308 prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support", &vec, sizeof(vec));
1309 for (i = 0; i < prop_len; i += 2) {
1310 prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2, vec[i], vec[i + 1]);
1311 prom_parse_platform_support(vec[i], vec[i + 1], &supported);
1315 if (supported.radix_mmu && supported.radix_gtse &&
1316 IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1317 /* Radix preferred - but we require GTSE for now */
1318 prom_debug("Asking for radix with GTSE\n");
1319 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1320 ibm_architecture_vec.vec5.radix_ext = OV5_FEAT(OV5_RADIX_GTSE);
1321 } else if (supported.hash_mmu) {
1322 /* Default to hash mmu (if we can) */
1323 prom_debug("Asking for hash\n");
1324 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1326 /* We're probably on a legacy hypervisor */
1327 prom_debug("Assuming legacy hash support\n");
1330 if (supported.xive) {
1331 prom_debug("Asking for XIVE\n");
1332 ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1336 static void __init prom_send_capabilities(void)
1342 /* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1343 prom_check_platform_support();
1345 root = call_prom("open", 1, 1, ADDR("/"));
1347 /* We need to tell the FW about the number of cores we support.
1349 * To do that, we count the number of threads on the first core
1350 * (we assume this is the same for all cores) and use it to
1354 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1355 prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1358 ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1360 /* try calling the ibm,client-architecture-support method */
1361 prom_printf("Calling ibm,client-architecture-support...");
1362 if (call_prom_ret("call-method", 3, 2, &ret,
1363 ADDR("ibm,client-architecture-support"),
1365 ADDR(&ibm_architecture_vec)) == 0) {
1366 /* the call exists... */
1368 prom_printf("\nWARNING: ibm,client-architecture"
1369 "-support call FAILED!\n");
1370 call_prom("close", 1, 0, root);
1371 prom_printf(" done\n");
1374 call_prom("close", 1, 0, root);
1375 prom_printf(" not implemented\n");
1378 #ifdef __BIG_ENDIAN__
1382 /* no ibm,client-architecture-support call, try the old way */
1383 elfloader = call_prom("open", 1, 1,
1384 ADDR("/packages/elf-loader"));
1385 if (elfloader == 0) {
1386 prom_printf("couldn't open /packages/elf-loader\n");
1389 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1390 elfloader, ADDR(&fake_elf));
1391 call_prom("close", 1, 0, elfloader);
1393 #endif /* __BIG_ENDIAN__ */
1395 #endif /* CONFIG_PPC_PSERIES */
1398 * Memory allocation strategy... our layout is normally:
1400 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
1401 * rare cases, initrd might end up being before the kernel though.
1402 * We assume this won't override the final kernel at 0, we have no
1403 * provision to handle that in this version, but it should hopefully
1406 * alloc_top is set to the top of RMO, eventually shrink down if the
1409 * alloc_bottom is set to the top of kernel/initrd
1411 * from there, allocations are done this way : rtas is allocated
1412 * topmost, and the device-tree is allocated from the bottom. We try
1413 * to grow the device-tree allocation as we progress. If we can't,
1414 * then we fail, we don't currently have a facility to restart
1415 * elsewhere, but that shouldn't be necessary.
1417 * Note that calls to reserve_mem have to be done explicitly, memory
1418 * allocated with either alloc_up or alloc_down isn't automatically
1424 * Allocates memory in the RMO upward from the kernel/initrd
1426 * When align is 0, this is a special case, it means to allocate in place
1427 * at the current location of alloc_bottom or fail (that is basically
1428 * extending the previous allocation). Used for the device-tree flattening
1430 static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1432 unsigned long base = alloc_bottom;
1433 unsigned long addr = 0;
1436 base = _ALIGN_UP(base, align);
1437 prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1439 prom_panic("alloc_up() called with mem not initialized\n");
1442 base = _ALIGN_UP(alloc_bottom, align);
1444 base = alloc_bottom;
1446 for(; (base + size) <= alloc_top;
1447 base = _ALIGN_UP(base + 0x100000, align)) {
1448 prom_debug(" trying: 0x%lx\n\r", base);
1449 addr = (unsigned long)prom_claim(base, size, 0);
1450 if (addr != PROM_ERROR && addr != 0)
1458 alloc_bottom = addr + size;
1460 prom_debug(" -> %lx\n", addr);
1461 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1462 prom_debug(" alloc_top : %lx\n", alloc_top);
1463 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1464 prom_debug(" rmo_top : %lx\n", rmo_top);
1465 prom_debug(" ram_top : %lx\n", ram_top);
1471 * Allocates memory downward, either from top of RMO, or if highmem
1472 * is set, from the top of RAM. Note that this one doesn't handle
1473 * failures. It does claim memory if highmem is not set.
1475 static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1478 unsigned long base, addr = 0;
1480 prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1481 highmem ? "(high)" : "(low)");
1483 prom_panic("alloc_down() called with mem not initialized\n");
1486 /* Carve out storage for the TCE table. */
1487 addr = _ALIGN_DOWN(alloc_top_high - size, align);
1488 if (addr <= alloc_bottom)
1490 /* Will we bump into the RMO ? If yes, check out that we
1491 * didn't overlap existing allocations there, if we did,
1492 * we are dead, we must be the first in town !
1494 if (addr < rmo_top) {
1495 /* Good, we are first */
1496 if (alloc_top == rmo_top)
1497 alloc_top = rmo_top = addr;
1501 alloc_top_high = addr;
1505 base = _ALIGN_DOWN(alloc_top - size, align);
1506 for (; base > alloc_bottom;
1507 base = _ALIGN_DOWN(base - 0x100000, align)) {
1508 prom_debug(" trying: 0x%lx\n\r", base);
1509 addr = (unsigned long)prom_claim(base, size, 0);
1510 if (addr != PROM_ERROR && addr != 0)
1519 prom_debug(" -> %lx\n", addr);
1520 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1521 prom_debug(" alloc_top : %lx\n", alloc_top);
1522 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1523 prom_debug(" rmo_top : %lx\n", rmo_top);
1524 prom_debug(" ram_top : %lx\n", ram_top);
1530 * Parse a "reg" cell
1532 static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1535 unsigned long r = 0;
1537 /* Ignore more than 2 cells */
1538 while (s > sizeof(unsigned long) / 4) {
1542 r = be32_to_cpu(*p++);
1546 r |= be32_to_cpu(*(p++));
1554 * Very dumb function for adding to the memory reserve list, but
1555 * we don't need anything smarter at this point
1557 * XXX Eventually check for collisions. They should NEVER happen.
