Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 <asm/module.h> | |
24 | #include <asm/uaccess.h> | |
21c4ff80 BH |
25 | #include <asm/firmware.h> |
26 | ||
27 | #include "setup.h" | |
1da177e4 LT |
28 | |
29 | /* FIXME: We don't do .init separately. To do this, we'd need to have | |
30 | a separate r2 value in the init and core section, and stub between | |
31 | them, too. | |
32 | ||
33 | Using a magic allocator which places modules within 32MB solves | |
34 | this, and makes other things simpler. Anton? | |
35 | --RR. */ | |
36 | #if 0 | |
37 | #define DEBUGP printk | |
38 | #else | |
39 | #define DEBUGP(fmt , ...) | |
40 | #endif | |
41 | ||
42 | /* There's actually a third entry here, but it's unused */ | |
43 | struct ppc64_opd_entry | |
44 | { | |
45 | unsigned long funcaddr; | |
46 | unsigned long r2; | |
47 | }; | |
48 | ||
49 | /* Like PPC32, we need little trampolines to do > 24-bit jumps (into | |
50 | the kernel itself). But on PPC64, these need to be used for every | |
51 | jump, actually, to reset r2 (TOC+0x8000). */ | |
52 | struct ppc64_stub_entry | |
53 | { | |
54 | /* 28 byte jump instruction sequence (7 instructions) */ | |
55 | unsigned char jump[28]; | |
56 | unsigned char unused[4]; | |
57 | /* Data for the above code */ | |
58 | struct ppc64_opd_entry opd; | |
59 | }; | |
60 | ||
61 | /* We use a stub to fix up r2 (TOC ptr) and to jump to the (external) | |
62 | function which may be more than 24-bits away. We could simply | |
63 | patch the new r2 value and function pointer into the stub, but it's | |
64 | significantly shorter to put these values at the end of the stub | |
65 | code, and patch the stub address (32-bits relative to the TOC ptr, | |
66 | r2) into the stub. */ | |
67 | static struct ppc64_stub_entry ppc64_stub = | |
68 | { .jump = { | |
69 | 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */ | |
70 | 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */ | |
71 | /* Save current r2 value in magic place on the stack. */ | |
72 | 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */ | |
73 | 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */ | |
74 | 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */ | |
75 | 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */ | |
76 | 0x4e, 0x80, 0x04, 0x20 /* bctr */ | |
77 | } }; | |
78 | ||
79 | /* Count how many different 24-bit relocations (different symbol, | |
80 | different addend) */ | |
81 | static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) | |
82 | { | |
83 | unsigned int i, j, ret = 0; | |
84 | ||
85 | /* FIXME: Only count external ones --RR */ | |
86 | /* Sure, this is order(n^2), but it's usually short, and not | |
87 | time critical */ | |
88 | for (i = 0; i < num; i++) { | |
89 | /* Only count 24-bit relocs, others don't need stubs */ | |
90 | if (ELF64_R_TYPE(rela[i].r_info) != R_PPC_REL24) | |
91 | continue; | |
92 | for (j = 0; j < i; j++) { | |
93 | /* If this addend appeared before, it's | |
94 | already been counted */ | |
95 | if (rela[i].r_info == rela[j].r_info | |
96 | && rela[i].r_addend == rela[j].r_addend) | |
97 | break; | |
98 | } | |
99 | if (j == i) ret++; | |
100 | } | |
101 | return ret; | |
102 | } | |
103 | ||
104 | void *module_alloc(unsigned long size) | |
105 | { | |
106 | if (size == 0) | |
107 | return NULL; | |
108 | ||
109 | return vmalloc_exec(size); | |
110 | } | |
111 | ||
112 | /* Free memory returned from module_alloc */ | |
