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 | */ | |
c7d1f6af AB |
18 | |
19 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
20 | ||
1da177e4 LT |
21 | #include <linux/module.h> |
22 | #include <linux/elf.h> | |
23 | #include <linux/moduleloader.h> | |
24 | #include <linux/err.h> | |
25 | #include <linux/vmalloc.h> | |
f48cb8b4 | 26 | #include <linux/ftrace.h> |
73c9ceab | 27 | #include <linux/bug.h> |
83775b85 | 28 | #include <linux/uaccess.h> |
1da177e4 | 29 | #include <asm/module.h> |
21c4ff80 | 30 | #include <asm/firmware.h> |
b7bcda63 | 31 | #include <asm/code-patching.h> |
eda09fbd | 32 | #include <linux/sort.h> |
b88c4767 | 33 | #include <asm/setup.h> |
336a7b5d | 34 | #include <asm/sections.h> |
1da177e4 LT |
35 | |
36 | /* FIXME: We don't do .init separately. To do this, we'd need to have | |
37 | a separate r2 value in the init and core section, and stub between | |
38 | them, too. | |
39 | ||
40 | Using a magic allocator which places modules within 32MB solves | |
41 | this, and makes other things simpler. Anton? | |
42 | --RR. */ | |
1da177e4 | 43 | |
d2fae548 | 44 | #if defined(_CALL_ELF) && _CALL_ELF == 2 |
008d7a91 RR |
45 | |
46 | /* An address is simply the address of the function. */ | |
47 | typedef unsigned long func_desc_t; | |
48 | ||
49 | static func_desc_t func_desc(unsigned long addr) | |
50 | { | |
51 | return addr; | |
52 | } | |
53 | static unsigned long func_addr(unsigned long addr) | |
54 | { | |
55 | return addr; | |
56 | } | |
57 | static unsigned long stub_func_addr(func_desc_t func) | |
58 | { | |
59 | return func; | |
60 | } | |
61 | ||
62 | /* PowerPC64 specific values for the Elf64_Sym st_other field. */ | |
63 | #define STO_PPC64_LOCAL_BIT 5 | |
64 | #define STO_PPC64_LOCAL_MASK (7 << STO_PPC64_LOCAL_BIT) | |
65 | #define PPC64_LOCAL_ENTRY_OFFSET(other) \ | |
66 | (((1 << (((other) & STO_PPC64_LOCAL_MASK) >> STO_PPC64_LOCAL_BIT)) >> 2) << 2) | |
67 | ||
68 | static unsigned int local_entry_offset(const Elf64_Sym *sym) | |
69 | { | |
70 | /* sym->st_other indicates offset to local entry point | |
71 | * (otherwise it will assume r12 is the address of the start | |
72 | * of function and try to derive r2 from it). */ | |
73 | return PPC64_LOCAL_ENTRY_OFFSET(sym->st_other); | |
74 | } | |
d2fae548 | 75 | #else |
008d7a91 RR |
76 | |
77 | /* An address is address of the OPD entry, which contains address of fn. */ | |
78 | typedef struct ppc64_opd_entry func_desc_t; | |
79 | ||
80 | static func_desc_t func_desc(unsigned long addr) | |
81 | { | |
82 | return *(struct ppc64_opd_entry *)addr; | |
83 | } | |
84 | static unsigned long func_addr(unsigned long addr) | |
85 | { | |
86 | return func_desc(addr).funcaddr; | |
87 | } | |
88 | static unsigned long stub_func_addr(func_desc_t func) | |
89 | { | |
90 | return func.funcaddr; | |
91 | } | |
92 | static unsigned int local_entry_offset(const Elf64_Sym *sym) | |
93 | { | |
94 | return 0; | |
95 | } | |
d2fae548 RR |
96 | #endif |
97 | ||
f17c4e01 ME |
98 | #define STUB_MAGIC 0x73747562 /* stub */ |
99 | ||
1da177e4 LT |
100 | /* Like PPC32, we need little trampolines to do > 24-bit jumps (into |
101 | the kernel itself). But on PPC64, these need to be used for every | |
102 | jump, actually, to reset r2 (TOC+0x8000). */ | |
103 | struct ppc64_stub_entry | |
104 | { | |
83775b85 AB |
105 | /* 28 byte jump instruction sequence (7 instructions). We only |
106 | * need 6 instructions on ABIv2 but we always allocate 7 so | |
107 | * so we don't have to modify the trampoline load instruction. */ | |
0e60e46e | 108 | u32 jump[7]; |
f17c4e01 ME |
109 | /* Used by ftrace to identify stubs */ |
110 | u32 magic; | |