1558 * If problems seem to show up, it would be a good start to track
1561 static void __init reserve_mem(u64 base, u64 size)
1563 u64 top = base + size;
1564 unsigned long cnt = mem_reserve_cnt;
1569 /* We need to always keep one empty entry so that we
1570 * have our terminator with "size" set to 0 since we are
1571 * dumb and just copy this entire array to the boot params
1573 base = _ALIGN_DOWN(base, PAGE_SIZE);
1574 top = _ALIGN_UP(top, PAGE_SIZE);
1577 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1578 prom_panic("Memory reserve map exhausted !\n");
1579 mem_reserve_map[cnt].base = cpu_to_be64(base);
1580 mem_reserve_map[cnt].size = cpu_to_be64(size);
1581 mem_reserve_cnt = cnt + 1;
1585 * Initialize memory allocation mechanism, parse "memory" nodes and
1586 * obtain that way the top of memory and RMO to setup out local allocator
1588 static void __init prom_init_mem(void)
1598 * We iterate the memory nodes to find
1599 * 1) top of RMO (first node)
1602 val = cpu_to_be32(2);
1603 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1604 rac = be32_to_cpu(val);
1605 val = cpu_to_be32(1);
1606 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1607 rsc = be32_to_cpu(val);
1608 prom_debug("root_addr_cells: %x\n", rac);
1609 prom_debug("root_size_cells: %x\n", rsc);
1611 prom_debug("scanning memory:\n");
1613 for (node = 0; prom_next_node(&node); ) {
1615 prom_getprop(node, "device_type", type, sizeof(type));
1619 * CHRP Longtrail machines have no device_type
1620 * on the memory node, so check the name instead...
1622 prom_getprop(node, "name", type, sizeof(type));
1624 if (prom_strcmp(type, "memory"))
1627 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1628 if (plen > sizeof(regbuf)) {
1629 prom_printf("memory node too large for buffer !\n");
1630 plen = sizeof(regbuf);
1633 endp = p + (plen / sizeof(cell_t));
1636 memset(prom_scratch, 0, sizeof(prom_scratch));
1637 call_prom("package-to-path", 3, 1, node, prom_scratch,
1638 sizeof(prom_scratch) - 1);
1639 prom_debug(" node %s :\n", prom_scratch);
1640 #endif /* DEBUG_PROM */
1642 while ((endp - p) >= (rac + rsc)) {
1643 unsigned long base, size;
1645 base = prom_next_cell(rac, &p);
1646 size = prom_next_cell(rsc, &p);
1650 prom_debug(" %lx %lx\n", base, size);
1651 if (base == 0 && (of_platform & PLATFORM_LPAR))
1653 if ((base + size) > ram_top)
1654 ram_top = base + size;
1658 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1661 * If prom_memory_limit is set we reduce the upper limits *except* for
1662 * alloc_top_high. This must be the real top of RAM so we can put
1666 alloc_top_high = ram_top;
1668 if (prom_memory_limit) {
1669 if (prom_memory_limit <= alloc_bottom) {
1670 prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1672 prom_memory_limit = 0;
1673 } else if (prom_memory_limit >= ram_top) {
1674 prom_printf("Ignoring mem=%lx >= ram_top.\n",
1676 prom_memory_limit = 0;
1678 ram_top = prom_memory_limit;
1679 rmo_top = min(rmo_top, prom_memory_limit);
1684 * Setup our top alloc point, that is top of RMO or top of
1685 * segment 0 when running non-LPAR.
1686 * Some RS64 machines have buggy firmware where claims up at
1687 * 1GB fail. Cap at 768MB as a workaround.
1688 * Since 768MB is plenty of room, and we need to cap to something
1689 * reasonable on 32-bit, cap at 768MB on all machines.
1693 rmo_top = min(0x30000000ul, rmo_top);
1694 alloc_top = rmo_top;
1695 alloc_top_high = ram_top;
1698 * Check if we have an initrd after the kernel but still inside
1699 * the RMO. If we do move our bottom point to after it.
1701 if (prom_initrd_start &&
1702 prom_initrd_start < rmo_top &&
1703 prom_initrd_end > alloc_bottom)
1704 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1706 prom_printf("memory layout at init:\n");
1707 prom_printf(" memory_limit : %lx (16 MB aligned)\n",
1709 prom_printf(" alloc_bottom : %lx\n", alloc_bottom);
1710 prom_printf(" alloc_top : %lx\n", alloc_top);
1711 prom_printf(" alloc_top_hi : %lx\n", alloc_top_high);
1712 prom_printf(" rmo_top : %lx\n", rmo_top);
1713 prom_printf(" ram_top : %lx\n", ram_top);
1716 static void __init prom_close_stdin(void)
1721 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1722 stdin = be32_to_cpu(val);
1723 call_prom("close", 1, 0, stdin);
1727 #ifdef CONFIG_PPC_SVM
1728 static int prom_rtas_hcall(uint64_t args)
1730 register uint64_t arg1 asm("r3") = H_RTAS;
1731 register uint64_t arg2 asm("r4") = args;
1733 asm volatile("sc 1\n" : "=r" (arg1) :
1739 static struct rtas_args __prombss os_term_args;
1741 static void __init prom_rtas_os_term(char *str)
1747 prom_debug("%s: start...\n", __func__);
1748 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1749 prom_debug("rtas_node: %x\n", rtas_node);
1750 if (!PHANDLE_VALID(rtas_node))
1754 prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1755 token = be32_to_cpu(val);
1756 prom_debug("ibm,os-term: %x\n", token);
1758 prom_panic("Could not get token for ibm,os-term\n");
1759 os_term_args.token = cpu_to_be32(token);
1760 os_term_args.nargs = cpu_to_be32(1);
1761 os_term_args.nret = cpu_to_be32(1);
1762 os_term_args.args[0] = cpu_to_be32(__pa(str));
1763 prom_rtas_hcall((uint64_t)&os_term_args);
1765 #endif /* CONFIG_PPC_SVM */
1768 * Allocate room for and instantiate RTAS
1770 static void __init prom_instantiate_rtas(void)
1774 u32 base, entry = 0;
1778 prom_debug("prom_instantiate_rtas: start...\n");
1780 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1781 prom_debug("rtas_node: %x\n", rtas_node);
1782 if (!PHANDLE_VALID(rtas_node))
1786 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1787 size = be32_to_cpu(val);
1791 base = alloc_down(size, PAGE_SIZE, 0);
1793 prom_panic("Could not allocate memory for RTAS\n");
1795 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1796 if (!IHANDLE_VALID(rtas_inst)) {
1797 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1801 prom_printf("instantiating rtas at 0x%x...", base);
1803 if (call_prom_ret("call-method", 3, 2, &entry,
1804 ADDR("instantiate-rtas"),
1805 rtas_inst, base) != 0
1807 prom_printf(" failed\n");
1810 prom_printf(" done\n");
1812 reserve_mem(base, size);
1814 val = cpu_to_be32(base);
1815 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1817 val = cpu_to_be32(entry);
1818 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1821 /* Check if it supports "query-cpu-stopped-state" */
1822 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1823 &val, sizeof(val)) != PROM_ERROR)
1824 rtas_has_query_cpu_stopped = true;
1826 prom_debug("rtas base = 0x%x\n", base);
