1 /* Kernel module help for PPC64.
2 Copyright (C) 2001, 2003 Rusty Russell IBM Corporation.
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 #include <linux/module.h>
19 #include <linux/elf.h>
20 #include <linux/moduleloader.h>
21 #include <linux/err.h>
22 #include <linux/vmalloc.h>
23 #include <linux/bug.h>
24 #include <asm/module.h>
25 #include <asm/uaccess.h>
26 #include <asm/firmware.h>
27 #include <linux/sort.h>
31 /* FIXME: We don't do .init separately. To do this, we'd need to have
32 a separate r2 value in the init and core section, and stub between
35 Using a magic allocator which places modules within 32MB solves
36 this, and makes other things simpler. Anton?
41 #define DEBUGP(fmt , ...)
44 /* There's actually a third entry here, but it's unused */
45 struct ppc64_opd_entry
47 unsigned long funcaddr;
51 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into
52 the kernel itself). But on PPC64, these need to be used for every
53 jump, actually, to reset r2 (TOC+0x8000). */
54 struct ppc64_stub_entry
56 /* 28 byte jump instruction sequence (7 instructions) */
57 unsigned char jump[28];
58 unsigned char unused[4];
59 /* Data for the above code */
60 struct ppc64_opd_entry opd;
63 /* We use a stub to fix up r2 (TOC ptr) and to jump to the (external)
64 function which may be more than 24-bits away. We could simply
65 patch the new r2 value and function pointer into the stub, but it's
66 significantly shorter to put these values at the end of the stub
67 code, and patch the stub address (32-bits relative to the TOC ptr,
69 static struct ppc64_stub_entry ppc64_stub =
71 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */
72 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */
73 /* Save current r2 value in magic place on the stack. */
74 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */
75 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */
76 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */
77 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */
78 0x4e, 0x80, 0x04, 0x20 /* bctr */
81 /* Count how many different 24-bit relocations (different symbol,
83 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num)
85 unsigned int i, r_info, r_addend, _count_relocs;
87 /* FIXME: Only count external ones --RR */
91 for (i = 0; i < num; i++)
92 /* Only count 24-bit relocs, others don't need stubs */
93 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
94 (r_info != ELF64_R_SYM(rela[i].r_info) ||
95 r_addend != rela[i].r_addend)) {
97 r_info = ELF64_R_SYM(rela[i].r_info);
98 r_addend = rela[i].r_addend;
101 return _count_relocs;
104 static int relacmp(const void *_x, const void *_y)
106 const Elf64_Rela *x, *y;
108 y = (Elf64_Rela *)_x;
109 x = (Elf64_Rela *)_y;
111 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to
112 * make the comparison cheaper/faster. It won't affect the sorting or
113 * the counting algorithms' performance
115 if (x->r_info < y->r_info)
117 else if (x->r_info > y->r_info)
119 else if (x->r_addend < y->r_addend)
121 else if (x->r_addend > y->r_addend)
127 static void relaswap(void *_x, void *_y, int size)
129 uint64_t *x, *y, tmp;
135 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) {
142 /* Get size of potential trampolines required. */
143 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
144 const Elf64_Shdr *sechdrs)
146 /* One extra reloc so it's always 0-funcaddr terminated */
147 unsigned long relocs = 1;
150 /* Every relocated section... */
151 for (i = 1; i < hdr->e_shnum; i++) {
152 if (sechdrs[i].sh_type == SHT_RELA) {
153 DEBUGP("Found relocations in section %u\n", i);
154 DEBUGP("Ptr: %p. Number: %lu\n",
155 (void *)sechdrs[i].sh_addr,
156 sechdrs[i].sh_size / sizeof(Elf64_Rela));
158 /* Sort the relocation information based on a symbol and
159 * addend key. This is a stable O(n*log n) complexity
160 * alogrithm but it will reduce the complexity of
161 * count_relocs() to linear complexity O(n)
163 sort((void *)sechdrs[i].sh_addr,
164 sechdrs[i].sh_size / sizeof(Elf64_Rela),
165 sizeof(Elf64_Rela), relacmp, relaswap);
167 relocs += count_relocs((void *)sechdrs[i].sh_addr,
169 / sizeof(Elf64_Rela));
173 DEBUGP("Looks like a total of %lu stubs, max\n", relocs);
174 return relocs * sizeof(struct ppc64_stub_entry);
177 static void dedotify_versions(struct modversion_info *vers,
180 struct modversion_info *end;
182 for (end = (void *)vers + size; vers < end; vers++)
183 if (vers->name[0] == '.')
