1 /* Kernel module help for PPC64. 2 Copyright (C) 2001, 2003 Rusty Russell IBM Corporation. 3 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. 8 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. 13 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 17 */ 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/firmware.h> 26 #include <asm/code-patching.h> 27 #include <linux/sort.h> 28 29 #include "setup.h" 30 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 33 them, too. 34 35 Using a magic allocator which places modules within 32MB solves 36 this, and makes other things simpler. Anton? 37 --RR. */ 38 #if 0 39 #define DEBUGP printk 40 #else 41 #define DEBUGP(fmt , ...) 42 #endif 43 44 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into 45 the kernel itself). But on PPC64, these need to be used for every 46 jump, actually, to reset r2 (TOC+0x8000). */ 47 struct ppc64_stub_entry 48 { 49 /* 28 byte jump instruction sequence (7 instructions) */ 50 unsigned char jump[28]; 51 unsigned char unused[4]; 52 /* Data for the above code */ 53 struct ppc64_opd_entry opd; 54 }; 55 56 /* We use a stub to fix up r2 (TOC ptr) and to jump to the (external) 57 function which may be more than 24-bits away. We could simply 58 patch the new r2 value and function pointer into the stub, but it's 59 significantly shorter to put these values at the end of the stub 60 code, and patch the stub address (32-bits relative to the TOC ptr, 61 r2) into the stub. */ 62 static struct ppc64_stub_entry ppc64_stub = 63 { .jump = { 64 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */ 65 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */ 66 /* Save current r2 value in magic place on the stack. */ 67 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */ 68 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */ 69 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */ 70 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */ 71 0x4e, 0x80, 0x04, 0x20 /* bctr */ 72 } }; 73 74 /* Count how many different 24-bit relocations (different symbol, 75 different addend) */ 76 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) 77 { 78 unsigned int i, r_info, r_addend, _count_relocs; 79 80 /* FIXME: Only count external ones --RR */ 81 _count_relocs = 0; 82 r_info = 0; 83 r_addend = 0; 84 for (i = 0; i < num; i++) 85 /* Only count 24-bit relocs, others don't need stubs */ 86 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 && 87 (r_info != ELF64_R_SYM(rela[i].r_info) || 88 r_addend != rela[i].r_addend)) { 89 _count_relocs++; 90 r_info = ELF64_R_SYM(rela[i].r_info); 91 r_addend = rela[i].r_addend; 92 } 93 94 return _count_relocs; 95 } 96 97 static int relacmp(const void *_x, const void *_y) 98 { 99 const Elf64_Rela *x, *y; 100 101 y = (Elf64_Rela *)_x; 102 x = (Elf64_Rela *)_y; 103 104 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to 105 * make the comparison cheaper/faster. It won't affect the sorting or 106 * the counting algorithms' performance 107 */ 108 if (x->r_info < y->r_info) 109 return -1; 110 else if (x->r_info > y->r_info) 111 return 1; 112 else if (x->r_addend < y->r_addend) 113 return -1; 114 else if (x->r_addend > y->r_addend) 115 return 1; 116 else 117 return 0; 118 } 119 120 static void relaswap(void *_x, void *_y, int size) 121 { 122 uint64_t *x, *y, tmp; 123 int i; 124 125 y = (uint64_t *)_x; 126 x = (uint64_t *)_y; 127 128 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) { 129 tmp = x[i]; 130 x[i] = y[i]; 131 y[i] = tmp; 132 } 133 } 134 135 /* Get size of potential trampolines required. */ 136 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, 137 const Elf64_Shdr *sechdrs) 138 { 139 /* One extra reloc so it's always 0-funcaddr terminated */ 140 unsigned long relocs = 1; 141 unsigned i; 142 143 /* Every relocated section... */ 144 for (i = 1; i < hdr->e_shnum; i++) { 145 if (sechdrs[i].sh_type == SHT_RELA) { 146 DEBUGP("Found relocations in section %u\n", i); 147 DEBUGP("Ptr: %p. Number: %lu\n", 148 (void *)sechdrs[i].sh_addr, 149 sechdrs[i].sh_size / sizeof(Elf64_Rela)); 150 151 /* Sort the relocation information based on a symbol and 152 * addend key. This is a stable O(n*log n) complexity 153 * alogrithm but it will reduce the complexity of 154 * count_relocs() to linear complexity O(n) 155 */ 156 sort((void *)sechdrs[i].sh_addr, 157 sechdrs[i].sh_size / sizeof(Elf64_Rela), 158 sizeof(Elf64_Rela), relacmp, relaswap); 159 160 relocs += count_relocs((void *)sechdrs[i].sh_addr, 161 sechdrs[i].sh_size 162 / sizeof(Elf64_Rela)); 163 } 164 } 165 166 DEBUGP("Looks like a total of %lu stubs, max\n", relocs); 167 return relocs * sizeof(struct ppc64_stub_entry); 168 } 169 170 static void dedotify_versions(struct modversion_info *vers, 171 unsigned long size) 172 { 173 struct modversion_info *end; 174 175 for (end = (void *)vers + size; vers < end; vers++) 176 if (vers->name[0] == '.') 