1827 prom_debug("rtas entry = 0x%x\n", entry);
1828 prom_debug("rtas size = 0x%x\n", size);
1830 prom_debug("prom_instantiate_rtas: end...\n");
1835 * Allocate room for and instantiate Stored Measurement Log (SML)
1837 static void __init prom_instantiate_sml(void)
1839 phandle ibmvtpm_node;
1840 ihandle ibmvtpm_inst;
1841 u32 entry = 0, size = 0, succ = 0;
1845 prom_debug("prom_instantiate_sml: start...\n");
1847 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1848 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1849 if (!PHANDLE_VALID(ibmvtpm_node))
1852 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1853 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1854 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1858 if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1859 &val, sizeof(val)) != PROM_ERROR) {
1860 if (call_prom_ret("call-method", 2, 2, &succ,
1861 ADDR("reformat-sml-to-efi-alignment"),
1862 ibmvtpm_inst) != 0 || succ == 0) {
1863 prom_printf("Reformat SML to EFI alignment failed\n");
1867 if (call_prom_ret("call-method", 2, 2, &size,
1868 ADDR("sml-get-allocated-size"),
1869 ibmvtpm_inst) != 0 || size == 0) {
1870 prom_printf("SML get allocated size failed\n");
1874 if (call_prom_ret("call-method", 2, 2, &size,
1875 ADDR("sml-get-handover-size"),
1876 ibmvtpm_inst) != 0 || size == 0) {
1877 prom_printf("SML get handover size failed\n");
1882 base = alloc_down(size, PAGE_SIZE, 0);
1884 prom_panic("Could not allocate memory for sml\n");
1886 prom_printf("instantiating sml at 0x%llx...", base);
1888 memset((void *)base, 0, size);
1890 if (call_prom_ret("call-method", 4, 2, &entry,
1891 ADDR("sml-handover"),
1892 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1893 prom_printf("SML handover failed\n");
1896 prom_printf(" done\n");
1898 reserve_mem(base, size);
1900 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1901 &base, sizeof(base));
1902 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1903 &size, sizeof(size));
1905 prom_debug("sml base = 0x%llx\n", base);
1906 prom_debug("sml size = 0x%x\n", size);
1908 prom_debug("prom_instantiate_sml: end...\n");
1912 * Allocate room for and initialize TCE tables
1914 #ifdef __BIG_ENDIAN__
1915 static void __init prom_initialize_tce_table(void)
1919 char compatible[64], type[64], model[64];
1920 char *path = prom_scratch;
1922 u32 minalign, minsize;
1923 u64 tce_entry, *tce_entryp;
1924 u64 local_alloc_top, local_alloc_bottom;
1930 prom_debug("starting prom_initialize_tce_table\n");
1932 /* Cache current top of allocs so we reserve a single block */
1933 local_alloc_top = alloc_top_high;
1934 local_alloc_bottom = local_alloc_top;
1936 /* Search all nodes looking for PHBs. */
1937 for (node = 0; prom_next_node(&node); ) {
1941 prom_getprop(node, "compatible",
1942 compatible, sizeof(compatible));
1943 prom_getprop(node, "device_type", type, sizeof(type));
1944 prom_getprop(node, "model", model, sizeof(model));
1946 if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
1949 /* Keep the old logic intact to avoid regression. */
1950 if (compatible[0] != 0) {
1951 if ((prom_strstr(compatible, "python") == NULL) &&
1952 (prom_strstr(compatible, "Speedwagon") == NULL) &&
1953 (prom_strstr(compatible, "Winnipeg") == NULL))
1955 } else if (model[0] != 0) {
1956 if ((prom_strstr(model, "ython") == NULL) &&
1957 (prom_strstr(model, "peedwagon") == NULL) &&
1958 (prom_strstr(model, "innipeg") == NULL))
1962 if (prom_getprop(node, "tce-table-minalign", &minalign,
1963 sizeof(minalign)) == PROM_ERROR)
1965 if (prom_getprop(node, "tce-table-minsize", &minsize,
1966 sizeof(minsize)) == PROM_ERROR)
1967 minsize = 4UL << 20;
1970 * Even though we read what OF wants, we just set the table
1971 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
1972 * By doing this, we avoid the pitfalls of trying to DMA to
1973 * MMIO space and the DMA alias hole.
1975 minsize = 4UL << 20;
1977 /* Align to the greater of the align or size */
1978 align = max(minalign, minsize);
1979 base = alloc_down(minsize, align, 1);
1981 prom_panic("ERROR, cannot find space for TCE table.\n");
1982 if (base < local_alloc_bottom)
1983 local_alloc_bottom = base;
1985 /* It seems OF doesn't null-terminate the path :-( */
1986 memset(path, 0, sizeof(prom_scratch));
1987 /* Call OF to setup the TCE hardware */
1988 if (call_prom("package-to-path", 3, 1, node,
1989 path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
1990 prom_printf("package-to-path failed\n");
1993 /* Save away the TCE table attributes for later use. */
1994 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1995 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1997 prom_debug("TCE table: %s\n", path);
1998 prom_debug("\tnode = 0x%x\n", node);
1999 prom_debug("\tbase = 0x%llx\n", base);
2000 prom_debug("\tsize = 0x%x\n", minsize);
2002 /* Initialize the table to have a one-to-one mapping
2003 * over the allocated size.
2005 tce_entryp = (u64 *)base;
2006 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2007 tce_entry = (i << PAGE_SHIFT);
2009 *tce_entryp = tce_entry;
2012 prom_printf("opening PHB %s", path);
2013 phb_node = call_prom("open", 1, 1, path);
2015 prom_printf("... failed\n");
2017 prom_printf("... done\n");
2019 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2020 phb_node, -1, minsize,
2021 (u32) base, (u32) (base >> 32));
2022 call_prom("close", 1, 0, phb_node);
2025 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2027 /* These are only really needed if there is a memory limit in
2028 * effect, but we don't know so export them always. */
2029 prom_tce_alloc_start = local_alloc_bottom;
2030 prom_tce_alloc_end = local_alloc_top;
2032 /* Flag the first invalid entry */
2033 prom_debug("ending prom_initialize_tce_table\n");
2035 #endif /* __BIG_ENDIAN__ */
2036 #endif /* CONFIG_PPC64 */
2039 * With CHRP SMP we need to use the OF to start the other processors.
2040 * We can't wait until smp_boot_cpus (the OF is trashed by then)
2041 * so we have to put the processors into a holding pattern controlled
2042 * by the kernel (not OF) before we destroy the OF.
2044 * This uses a chunk of low memory, puts some holding pattern
2045 * code there and sends the other processors off to there until
2046 * smp_boot_cpus tells them to do something. The holding pattern
2047 * checks that address until its cpu # is there, when it is that
2048 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
2049 * of setting those values.
2051 * We also use physical address 0x4 here to tell when a cpu
2052 * is in its holding pattern code.