113 | void module_free(struct module *mod, void *module_region) | |
114 | { | |
115 | vfree(module_region); | |
116 | /* FIXME: If module_region == mod->init_region, trim exception | |
117 | table entries. */ | |
118 | } | |
119 | ||
120 | /* Get size of potential trampolines required. */ | |
121 | static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, | |
122 | const Elf64_Shdr *sechdrs) | |
123 | { | |
124 | /* One extra reloc so it's always 0-funcaddr terminated */ | |
125 | unsigned long relocs = 1; | |
126 | unsigned i; | |
127 | ||
128 | /* Every relocated section... */ | |
129 | for (i = 1; i < hdr->e_shnum; i++) { | |
130 | if (sechdrs[i].sh_type == SHT_RELA) { | |
131 | DEBUGP("Found relocations in section %u\n", i); | |
132 | DEBUGP("Ptr: %p. Number: %lu\n", | |
133 | (void *)sechdrs[i].sh_addr, | |
134 | sechdrs[i].sh_size / sizeof(Elf64_Rela)); | |
135 | relocs += count_relocs((void *)sechdrs[i].sh_addr, | |
136 | sechdrs[i].sh_size | |
137 | / sizeof(Elf64_Rela)); | |
138 | } | |
139 | } | |
140 | ||
141 | DEBUGP("Looks like a total of %lu stubs, max\n", relocs); | |
142 | return relocs * sizeof(struct ppc64_stub_entry); | |
143 | } | |
144 | ||
145 | static void dedotify_versions(struct modversion_info *vers, | |
146 | unsigned long size) | |
147 | { | |
148 | struct modversion_info *end; | |
149 | ||
150 | for (end = (void *)vers + size; vers < end; vers++) | |
151 | if (vers->name[0] == '.') | |
152 | memmove(vers->name, vers->name+1, strlen(vers->name)); | |
153 | } | |
154 | ||
155 | /* Undefined symbols which refer to .funcname, hack to funcname */ | |
156 | static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) | |
157 | { | |
158 | unsigned int i; | |
159 | ||
160 | for (i = 1; i < numsyms; i++) { | |
161 | if (syms[i].st_shndx == SHN_UNDEF) { | |
162 | char *name = strtab + syms[i].st_name; | |
163 | if (name[0] == '.') | |
164 | memmove(name, name+1, strlen(name)); | |
165 | } | |
166 | } | |
167 | } | |
168 | ||
169 | int module_frob_arch_sections(Elf64_Ehdr *hdr, | |
170 | Elf64_Shdr *sechdrs, | |
171 | char *secstrings, | |
172 | struct module *me) | |
173 | { | |
174 | unsigned int i; | |
175 | ||
176 | /* Find .toc and .stubs sections, symtab and strtab */ | |
177 | for (i = 1; i < hdr->e_shnum; i++) { | |
178 | char *p; | |
179 | if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) | |
180 | me->arch.stubs_section = i; | |
181 | else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) | |
182 | me->arch.toc_section = i; | |
183 | else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) | |
184 | dedotify_versions((void *)hdr + sechdrs[i].sh_offset, | |
185 | sechdrs[i].sh_size); | |
186 | ||
187 | /* We don't handle .init for the moment: rename to _init */ | |
188 | while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) | |
189 | p[0] = '_'; | |
190 | ||
191 | if (sechdrs[i].sh_type == SHT_SYMTAB) | |
192 | dedotify((void *)hdr + sechdrs[i].sh_offset, | |
193 | sechdrs[i].sh_size / sizeof(Elf64_Sym), | |
194 | (void *)hdr | |
195 | + sechdrs[sechdrs[i].sh_link].sh_offset); | |
196 | } | |
f749edae AM |
197 | |
198 | if (!me->arch.stubs_section) { | |
199 | printk("%s: doesn't contain .stubs.\n", me->name); | |
1da177e4 LT |
200 | return -ENOEXEC; |
201 | } | |
202 | ||
f749edae AM |
203 | /* If we don't have a .toc, just use .stubs. We need to set r2 |
204 | to some reasonable value in case the module calls out to | |