184 memmove(vers->name, vers->name+1, strlen(vers->name));
187 /* Undefined symbols which refer to .funcname, hack to funcname */
188 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
192 for (i = 1; i < numsyms; i++) {
193 if (syms[i].st_shndx == SHN_UNDEF) {
194 char *name = strtab + syms[i].st_name;
196 memmove(name, name+1, strlen(name));
201 int module_frob_arch_sections(Elf64_Ehdr *hdr,
208 /* Find .toc and .stubs sections, symtab and strtab */
209 for (i = 1; i < hdr->e_shnum; i++) {
211 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
212 me->arch.stubs_section = i;
213 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0)
214 me->arch.toc_section = i;
215 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0)
216 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
219 /* We don't handle .init for the moment: rename to _init */
220 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init")))
223 if (sechdrs[i].sh_type == SHT_SYMTAB)
224 dedotify((void *)hdr + sechdrs[i].sh_offset,
225 sechdrs[i].sh_size / sizeof(Elf64_Sym),
227 + sechdrs[sechdrs[i].sh_link].sh_offset);
230 if (!me->arch.stubs_section) {
231 printk("%s: doesn't contain .stubs.\n", me->name);
235 /* If we don't have a .toc, just use .stubs. We need to set r2
236 to some reasonable value in case the module calls out to
237 other functions via a stub, or if a function pointer escapes
238 the module by some means. */
239 if (!me->arch.toc_section)
240 me->arch.toc_section = me->arch.stubs_section;
242 /* Override the stubs size */
243 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs);
247 int apply_relocate(Elf64_Shdr *sechdrs,
249 unsigned int symindex,
253 printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n", me->name);
257 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this
258 gives the value maximum span in an instruction which uses a signed
260 static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me)
262 return sechdrs[me->arch.toc_section].sh_addr + 0x8000;
265 /* Both low and high 16 bits are added as SIGNED additions, so if low
266 16 bits has high bit set, high 16 bits must be adjusted. These
267 macros do that (stolen from binutils). */
268 #define PPC_LO(v) ((v) & 0xffff)
269 #define PPC_HI(v) (((v) >> 16) & 0xffff)
270 #define PPC_HA(v) PPC_HI ((v) + 0x8000)
272 /* Patch stub to reference function and correct r2 value. */
273 static inline int create_stub(Elf64_Shdr *sechdrs,
274 struct ppc64_stub_entry *entry,
275 struct ppc64_opd_entry *opd,
278 Elf64_Half *loc1, *loc2;
283 loc1 = (Elf64_Half *)&entry->jump[2];
284 loc2 = (Elf64_Half *)&entry->jump[6];
286 /* Stub uses address relative to r2. */
287 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
288 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
289 printk("%s: Address %p of stub out of range of %p.\n",
290 me->name, (void *)reladdr, (void *)my_r2);
293 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr);
295 *loc1 = PPC_HA(reladdr);
296 *loc2 = PPC_LO(reladdr);
297 entry->opd.funcaddr = opd->funcaddr;
298 entry->opd.r2 = opd->r2;
302 /* Create stub to jump to function described in this OPD: we need the
303 stub to set up the TOC ptr (r2) for the function. */
304 static unsigned long stub_for_addr(Elf64_Shdr *sechdrs,
305 unsigned long opdaddr,
308 struct ppc64_stub_entry *stubs;
309 struct ppc64_opd_entry *opd = (void *)opdaddr;
310 unsigned int i, num_stubs;
312 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
314 /* Find this stub, or if that fails, the next avail. entry */
315 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
316 for (i = 0; stubs[i].opd.funcaddr; i++) {
317 BUG_ON(i >= num_stubs);
319 if (stubs[i].opd.funcaddr == opd->funcaddr)
320 return (unsigned long)&stubs[i];
323 if (!create_stub(sechdrs, &stubs[i], opd, me))
326 return (unsigned long)&stubs[i];
329 /* We expect a noop next: if it is, replace it with instruction to
331 static int restore_r2(u32 *instruction, struct module *me)
333 if (*instruction != 0x60000000) {
334 printk("%s: Expect noop after relocate, got %08x\n",
335 me->name, *instruction);
338 *instruction = 0xe8410028; /* ld r2,40(r1) */
342 int apply_relocate_add(Elf64_Shdr *sechdrs,
344 unsigned int symindex,
349 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
351 unsigned long *location;
354 DEBUGP("Applying ADD relocate section %u to %u\n", relsec,
355 sechdrs[relsec].sh_info);
356 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
357 /* This is where to make the change */
358 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
360 /* This is the symbol it is referring to */
361 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
362 + ELF64_R_SYM(rela[i].r_info);
364 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n",
365 location, (long)ELF64_R_TYPE(rela[i].r_info),
366 strtab + sym->st_name, (unsigned long)sym->st_value,
367 (long)rela[i].r_addend);
369 /* `Everything is relative'. */
370 value = sym->st_value + rela[i].r_addend;
372 switch (ELF64_R_TYPE(rela[i].r_info)) {
375 *(u32 *)location = value;
380 *(unsigned long *)location = value;
384 *(unsigned long *)location = my_r2(sechdrs, me);
388 /* Subtract TOC pointer */
389 value -= my_r2(sechdrs, me);
390 if (value + 0x8000 > 0xffff) {
391 printk("%s: bad TOC16 relocation (%lu)\n",
395 *((uint16_t *) location)
396 = (*((uint16_t *) location) & ~0xffff)
400 case R_PPC64_TOC16_DS:
401 /* Subtract TOC pointer */
402 value -= my_r2(sechdrs, me);
403 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
404 printk("%s: bad TOC16_DS relocation (%lu)\n",
408 *((uint16_t *) location)
409 = (*((uint16_t *) location) & ~0xfffc)
414 /* FIXME: Handle weak symbols here --RR */
415 if (sym->st_shndx == SHN_UNDEF) {
416 /* External: go via stub */
417 value = stub_for_addr(sechdrs, value, me);
420 if (!restore_r2((u32 *)location + 1, me))
424 /* Convert value to relative */
425 value -= (unsigned long)location;
426 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
427 printk("%s: REL24 %li out of range!\n",
428 me->name, (long int)value);
432 /* Only replace bits 2 through 26 */
433 *(uint32_t *)location
434 = (*(uint32_t *)location & ~0x03fffffc)
435 | (value & 0x03fffffc);
439 /* 64 bits relative (used by features fixups) */
440 *location = value - (unsigned long)location;
444 printk("%s: Unknown ADD relocation: %lu\n",
446 (unsigned long)ELF64_R_TYPE(rela[i].r_info));