2057 * We want to reference the copy of __secondary_hold_* in the
2058 * 0 - 0x100 address range
2060 #define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
2062 static void __init prom_hold_cpus(void)
2067 unsigned long *spinloop
2068 = (void *) LOW_ADDR(__secondary_hold_spinloop);
2069 unsigned long *acknowledge
2070 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
2071 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2074 * On pseries, if RTAS supports "query-cpu-stopped-state",
2075 * we skip this stage, the CPUs will be started by the
2076 * kernel using RTAS.
2078 if ((of_platform == PLATFORM_PSERIES ||
2079 of_platform == PLATFORM_PSERIES_LPAR) &&
2080 rtas_has_query_cpu_stopped) {
2081 prom_printf("prom_hold_cpus: skipped\n");
2085 prom_debug("prom_hold_cpus: start...\n");
2086 prom_debug(" 1) spinloop = 0x%lx\n", (unsigned long)spinloop);
2087 prom_debug(" 1) *spinloop = 0x%lx\n", *spinloop);
2088 prom_debug(" 1) acknowledge = 0x%lx\n",
2089 (unsigned long)acknowledge);
2090 prom_debug(" 1) *acknowledge = 0x%lx\n", *acknowledge);
2091 prom_debug(" 1) secondary_hold = 0x%lx\n", secondary_hold);
2093 /* Set the common spinloop variable, so all of the secondary cpus
2094 * will block when they are awakened from their OF spinloop.
2095 * This must occur for both SMP and non SMP kernels, since OF will
2096 * be trashed when we move the kernel.
2101 for (node = 0; prom_next_node(&node); ) {
2102 unsigned int cpu_no;
2106 prom_getprop(node, "device_type", type, sizeof(type));
2107 if (prom_strcmp(type, "cpu") != 0)
2110 /* Skip non-configured cpus. */
2111 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2112 if (prom_strcmp(type, "okay") != 0)
2115 reg = cpu_to_be32(-1); /* make sparse happy */
2116 prom_getprop(node, "reg", ®, sizeof(reg));
2117 cpu_no = be32_to_cpu(reg);
2119 prom_debug("cpu hw idx = %u\n", cpu_no);
2121 /* Init the acknowledge var which will be reset by
2122 * the secondary cpu when it awakens from its OF
2125 *acknowledge = (unsigned long)-1;
2127 if (cpu_no != prom.cpu) {
2128 /* Primary Thread of non-boot cpu or any thread */
2129 prom_printf("starting cpu hw idx %u... ", cpu_no);
2130 call_prom("start-cpu", 3, 0, node,
2131 secondary_hold, cpu_no);
2133 for (i = 0; (i < 100000000) &&
2134 (*acknowledge == ((unsigned long)-1)); i++ )
2137 if (*acknowledge == cpu_no)
2138 prom_printf("done\n");
2140 prom_printf("failed: %lx\n", *acknowledge);
2144 prom_printf("boot cpu hw idx %u\n", cpu_no);
2145 #endif /* CONFIG_SMP */
2148 prom_debug("prom_hold_cpus: end...\n");
2152 static void __init prom_init_client_services(unsigned long pp)
2154 /* Get a handle to the prom entry point before anything else */
2157 /* get a handle for the stdout device */
2158 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2159 if (!PHANDLE_VALID(prom.chosen))
2160 prom_panic("cannot find chosen"); /* msg won't be printed :( */
2162 /* get device tree root */
2163 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2164 if (!PHANDLE_VALID(prom.root))
2165 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2172 * For really old powermacs, we need to map things we claim.
2173 * For that, we need the ihandle of the mmu.
2174 * Also, on the longtrail, we need to work around other bugs.
2176 static void __init prom_find_mmu(void)
2181 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2182 if (!PHANDLE_VALID(oprom))
2184 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2186 version[sizeof(version) - 1] = 0;
2187 /* XXX might need to add other versions here */
2188 if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2189 of_workarounds = OF_WA_CLAIM;
2190 else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2191 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2192 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2195 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2196 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2197 sizeof(prom.mmumap));
2198 prom.mmumap = be32_to_cpu(prom.mmumap);
2199 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2200 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
2203 #define prom_find_mmu()
2206 static void __init prom_init_stdout(void)
2208 char *path = of_stdout_device;
2210 phandle stdout_node;
2213 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2214 prom_panic("cannot find stdout");
2216 prom.stdout = be32_to_cpu(val);
2218 /* Get the full OF pathname of the stdout device */
2219 memset(path, 0, 256);
2220 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2221 prom_printf("OF stdout device is: %s\n", of_stdout_device);
2222 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2223 path, prom_strlen(path) + 1);
2225 /* instance-to-package fails on PA-Semi */
2226 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2227 if (stdout_node != PROM_ERROR) {
2228 val = cpu_to_be32(stdout_node);
2230 /* If it's a display, note it */
2231 memset(type, 0, sizeof(type));
2232 prom_getprop(stdout_node, "device_type", type, sizeof(type));
2233 if (prom_strcmp(type, "display") == 0)
2234 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2238 static int __init prom_find_machine_type(void)
2247 /* Look for a PowerMac or a Cell */
2248 len = prom_getprop(prom.root, "compatible",
2249 compat, sizeof(compat)-1);
2253 char *p = &compat[i];
2254 int sl = prom_strlen(p);
2257 if (prom_strstr(p, "Power Macintosh") ||
2258 prom_strstr(p, "MacRISC"))
2259 return PLATFORM_POWERMAC;
2261 /* We must make sure we don't detect the IBM Cell
2262 * blades as pSeries due to some firmware issues,
2265 if (prom_strstr(p, "IBM,CBEA") ||
2266 prom_strstr(p, "IBM,CPBW-1.0"))
2267 return PLATFORM_GENERIC;
2268 #endif /* CONFIG_PPC64 */
2273 /* Try to figure out if it's an IBM pSeries or any other
2274 * PAPR compliant platform. We assume it is if :
2275 * - /device_type is "chrp" (please, do NOT use that for future
2279 len = prom_getprop(prom.root, "device_type",
2280 compat, sizeof(compat)-1);
2282 return PLATFORM_GENERIC;
2283 if (prom_strcmp(compat, "chrp"))
2284 return PLATFORM_GENERIC;
2286 /* Default to pSeries. We need to know if we are running LPAR */
2287 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2288 if (!PHANDLE_VALID(rtas))
2289 return PLATFORM_GENERIC;
2290 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2291 if (x != PROM_ERROR) {
2292 prom_debug("Hypertas detected, assuming LPAR !\n");
2293 return PLATFORM_PSERIES_LPAR;
2295 return PLATFORM_PSERIES;
2297 return PLATFORM_GENERIC;
2301 static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2303 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2307 * If we have a display that we don't know how to drive,
2308 * we will want to try to execute OF's open method for it
2309 * later. However, OF will probably fall over if we do that
2310 * we've taken over the MMU.
2311 * So we check whether we will need to open the display,
2312 * and if so, open it now.