205 | other functions via a stub, or if a function pointer escapes | |
206 | the module by some means. */ | |
207 | if (!me->arch.toc_section) | |
208 | me->arch.toc_section = me->arch.stubs_section; | |
209 | ||
1da177e4 LT |
210 | /* Override the stubs size */ |
211 | sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); | |
212 | return 0; | |
213 | } | |
214 | ||
215 | int apply_relocate(Elf64_Shdr *sechdrs, | |
216 | const char *strtab, | |
217 | unsigned int symindex, | |
218 | unsigned int relsec, | |
219 | struct module *me) | |
220 | { | |
221 | printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n", me->name); | |
222 | return -ENOEXEC; | |
223 | } | |
224 | ||
225 | /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this | |
226 | gives the value maximum span in an instruction which uses a signed | |
227 | offset) */ | |
228 | static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me) | |
229 | { | |
230 | return sechdrs[me->arch.toc_section].sh_addr + 0x8000; | |
231 | } | |
232 | ||
233 | /* Both low and high 16 bits are added as SIGNED additions, so if low | |
234 | 16 bits has high bit set, high 16 bits must be adjusted. These | |
235 | macros do that (stolen from binutils). */ | |
236 | #define PPC_LO(v) ((v) & 0xffff) | |
237 | #define PPC_HI(v) (((v) >> 16) & 0xffff) | |
238 | #define PPC_HA(v) PPC_HI ((v) + 0x8000) | |
239 | ||
240 | /* Patch stub to reference function and correct r2 value. */ | |
241 | static inline int create_stub(Elf64_Shdr *sechdrs, | |
242 | struct ppc64_stub_entry *entry, | |
243 | struct ppc64_opd_entry *opd, | |
244 | struct module *me) | |
245 | { | |
246 | Elf64_Half *loc1, *loc2; | |
247 | long reladdr; | |
248 | ||
249 | *entry = ppc64_stub; | |
250 | ||
251 | loc1 = (Elf64_Half *)&entry->jump[2]; | |
252 | loc2 = (Elf64_Half *)&entry->jump[6]; | |
253 | ||
254 | /* Stub uses address relative to r2. */ | |
255 | reladdr = (unsigned long)entry - my_r2(sechdrs, me); | |
256 | if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { | |
257 | printk("%s: Address %p of stub out of range of %p.\n", | |
258 | me->name, (void *)reladdr, (void *)my_r2); | |
259 | return 0; | |
260 | } | |
261 | DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr); | |
262 | ||
263 | *loc1 = PPC_HA(reladdr); | |
264 | *loc2 = PPC_LO(reladdr); | |
265 | entry->opd.funcaddr = opd->funcaddr; | |
266 | entry->opd.r2 = opd->r2; | |
267 | return 1; | |
268 | } | |
269 | ||
270 | /* Create stub to jump to function described in this OPD: we need the | |
271 | stub to set up the TOC ptr (r2) for the function. */ | |
272 | static unsigned long stub_for_addr(Elf64_Shdr *sechdrs, | |
273 | unsigned long opdaddr, | |
274 | struct module *me) | |
275 | { | |
276 | struct ppc64_stub_entry *stubs; | |
277 | struct ppc64_opd_entry *opd = (void *)opdaddr; | |
278 | unsigned int i, num_stubs; | |
279 | ||
280 | num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); | |
281 | ||
282 | /* Find this stub, or if that fails, the next avail. entry */ | |
283 | stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; | |
284 | for (i = 0; stubs[i].opd.funcaddr; i++) { | |
285 | BUG_ON(i >= num_stubs); | |
286 | ||
287 | if (stubs[i].opd.funcaddr == opd->funcaddr) | |
288 | return (unsigned long)&stubs[i]; | |
289 | } | |
290 | ||
291 | if (!create_stub(sechdrs, &stubs[i], opd, me)) | |
292 | return 0; | |
293 | ||
294 | return (unsigned long)&stubs[i]; | |