177 memmove(vers->name, vers->name+1, strlen(vers->name)); 178 } 179 180 /* Undefined symbols which refer to .funcname, hack to funcname */ 181 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) 182 { 183 unsigned int i; 184 185 for (i = 1; i < numsyms; i++) { 186 if (syms[i].st_shndx == SHN_UNDEF) { 187 char *name = strtab + syms[i].st_name; 188 if (name[0] == '.') 189 memmove(name, name+1, strlen(name)); 190 } 191 } 192 } 193 194 int module_frob_arch_sections(Elf64_Ehdr *hdr, 195 Elf64_Shdr *sechdrs, 196 char *secstrings, 197 struct module *me) 198 { 199 unsigned int i; 200 201 /* Find .toc and .stubs sections, symtab and strtab */ 202 for (i = 1; i < hdr->e_shnum; i++) { 203 char *p; 204 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) 205 me->arch.stubs_section = i; 206 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) 207 me->arch.toc_section = i; 208 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) 209 dedotify_versions((void *)hdr + sechdrs[i].sh_offset, 210 sechdrs[i].sh_size); 211 212 /* We don't handle .init for the moment: rename to _init */ 213 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) 214 p[0] = '_'; 215 216 if (sechdrs[i].sh_type == SHT_SYMTAB) 217 dedotify((void *)hdr + sechdrs[i].sh_offset, 218 sechdrs[i].sh_size / sizeof(Elf64_Sym), 219 (void *)hdr 220 + sechdrs[sechdrs[i].sh_link].sh_offset); 221 } 222 223 if (!me->arch.stubs_section) { 224 printk("%s: doesn't contain .stubs.\n", me->name); 225 return -ENOEXEC; 226 } 227 228 /* If we don't have a .toc, just use .stubs. We need to set r2 229 to some reasonable value in case the module calls out to 230 other functions via a stub, or if a function pointer escapes 231 the module by some means. */ 232 if (!me->arch.toc_section) 233 me->arch.toc_section = me->arch.stubs_section; 234 235 /* Override the stubs size */ 236 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); 237 return 0; 238 } 239 240 int apply_relocate(Elf64_Shdr *sechdrs, 241 const char *strtab, 242 unsigned int symindex, 243 unsigned int relsec, 244 struct module *me) 245 { 246 printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n", me->name); 247 return -ENOEXEC; 248 } 249 250 /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this 251 gives the value maximum span in an instruction which uses a signed 252 offset) */ 253 static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me) 254 { 255 return sechdrs[me->arch.toc_section].sh_addr + 0x8000; 256 } 257 258 /* Both low and high 16 bits are added as SIGNED additions, so if low 259 16 bits has high bit set, high 16 bits must be adjusted. These 260 macros do that (stolen from binutils). */ 261 #define PPC_LO(v) ((v) & 0xffff) 262 #define PPC_HI(v) (((v) >> 16) & 0xffff) 263 #define PPC_HA(v) PPC_HI ((v) + 0x8000) 264 265 /* Patch stub to reference function and correct r2 value. */ 266 static inline int create_stub(Elf64_Shdr *sechdrs, 267 struct ppc64_stub_entry *entry, 268 struct ppc64_opd_entry *opd, 269 struct module *me) 270 { 271 Elf64_Half *loc1, *loc2; 272 long reladdr; 273 274 *entry = ppc64_stub; 275 276 loc1 = (Elf64_Half *)&entry->jump[2]; 277 loc2 = (Elf64_Half *)&entry->jump[6]; 278 279 /* Stub uses address relative to r2. */ 280 reladdr = (unsigned long)entry - my_r2(sechdrs, me); 281 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { 282 printk("%s: Address %p of stub out of range of %p.