2314 static void __init prom_check_displays(void)
2316 char type[16], *path;
2321 static const unsigned char default_colors[] __initconst = {
2339 const unsigned char *clut;
2341 prom_debug("Looking for displays\n");
2342 for (node = 0; prom_next_node(&node); ) {
2343 memset(type, 0, sizeof(type));
2344 prom_getprop(node, "device_type", type, sizeof(type));
2345 if (prom_strcmp(type, "display") != 0)
2348 /* It seems OF doesn't null-terminate the path :-( */
2349 path = prom_scratch;
2350 memset(path, 0, sizeof(prom_scratch));
2353 * leave some room at the end of the path for appending extra
2356 if (call_prom("package-to-path", 3, 1, node, path,
2357 sizeof(prom_scratch) - 10) == PROM_ERROR)
2359 prom_printf("found display : %s, opening... ", path);
2361 ih = call_prom("open", 1, 1, path);
2363 prom_printf("failed\n");
2368 prom_printf("done\n");
2369 prom_setprop(node, path, "linux,opened", NULL, 0);
2371 /* Setup a usable color table when the appropriate
2372 * method is available. Should update this to set-colors */
2373 clut = default_colors;
2374 for (i = 0; i < 16; i++, clut += 3)
2375 if (prom_set_color(ih, i, clut[0], clut[1],
2379 #ifdef CONFIG_LOGO_LINUX_CLUT224
2380 clut = PTRRELOC(logo_linux_clut224.clut);
2381 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2382 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2385 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
2387 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2388 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2390 u32 width, height, pitch, addr;
2392 prom_printf("Setting btext !\n");
2393 prom_getprop(node, "width", &width, 4);
2394 prom_getprop(node, "height", &height, 4);
2395 prom_getprop(node, "linebytes", &pitch, 4);
2396 prom_getprop(node, "address", &addr, 4);
2397 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2398 width, height, pitch, addr);
2399 btext_setup_display(width, height, 8, pitch, addr);
2400 btext_prepare_BAT();
2402 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2407 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2408 static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2409 unsigned long needed, unsigned long align)
2413 *mem_start = _ALIGN(*mem_start, align);
2414 while ((*mem_start + needed) > *mem_end) {
2415 unsigned long room, chunk;
2417 prom_debug("Chunk exhausted, claiming more at %lx...\n",
2419 room = alloc_top - alloc_bottom;
2420 if (room > DEVTREE_CHUNK_SIZE)
2421 room = DEVTREE_CHUNK_SIZE;
2422 if (room < PAGE_SIZE)
2423 prom_panic("No memory for flatten_device_tree "
2425 chunk = alloc_up(room, 0);
2427 prom_panic("No memory for flatten_device_tree "
2428 "(claim failed)\n");
2429 *mem_end = chunk + room;
2432 ret = (void *)*mem_start;
2433 *mem_start += needed;
2438 #define dt_push_token(token, mem_start, mem_end) do { \
2439 void *room = make_room(mem_start, mem_end, 4, 4); \
2440 *(__be32 *)room = cpu_to_be32(token); \
2443 static unsigned long __init dt_find_string(char *str)
2447 s = os = (char *)dt_string_start;
2449 while (s < (char *)dt_string_end) {
2450 if (prom_strcmp(s, str) == 0)
2452 s += prom_strlen(s) + 1;
2458 * The Open Firmware 1275 specification states properties must be 31 bytes or
2459 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2461 #define MAX_PROPERTY_NAME 64
2463 static void __init scan_dt_build_strings(phandle node,
2464 unsigned long *mem_start,
2465 unsigned long *mem_end)
2467 char *prev_name, *namep, *sstart;
2471 sstart = (char *)dt_string_start;
2473 /* get and store all property names */
2476 /* 64 is max len of name including nul. */
2477 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2478 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2479 /* No more nodes: unwind alloc */
2480 *mem_start = (unsigned long)namep;
2485 if (prom_strcmp(namep, "name") == 0) {
2486 *mem_start = (unsigned long)namep;
2490 /* get/create string entry */
2491 soff = dt_find_string(namep);
2493 *mem_start = (unsigned long)namep;
2494 namep = sstart + soff;
2496 /* Trim off some if we can */
2497 *mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2498 dt_string_end = *mem_start;
2503 /* do all our children */
2504 child = call_prom("child", 1, 1, node);
2505 while (child != 0) {
2506 scan_dt_build_strings(child, mem_start, mem_end);
2507 child = call_prom("peer", 1, 1, child);
2511 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2512 unsigned long *mem_end)
2515 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2517 unsigned char *valp;
2518 static char pname[MAX_PROPERTY_NAME] __prombss;
2519 int l, room, has_phandle = 0;
2521 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2523 /* get the node's full name */
2524 namep = (char *)*mem_start;
2525 room = *mem_end - *mem_start;
2528 l = call_prom("package-to-path", 3, 1, node, namep, room);
2530 /* Didn't fit? Get more room. */
2532 if (l >= *mem_end - *mem_start)
2533 namep = make_room(mem_start, mem_end, l+1, 1);
2534 call_prom("package-to-path", 3, 1, node, namep, l);
2538 /* Fixup an Apple bug where they have bogus \0 chars in the
2539 * middle of the path in some properties, and extract
2540 * the unit name (everything after the last '/').