295 | } | |
296 | ||
297 | /* We expect a noop next: if it is, replace it with instruction to | |
298 | restore r2. */ | |
299 | static int restore_r2(u32 *instruction, struct module *me) | |
300 | { | |
301 | if (*instruction != 0x60000000) { | |
302 | printk("%s: Expect noop after relocate, got %08x\n", | |
303 | me->name, *instruction); | |
304 | return 0; | |
305 | } | |
306 | *instruction = 0xe8410028; /* ld r2,40(r1) */ | |
307 | return 1; | |
308 | } | |
309 | ||
310 | int apply_relocate_add(Elf64_Shdr *sechdrs, | |
311 | const char *strtab, | |
312 | unsigned int symindex, | |
313 | unsigned int relsec, | |
314 | struct module *me) | |
315 | { | |
316 | unsigned int i; | |
317 | Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; | |
318 | Elf64_Sym *sym; | |
319 | unsigned long *location; | |
320 | unsigned long value; | |
321 | ||
322 | DEBUGP("Applying ADD relocate section %u to %u\n", relsec, | |
323 | sechdrs[relsec].sh_info); | |
324 | for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { | |
325 | /* This is where to make the change */ | |
326 | location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr | |
327 | + rela[i].r_offset; | |
328 | /* This is the symbol it is referring to */ | |
329 | sym = (Elf64_Sym *)sechdrs[symindex].sh_addr | |
330 | + ELF64_R_SYM(rela[i].r_info); | |
331 | ||
332 | DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n", | |
333 | location, (long)ELF64_R_TYPE(rela[i].r_info), | |
334 | strtab + sym->st_name, (unsigned long)sym->st_value, | |
335 | (long)rela[i].r_addend); | |
336 | ||
337 | /* `Everything is relative'. */ | |
338 | value = sym->st_value + rela[i].r_addend; | |
339 | ||
340 | switch (ELF64_R_TYPE(rela[i].r_info)) { | |
341 | case R_PPC64_ADDR32: | |
342 | /* Simply set it */ | |
343 | *(u32 *)location = value; | |
344 | break; | |
345 | ||
346 | case R_PPC64_ADDR64: | |
347 | /* Simply set it */ | |
348 | *(unsigned long *)location = value; | |
349 | break; | |
350 | ||
351 | case R_PPC64_TOC: | |
352 | *(unsigned long *)location = my_r2(sechdrs, me); | |
353 | break; | |
354 | ||
9149ccfa | 355 | case R_PPC64_TOC16: |
f749edae | 356 | /* Subtract TOC pointer */ |
9149ccfa PB |
357 | value -= my_r2(sechdrs, me); |
358 | if (value + 0x8000 > 0xffff) { | |
359 | printk("%s: bad TOC16 relocation (%lu)\n", | |
360 | me->name, value); | |
361 | return -ENOEXEC; | |
362 | } | |
363 | *((uint16_t *) location) | |
364 | = (*((uint16_t *) location) & ~0xffff) | |
365 | | (value & 0xffff); | |
366 | break; | |
367 | ||
1da177e4 | 368 | case R_PPC64_TOC16_DS: |
f749edae | 369 | /* Subtract TOC pointer */ |
1da177e4 LT |
370 | value -= my_r2(sechdrs, me); |
371 | if ((value & 3) != 0 || value + 0x8000 > 0xffff) { | |
372 | printk("%s: bad TOC16_DS relocation (%lu)\n", | |
373 | me->name, value); | |
374 | return -ENOEXEC; | |
375 | } | |
376 | *((uint16_t *) location) | |
377 | = (*((uint16_t *) location) & ~0xfffc) | |
378 | | (value & 0xfffc); | |
379 | break; | |
380 | ||
381 | case R_PPC_REL24: | |
382 | /* FIXME: Handle weak symbols here --RR */ | |
383 | if (sym->st_shndx == SHN_UNDEF) { | |
384 | /* External: go via stub */ | |
385 | value = stub_for_addr(sechdrs, value, me); | |
386 | if (!value) | |
387 | return -ENOENT; | |
388 | if (!restore_r2((u32 *)location + 1, me)) | |
389 | return -ENOEXEC; | |
390 | } | |
391 | ||
392 | /* Convert value to relative */ | |
393 | value -= (unsigned long)location; | |