\n", 283 me->name, (void *)reladdr, (void *)my_r2); 284 return 0; 285 } 286 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr); 287 288 *loc1 = PPC_HA(reladdr); 289 *loc2 = PPC_LO(reladdr); 290 entry->opd.funcaddr = opd->funcaddr; 291 entry->opd.r2 = opd->r2; 292 return 1; 293 } 294 295 /* Create stub to jump to function described in this OPD: we need the 296 stub to set up the TOC ptr (r2) for the function. */ 297 static unsigned long stub_for_addr(Elf64_Shdr *sechdrs, 298 unsigned long opdaddr, 299 struct module *me) 300 { 301 struct ppc64_stub_entry *stubs; 302 struct ppc64_opd_entry *opd = (void *)opdaddr; 303 unsigned int i, num_stubs; 304 305 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); 306 307 /* Find this stub, or if that fails, the next avail. entry */ 308 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; 309 for (i = 0; stubs[i].opd.funcaddr; i++) { 310 BUG_ON(i >= num_stubs); 311 312 if (stubs[i].opd.funcaddr == opd->funcaddr) 313 return (unsigned long)&stubs[i]; 314 } 315 316 if (!create_stub(sechdrs, &stubs[i], opd, me)) 317 return 0; 318 319 return (unsigned long)&stubs[i]; 320 } 321 322 /* We expect a noop next: if it is, replace it with instruction to 323 restore r2. */ 324 static int restore_r2(u32 *instruction, struct module *me) 325 { 326 if (*instruction != PPC_NOP_INSTR) { 327 printk("%s: Expect noop after relocate, got %08x\n", 328 me->name, *instruction); 329 return 0; 330 } 331 *instruction = 0xe8410028; /* ld r2,40(r1) */ 332 return 1; 333 } 334 335 int apply_relocate_add(Elf64_Shdr *sechdrs, 336 const char *strtab, 337 unsigned int symindex, 338 unsigned int relsec, 339 struct module *me) 340 { 341 unsigned int i; 342 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; 343 Elf64_Sym *sym; 344 unsigned long *location; 345 unsigned long value; 346 347 DEBUGP("Applying ADD relocate section %u to %u\n", relsec, 348 sechdrs[relsec].sh_info); 349 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { 350 /* This is where to make the change */ 351 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr 352 + rela[i].r_offset; 353 /* This is the symbol it is referring to */ 354 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr 355 + ELF64_R_SYM(rela[i].r_info); 356 357 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n", 358 location, (long)ELF64_R_TYPE(rela[i].r_info), 359 strtab + sym->st_name, (unsigned long)sym->st_value, 360 (long)rela[i].r_addend); 361 362 /* `Everything is relative'. */ 363 value = sym->st_value + rela[i].r_addend; 364 365 switch (ELF64_R_TYPE(rela[i].r_info)) { 366 case R_PPC64_ADDR32: 367 /* Simply set it */ 368 *(u32 *)location = value; 369 break; 370 371 case R_PPC64_ADDR64: 372 /* Simply set it */ 373 *(unsigned long *)location = value; 374 break; 375 376 case R_PPC64_TOC: 377 *(unsigned long *)location = my_r2(sechdrs, me); 378 break; 379 380 case R_PPC64_TOC16: 381 /* Subtract TOC pointer */ 382 value -= my_r2(sechdrs, me); 383 if (value + 0x8000 > 0xffff) { 384 printk("%s: bad TOC16 relocation (%lu)\n", 385 me->name, value); 386 return -ENOEXEC; 387 } 388 *((uint16_t *) location) 389 = (*((uint16_t *) location) & ~0xffff) 390 | (value & 0xffff); 391 break; 392 393 case R_PPC64_TOC16_DS: 394 /* Subtract TOC pointer */ 395 value -= my_r2(sechdrs, me); 396 if ((value & 3) != 0 || value + 0x8000 > 0xffff) { 397 printk("%s: bad TOC16_DS relocation (%lu)\n", 398 me->name, value); 399 return -ENOEXEC; 400 } 401 *((uint16_t *) location) 402 = (*((uint16_t *) location) & ~0xfffc) 403 | (value & 0xfffc); 404 break; 405 406 case R_PPC_REL24: 407 /* FIXME: Handle weak symbols here --RR */ 408 if (sym->st_shndx == SHN_UNDEF) { 409 /* External: go via stub */ 410 value = stub_for_addr(sechdrs, value, me); 411 if (!value) 412 return -ENOENT; 413 if (!restore_r2((u32 *)location + 1, me)) 414 return -ENOEXEC; 415 } 416 417 /* Convert value to relative */ 418 value -= (unsigned long)location; 419 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ 420 printk("%s: REL24 %li out of range!\n", 421 me->name, (long int)value); 422 return -ENOEXEC; 423 } 424 425 /* Only replace bits 2 through 26 */ 426 *(uint32_t *)location 427 = (*(uint32_t *)location & ~0x03fffffc) 428 | (value & 0x03fffffc); 429 break; 430 431 case R_PPC64_REL64: 432 /* 64 bits relative (used by features fixups) */ 433 *location = value - (unsigned long)location; 434 break; 435 436 default: 437 printk("%s: Unknown ADD relocation: %lu\n", 438 me->name, 439 (unsigned long)ELF64_R_TYPE(rela[i].r_info)); 440 return -ENOEXEC; 441 } 442 } 443 444 return 0; 445 } 446