2542 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2549 *mem_start = _ALIGN((unsigned long)lp + 1, 4);
2552 /* get it again for debugging */
2553 path = prom_scratch;
2554 memset(path, 0, sizeof(prom_scratch));
2555 call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2557 /* get and store all properties */
2559 sstart = (char *)dt_string_start;
2561 if (call_prom("nextprop", 3, 1, node, prev_name,
2566 if (prom_strcmp(pname, "name") == 0) {
2571 /* find string offset */
2572 soff = dt_find_string(pname);
2574 prom_printf("WARNING: Can't find string index for"
2575 " <%s>, node %s\n", pname, path);
2578 prev_name = sstart + soff;
2581 l = call_prom("getproplen", 2, 1, node, pname);
2584 if (l == PROM_ERROR)
2587 /* push property head */
2588 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2589 dt_push_token(l, mem_start, mem_end);
2590 dt_push_token(soff, mem_start, mem_end);
2592 /* push property content */
2593 valp = make_room(mem_start, mem_end, l, 4);
2594 call_prom("getprop", 4, 1, node, pname, valp, l);
2595 *mem_start = _ALIGN(*mem_start, 4);
2597 if (!prom_strcmp(pname, "phandle"))
2601 /* Add a "phandle" property if none already exist */
2603 soff = dt_find_string("phandle");
2605 prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2607 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2608 dt_push_token(4, mem_start, mem_end);
2609 dt_push_token(soff, mem_start, mem_end);
2610 valp = make_room(mem_start, mem_end, 4, 4);
2611 *(__be32 *)valp = cpu_to_be32(node);
2615 /* do all our children */
2616 child = call_prom("child", 1, 1, node);
2617 while (child != 0) {
2618 scan_dt_build_struct(child, mem_start, mem_end);
2619 child = call_prom("peer", 1, 1, child);
2622 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2625 static void __init flatten_device_tree(void)
2628 unsigned long mem_start, mem_end, room;
2629 struct boot_param_header *hdr;
2634 * Check how much room we have between alloc top & bottom (+/- a
2635 * few pages), crop to 1MB, as this is our "chunk" size
2637 room = alloc_top - alloc_bottom - 0x4000;
2638 if (room > DEVTREE_CHUNK_SIZE)
2639 room = DEVTREE_CHUNK_SIZE;
2640 prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2642 /* Now try to claim that */
2643 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2645 prom_panic("Can't allocate initial device-tree chunk\n");
2646 mem_end = mem_start + room;
2648 /* Get root of tree */
2649 root = call_prom("peer", 1, 1, (phandle)0);
2650 if (root == (phandle)0)
2651 prom_panic ("couldn't get device tree root\n");
2653 /* Build header and make room for mem rsv map */
2654 mem_start = _ALIGN(mem_start, 4);
2655 hdr = make_room(&mem_start, &mem_end,
2656 sizeof(struct boot_param_header), 4);
2657 dt_header_start = (unsigned long)hdr;
2658 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2660 /* Start of strings */
2661 mem_start = PAGE_ALIGN(mem_start);
2662 dt_string_start = mem_start;
2663 mem_start += 4; /* hole */
2665 /* Add "phandle" in there, we'll need it */
2666 namep = make_room(&mem_start, &mem_end, 16, 1);
2667 prom_strcpy(namep, "phandle");
2668 mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2670 /* Build string array */
2671 prom_printf("Building dt strings...\n");
2672 scan_dt_build_strings(root, &mem_start, &mem_end);
2673 dt_string_end = mem_start;
2675 /* Build structure */
2676 mem_start = PAGE_ALIGN(mem_start);
2677 dt_struct_start = mem_start;
2678 prom_printf("Building dt structure...\n");
2679 scan_dt_build_struct(root, &mem_start, &mem_end);
2680 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2681 dt_struct_end = PAGE_ALIGN(mem_start);
2684 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2685 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2686 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2687 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2688 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2689 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2690 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2691 hdr->version = cpu_to_be32(OF_DT_VERSION);
2692 /* Version 16 is not backward compatible */
2693 hdr->last_comp_version = cpu_to_be32(0x10);
2695 /* Copy the reserve map in */
2696 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2701 prom_printf("reserved memory map:\n");
2702 for (i = 0; i < mem_reserve_cnt; i++)
2703 prom_printf(" %llx - %llx\n",
2704 be64_to_cpu(mem_reserve_map[i].base),
2705 be64_to_cpu(mem_reserve_map[i].size));
2708 /* Bump mem_reserve_cnt to cause further reservations to fail
2709 * since it's too late.
2711 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2713 prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2714 dt_string_start, dt_string_end);
2715 prom_printf("Device tree struct 0x%lx -> 0x%lx\n",
2716 dt_struct_start, dt_struct_end);
2719 #ifdef CONFIG_PPC_MAPLE
2720 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2721 * The values are bad, and it doesn't even have the right number of cells. */
2722 static void __init fixup_device_tree_maple(void)
2725 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2729 name = "/ht@0/isa@4";
2730 isa = call_prom("finddevice", 1, 1, ADDR(name));
2731 if (!PHANDLE_VALID(isa)) {
2732 name = "/ht@0/isa@6";
2733 isa = call_prom("finddevice", 1, 1, ADDR(name));
2734 rloc = 0x01003000; /* IO space; PCI device = 6 */
2736 if (!PHANDLE_VALID(isa))
2739 if (prom_getproplen(isa, "ranges") != 12)
2741 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2745 if (isa_ranges[0] != 0x1 ||
2746 isa_ranges[1] != 0xf4000000 ||
2747 isa_ranges[2] != 0x00010000)
2750 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2752 isa_ranges[0] = 0x1;
2753 isa_ranges[1] = 0x0;
2754 isa_ranges[2] = rloc;
2755 isa_ranges[3] = 0x0;
2756 isa_ranges[4] = 0x0;
2757 isa_ranges[5] = 0x00010000;
2758 prom_setprop(isa, name, "ranges",
2759 isa_ranges, sizeof(isa_ranges));
2762 #define CPC925_MC_START 0xf8000000
2763 #define CPC925_MC_LENGTH 0x1000000
2764 /* The values for memory-controller don't have right number of cells */
2765 static void __init fixup_device_tree_maple_memory_controller(void)
2769 char *name = "/hostbridge@f8000000";
2772 mc = call_prom("finddevice", 1, 1, ADDR(name));
2773 if (!PHANDLE_VALID(mc))
2776 if (prom_getproplen(mc, "reg") != 8)
2779 prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2780 prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2781 if ((ac != 2) || (sc != 2))
2784 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2787 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2790 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2793 mc_reg[1] = CPC925_MC_START;
2795 mc_reg[3] = CPC925_MC_LENGTH;
2796 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2799 #define fixup_device_tree_maple()
2800 #define fixup_device_tree_maple_memory_controller()
2803 #ifdef CONFIG_PPC_CHRP
2805 * Pegasos and BriQ lacks the "ranges" property in the isa node
2806 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2807 * Pegasos has the IDE configured in legacy mode, but advertised as native
2809 static void __init fixup_device_tree_chrp(void)
2813 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2817 name = "/pci@80000000/isa@c";
2818 ph = call_prom("finddevice", 1, 1, ADDR(name));
2819 if (!PHANDLE_VALID(ph)) {
2820 name = "/pci@ff500000/isa@6";
2821 ph = call_prom("finddevice", 1, 1, ADDR(name));
2822 rloc = 0x01003000; /* IO space; PCI device = 6 */
2824 if (PHANDLE_VALID(ph)) {
2825 rc = prom_getproplen(ph, "ranges");
2826 if (rc == 0 || rc == PROM_ERROR) {
2827 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2834 prop[5] = 0x00010000;
2835 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2839 name = "/pci@80000000/ide@C,1";
2840 ph = call_prom("finddevice", 1, 1, ADDR(name));
2841 if (PHANDLE_VALID(ph)) {
2842 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2845 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2846 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2847 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2848 if (rc == sizeof(u32)) {
2850 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2855 #define fixup_device_tree_chrp()
2858 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2859 static void __init fixup_device_tree_pmac(void)
2861 phandle u3, i2c, mpic;
2866 /* Some G5s have a missing interrupt definition, fix it up here */
2867 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2868 if (!PHANDLE_VALID(u3))
2870 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2871 if (!PHANDLE_VALID(i2c))
2873 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2874 if (!PHANDLE_VALID(mpic))
2877 /* check if proper rev of u3 */
2878 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2881 if (u3_rev < 0x35 || u3_rev > 0x39)
2883 /* does it need fixup ? */
2884 if (prom_getproplen(i2c, "interrupts") > 0)
2887 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2889 /* interrupt on this revision of u3 is number 0 and level */
2892 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2893 &interrupts, sizeof(interrupts));
2895 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2896 &parent, sizeof(parent));
2899 #define fixup_device_tree_pmac()
2902 #ifdef CONFIG_PPC_EFIKA
2904 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2905 * to talk to the phy. If the phy-handle property is missing, then this
2906 * function is called to add the appropriate nodes and link it to the
2909 static void __init fixup_device_tree_efika_add_phy(void)
2915 /* Check if /builtin/ethernet exists - bail if it doesn't */
2916 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2917 if (!PHANDLE_VALID(node))
2920 /* Check if the phy-handle property exists - bail if it does */
2921 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2926 * At this point the ethernet device doesn't have a phy described.