394 | if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ | |
395 | printk("%s: REL24 %li out of range!\n", | |
396 | me->name, (long int)value); | |
397 | return -ENOEXEC; | |
398 | } | |
399 | ||
400 | /* Only replace bits 2 through 26 */ | |
401 | *(uint32_t *)location | |
402 | = (*(uint32_t *)location & ~0x03fffffc) | |
403 | | (value & 0x03fffffc); | |
404 | break; | |
405 | ||
21c4ff80 BH |
406 | case R_PPC64_REL64: |
407 | /* 64 bits relative (used by features fixups) */ | |
408 | *location = value - (unsigned long)location; | |
409 | break; | |
410 | ||
1da177e4 LT |
411 | default: |
412 | printk("%s: Unknown ADD relocation: %lu\n", | |
413 | me->name, | |
414 | (unsigned long)ELF64_R_TYPE(rela[i].r_info)); | |
415 | return -ENOEXEC; | |
416 | } | |
417 | } | |
418 | ||
419 | return 0; | |
420 | } | |
421 | ||
422 | LIST_HEAD(module_bug_list); | |
423 | ||
21c4ff80 BH |
424 | static const Elf_Shdr *find_section(const Elf_Ehdr *hdr, |
425 | const Elf_Shdr *sechdrs, | |
426 | const char *name) | |
1da177e4 LT |
427 | { |
428 | char *secstrings; | |
429 | unsigned int i; | |
430 | ||
21c4ff80 BH |
431 | secstrings = (char *)hdr + sechdrs[hdr->e_shstrndx].sh_offset; |
432 | for (i = 1; i < hdr->e_shnum; i++) | |
433 | if (strcmp(secstrings+sechdrs[i].sh_name, name) == 0) | |
434 | return &sechdrs[i]; | |
435 | return NULL; | |
436 | } | |
437 | ||
438 | int module_finalize(const Elf_Ehdr *hdr, | |
439 | const Elf_Shdr *sechdrs, struct module *me) | |
440 | { | |
441 | const Elf_Shdr *sect; | |
442 | ||
1da177e4 LT |
443 | me->arch.bug_table = NULL; |
444 | me->arch.num_bugs = 0; | |
445 | ||
446 | /* Find the __bug_table section, if present */ | |
21c4ff80 BH |
447 | sect = find_section(hdr, sechdrs, "__bug_table"); |
448 | if (sect != NULL) { | |
449 | me->arch.bug_table = (void *) sect->sh_addr; | |
450 | me->arch.num_bugs = sect->sh_size / sizeof(struct bug_entry); | |
1da177e4 LT |
451 | } |
452 | ||
453 | /* | |
454 | * Strictly speaking this should have a spinlock to protect against | |
455 | * traversals, but since we only traverse on BUG()s, a spinlock | |
456 | * could potentially lead to deadlock and thus be counter-productive. | |
457 | */ | |
458 | list_add(&me->arch.bug_list, &module_bug_list); | |
459 | ||
21c4ff80 BH |
460 | /* Apply feature fixups */ |
461 | sect = find_section(hdr, sechdrs, "__ftr_fixup"); | |
462 | if (sect != NULL) | |
463 | do_feature_fixups(cur_cpu_spec->cpu_features, | |
464 | (void *)sect->sh_addr, | |
465 | (void *)sect->sh_addr + sect->sh_size); | |
466 | ||
467 | sect = find_section(hdr, sechdrs, "__fw_ftr_fixup"); | |
468 | if (sect != NULL) | |
469 | do_feature_fixups(powerpc_firmware_features, | |
470 | (void *)sect->sh_addr, | |
471 | (void *)sect->sh_addr + sect->sh_size); | |
472 | ||
1da177e4 LT |
473 | return 0; |
474 | } | |
475 | ||
476 | void module_arch_cleanup(struct module *mod) | |
477 | { | |
478 | list_del(&mod->arch.bug_list); | |
479 | } | |
480 | ||
481 | struct bug_entry *module_find_bug(unsigned long bugaddr) | |
482 | { | |
483 | struct mod_arch_specific *mod; | |
484 | unsigned int i; | |
485 | struct bug_entry *bug; | |
486 | ||
487 | list_for_each_entry(mod, &module_bug_list, bug_list) { | |
488 | bug = mod->bug_table; | |
489 | for (i = 0; i < mod->num_bugs; ++i, ++bug) | |
490 | if (bugaddr == bug->bug_addr) | |
491 | return bug; | |
492 | } | |
493 | return NULL; | |
494 | } |