2927 * Now we need to add the missing phy node and linkage
2930 /* Check for an MDIO bus node - if missing then create one */
2931 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2932 if (!PHANDLE_VALID(node)) {
2933 prom_printf("Adding Ethernet MDIO node\n");
2934 call_prom("interpret", 1, 1,
2935 " s\" /builtin\" find-device"
2937 " 1 encode-int s\" #address-cells\" property"
2938 " 0 encode-int s\" #size-cells\" property"
2939 " s\" mdio\" device-name"
2940 " s\" fsl,mpc5200b-mdio\" encode-string"
2941 " s\" compatible\" property"
2942 " 0xf0003000 0x400 reg"
2944 " 0x5 encode-int encode+"
2945 " 0x3 encode-int encode+"
2946 " s\" interrupts\" property"
2950 /* Check for a PHY device node - if missing then create one and
2951 * give it's phandle to the ethernet node */
2952 node = call_prom("finddevice", 1, 1,
2953 ADDR("/builtin/mdio/ethernet-phy"));
2954 if (!PHANDLE_VALID(node)) {
2955 prom_printf("Adding Ethernet PHY node\n");
2956 call_prom("interpret", 1, 1,
2957 " s\" /builtin/mdio\" find-device"
2959 " s\" ethernet-phy\" device-name"
2960 " 0x10 encode-int s\" reg\" property"
2964 " s\" /builtin/ethernet\" find-device"
2966 " s\" phy-handle\" property"
2971 static void __init fixup_device_tree_efika(void)
2973 int sound_irq[3] = { 2, 2, 0 };
2974 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2975 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2976 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2977 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2982 /* Check if we're really running on a EFIKA */
2983 node = call_prom("finddevice", 1, 1, ADDR("/"));
2984 if (!PHANDLE_VALID(node))
2987 rv = prom_getprop(node, "model", prop, sizeof(prop));
2988 if (rv == PROM_ERROR)
2990 if (prom_strcmp(prop, "EFIKA5K2"))
2993 prom_printf("Applying EFIKA device tree fixups\n");
2995 /* Claiming to be 'chrp' is death */
2996 node = call_prom("finddevice", 1, 1, ADDR("/"));
2997 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2998 if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
2999 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3001 /* CODEGEN,description is exposed in /proc/cpuinfo so
3003 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3004 if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3005 prom_setprop(node, "/", "CODEGEN,description",
3006 "Efika 5200B PowerPC System",
3007 sizeof("Efika 5200B PowerPC System"));
3009 /* Fixup bestcomm interrupts property */
3010 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3011 if (PHANDLE_VALID(node)) {
3012 len = prom_getproplen(node, "interrupts");
3014 prom_printf("Fixing bestcomm interrupts property\n");
3015 prom_setprop(node, "/builtin/bestcom", "interrupts",
3016 bcomm_irq, sizeof(bcomm_irq));
3020 /* Fixup sound interrupts property */
3021 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3022 if (PHANDLE_VALID(node)) {
3023 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3024 if (rv == PROM_ERROR) {
3025 prom_printf("Adding sound interrupts property\n");
3026 prom_setprop(node, "/builtin/sound", "interrupts",
3027 sound_irq, sizeof(sound_irq));
3031 /* Make sure ethernet phy-handle property exists */
3032 fixup_device_tree_efika_add_phy();
3035 #define fixup_device_tree_efika()
3038 #ifdef CONFIG_PPC_PASEMI_NEMO
3040 * CFE supplied on Nemo is broken in several ways, biggest
3041 * problem is that it reassigns ISA interrupts to unused mpic ints.
3042 * Add an interrupt-controller property for the io-bridge to use
3043 * and correct the ints so we can attach them to an irq_domain
3045 static void __init fixup_device_tree_pasemi(void)
3047 u32 interrupts[2], parent, rval, val = 0;
3048 char *name, *pci_name;
3051 /* Find the root pci node */
3052 name = "/pxp@0,e0000000";
3053 iob = call_prom("finddevice", 1, 1, ADDR(name));
3054 if (!PHANDLE_VALID(iob))
3057 /* check if interrupt-controller node set yet */
3058 if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3061 prom_printf("adding interrupt-controller property for SB600...\n");
3063 prom_setprop(iob, name, "interrupt-controller", &val, 0);
3065 pci_name = "/pxp@0,e0000000/pci@11";
3066 node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3069 for( ; prom_next_node(&node); ) {
3070 /* scan each node for one with an interrupt */
3071 if (!PHANDLE_VALID(node))
3074 rval = prom_getproplen(node, "interrupts");
3075 if (rval == 0 || rval == PROM_ERROR)
3078 prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3079 if ((interrupts[0] < 212) || (interrupts[0] > 222))
3082 /* found a node, update both interrupts and interrupt-parent */
3083 if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3084 interrupts[0] -= 203;
3085 if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3086 interrupts[0] -= 213;
3087 if (interrupts[0] == 221)
3089 if (interrupts[0] == 222)
3092 prom_setprop(node, pci_name, "interrupts", interrupts,
3093 sizeof(interrupts));
3094 prom_setprop(node, pci_name, "interrupt-parent", &parent,
3099 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3100 * so that generic isa-bridge code can add the SB600 and its on-board
3103 name = "/pxp@0,e0000000/io-bridge@0";
3104 iob = call_prom("finddevice", 1, 1, ADDR(name));
3105 if (!PHANDLE_VALID(iob))
3108 /* device_type is already set, just change it. */
3110 prom_printf("Changing device_type of SB600 node...\n");
3112 prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3114 #else /* !CONFIG_PPC_PASEMI_NEMO */
3115 static inline void fixup_device_tree_pasemi(void) { }
3118 static void __init fixup_device_tree(void)
3120 fixup_device_tree_maple();
3121 fixup_device_tree_maple_memory_controller();
3122 fixup_device_tree_chrp();
3123 fixup_device_tree_pmac();
3124 fixup_device_tree_efika();
3125 fixup_device_tree_pasemi();
3128 static void __init prom_find_boot_cpu(void)
3135 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3137 prom_cpu = be32_to_cpu(rval);
3139 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3141 if (!PHANDLE_VALID(cpu_pkg))
3144 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3145 prom.cpu = be32_to_cpu(rval);
3147 prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3150 static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3152 #ifdef CONFIG_BLK_DEV_INITRD
3153 if (r3 && r4 && r4 != 0xdeadbeef) {
3156 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3157 prom_initrd_end = prom_initrd_start + r4;
3159 val = cpu_to_be64(prom_initrd_start);
3160 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3162 val = cpu_to_be64(prom_initrd_end);
3163 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3166 reserve_mem(prom_initrd_start,
3167 prom_initrd_end - prom_initrd_start);
3169 prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3170 prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3172 #endif /* CONFIG_BLK_DEV_INITRD */
3176 #ifdef CONFIG_RELOCATABLE
3177 static void reloc_toc(void)
3181 static void unreloc_toc(void)
3185 static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
3188 unsigned long *toc_entry;
3190 /* Get the start of the TOC by using r2 directly. */
3191 asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
3193 for (i = 0; i < nr_entries; i++) {
3194 *toc_entry = *toc_entry + offset;
3199 static void reloc_toc(void)
3201 unsigned long offset = reloc_offset();
3202 unsigned long nr_entries =
3203 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3205 __reloc_toc(offset, nr_entries);
3210 static void unreloc_toc(void)
3212 unsigned long offset = reloc_offset();
3213 unsigned long nr_entries =
3214 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3218 __reloc_toc(-offset, nr_entries);
3223 #ifdef CONFIG_PPC_SVM
3225 * Perform the Enter Secure Mode ultracall.
3227 static int enter_secure_mode(unsigned long kbase, unsigned long fdt)
3229 register unsigned long r3 asm("r3") = UV_ESM;
3230 register unsigned long r4 asm("r4") = kbase;
3231 register unsigned long r5 asm("r5") = fdt;
3233 asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3239 * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3241 static void setup_secure_guest(unsigned long kbase, unsigned long fdt)
3245 if (!prom_svm_enable)
3248 /* Switch to secure mode. */
3249 prom_printf("Switching to secure mode.\n");
3252 * The ultravisor will do an integrity check of the kernel image but we
3253 * relocated it so the check will fail. Restore the original image by
3254 * relocating it back to the kernel virtual base address.
3256 if (IS_ENABLED(CONFIG_RELOCATABLE))
3257 relocate(KERNELBASE);
3259 ret = enter_secure_mode(kbase, fdt);
3261 /* Relocate the kernel again. */
3262 if (IS_ENABLED(CONFIG_RELOCATABLE))
3265 if (ret != U_SUCCESS) {
3266 prom_printf("Returned %d from switching to secure mode.\n", ret);
3267 prom_rtas_os_term("Switch to secure mode failed.\n");
3271 static void setup_secure_guest(unsigned long kbase, unsigned long fdt)
3274 #endif /* CONFIG_PPC_SVM */
3277 * We enter here early on, when the Open Firmware prom is still
3278 * handling exceptions and the MMU hash table for us.
3281 unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3283 unsigned long r6, unsigned long r7,
3284 unsigned long kbase)
3289 unsigned long offset = reloc_offset();
3296 * First zero the BSS
3298 memset(&__bss_start, 0, __bss_stop - __bss_start);
3301 * Init interface to Open Firmware, get some node references,
3304 prom_init_client_services(pp);
3307 * See if this OF is old enough that we need to do explicit maps
3308 * and other workarounds
3313 * Init prom stdout device
3317 prom_printf("Preparing to boot %s", linux_banner);
3320 * Get default machine type. At this point, we do not differentiate
3321 * between pSeries SMP and pSeries LPAR
3323 of_platform = prom_find_machine_type();
3324 prom_printf("Detected machine type: %x\n", of_platform);
3326 #ifndef CONFIG_NONSTATIC_KERNEL
3327 /* Bail if this is a kdump kernel. */
3328 if (PHYSICAL_START > 0)
3329 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3333 * Check for an initrd
3335 prom_check_initrd(r3, r4);
3338 * Do early parsing of command line
3340 early_cmdline_parse();
3342 #ifdef CONFIG_PPC_PSERIES
3344 * On pSeries, inform the firmware about our capabilities
3346 if (of_platform == PLATFORM_PSERIES ||
3347 of_platform == PLATFORM_PSERIES_LPAR)
3348 prom_send_capabilities();
3352 * Copy the CPU hold code
3354 if (of_platform != PLATFORM_POWERMAC)
3355 copy_and_flush(0, kbase, 0x100, 0);
3358 * Initialize memory management within prom_init
3363 * Determine which cpu is actually running right _now_
3365 prom_find_boot_cpu();
3368 * Initialize display devices
3370 prom_check_displays();
3372 #if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3374 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3375 * that uses the allocator, we need to make sure we get the top of memory
3376 * available for us here...
3378 if (of_platform == PLATFORM_PSERIES)
3379 prom_initialize_tce_table();
3383 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3384 * have a usable RTAS implementation.
3386 if (of_platform != PLATFORM_POWERMAC)
3387 prom_instantiate_rtas();
3390 /* instantiate sml */
3391 prom_instantiate_sml();
3395 * On non-powermacs, put all CPUs in spin-loops.
3397 * PowerMacs use a different mechanism to spin CPUs
3399 * (This must be done after instanciating RTAS)
3401 if (of_platform != PLATFORM_POWERMAC)
3405 * Fill in some infos for use by the kernel later on
3407 if (prom_memory_limit) {
3408 __be64 val = cpu_to_be64(prom_memory_limit);
3409 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3414 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3417 if (prom_iommu_force_on)
3418 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3421 if (prom_tce_alloc_start) {
3422 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3423 &prom_tce_alloc_start,
3424 sizeof(prom_tce_alloc_start));
3425 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3426 &prom_tce_alloc_end,
3427 sizeof(prom_tce_alloc_end));
3432 * Fixup any known bugs in the device-tree
3434 fixup_device_tree();
3437 * Now finally create the flattened device-tree
3439 prom_printf("copying OF device tree...\n");
3440 flatten_device_tree();
3443 * in case stdin is USB and still active on IBM machines...
3444 * Unfortunately quiesce crashes on some powermacs if we have
3445 * closed stdin already (in particular the powerbook 101).
3447 if (of_platform != PLATFORM_POWERMAC)
3451 * Call OF "quiesce" method to shut down pending DMA's from
3454 prom_printf("Quiescing Open Firmware ...\n");
3455 call_prom("quiesce", 0, 0);
3458 * And finally, call the kernel passing it the flattened device
3459 * tree and NULL as r5, thus triggering the new entry point which
3460 * is common to us and kexec
3462 hdr = dt_header_start;
3464 /* Don't print anything after quiesce under OPAL, it crashes OFW */
3465 prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3466 prom_debug("->dt_header_start=0x%lx\n", hdr);
3469 reloc_got2(-offset);
3474 /* Move to secure memory if we're supposed to be secure guests. */
3475 setup_secure_guest(kbase, hdr);
3477 __start(hdr, kbase, 0, 0, 0, 0, 0);