1da177e4 | 111 | /* Data for the above code */ |
008d7a91 | 112 | func_desc_t funcdata; |
1da177e4 LT |
113 | }; |
114 | ||
5c729a11 RR |
115 | /* |
116 | * PPC64 uses 24 bit jumps, but we need to jump into other modules or | |
117 | * the kernel which may be further. So we jump to a stub. | |
118 | * | |
119 | * For ELFv1 we need to use this to set up the new r2 value (aka TOC | |
120 | * pointer). For ELFv2 it's the callee's responsibility to set up the | |
121 | * new r2, but for both we need to save the old r2. | |
122 | * | |
123 | * We could simply patch the new r2 value and function pointer into | |
124 | * the stub, but it's significantly shorter to put these values at the | |
125 | * end of the stub code, and patch the stub address (32-bits relative | |
126 | * to the TOC ptr, r2) into the stub. | |
127 | */ | |
83775b85 AB |
128 | |
129 | static u32 ppc64_stub_insns[] = { | |
b1ce369e RR |
130 | 0x3d620000, /* addis r11,r2, <high> */ |
131 | 0x396b0000, /* addi r11,r11, <low> */ | |
fed8393e | 132 | /* Save current r2 value in magic place on the stack. */ |
d2fae548 | 133 | 0xf8410000|R2_STACK_OFFSET, /* std r2,R2_STACK_OFFSET(r1) */ |
b1ce369e | 134 | 0xe98b0020, /* ld r12,32(r11) */ |
5c729a11 RR |
135 | #if !defined(_CALL_ELF) || _CALL_ELF != 2 |
136 | /* Set up new r2 from function descriptor */ | |
872aa779 | 137 | 0xe84b0028, /* ld r2,40(r11) */ |
5c729a11 | 138 | #endif |
b1ce369e | 139 | 0x7d8903a6, /* mtctr r12 */ |
d2fae548 | 140 | 0x4e800420 /* bctr */ |
83775b85 AB |
141 | }; |
142 | ||
143 | #ifdef CONFIG_DYNAMIC_FTRACE | |
f17c4e01 ME |
144 | int module_trampoline_target(struct module *mod, unsigned long addr, |
145 | unsigned long *target) | |
83775b85 | 146 | { |
f17c4e01 ME |
147 | struct ppc64_stub_entry *stub; |
148 | func_desc_t funcdata; | |
149 | u32 magic; | |
83775b85 | 150 | |
f17c4e01 ME |
151 | if (!within_module_core(addr, mod)) { |
152 | pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name); | |
83775b85 | 153 | return -EFAULT; |
83775b85 AB |
154 | } |
155 | ||
f17c4e01 | 156 | stub = (struct ppc64_stub_entry *)addr; |
dd9fa162 | 157 | |
f17c4e01 ME |
158 | if (probe_kernel_read(&magic, &stub->magic, sizeof(magic))) { |
159 | pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name); | |
dd9fa162 | 160 | return -EFAULT; |
f17c4e01 | 161 | } |
dd9fa162 | 162 | |
f17c4e01 ME |
163 | if (magic != STUB_MAGIC) { |
164 | pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name); | |
165 | return -EFAULT; | |
166 | } | |
dd9fa162 | 167 | |
f17c4e01 ME |
168 | if (probe_kernel_read(&funcdata, &stub->funcdata, sizeof(funcdata))) { |
169 | pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name); | |
170 | return -EFAULT; | |
171 | } | |
dd9fa162 | 172 | |
f17c4e01 | 173 | *target = stub_func_addr(funcdata); |
dd9fa162 AB |
174 | |
175 | return 0; | |
176 | } | |
83775b85 | 177 | #endif |
1da177e4 LT |
178 | |
179 | /* Count how many different 24-bit relocations (different symbol, | |
180 | different addend) */ | |
181 | static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) | |
182 | { | |
eda09fbd | 183 | unsigned int i, r_info, r_addend, _count_relocs; |
1da177e4 LT |
184 | |
185 | /* FIXME: Only count external ones --RR */ | |
eda09fbd EM |
186 | _count_relocs = 0; |
187 | r_info = 0; | |
188 | r_addend = 0; | |
189 | for (i = 0; i < num; i++) | |
1da177e4 | 190 | /* Only count 24-bit relocs, others don't need stubs */ |
eda09fbd EM |
191 | if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 && |
192 | (r_info != ELF64_R_SYM(rela[i].r_info) || | |
193 | r_addend != rela[i].r_addend)) { | |
194 | _count_relocs++; | |
195 | r_info = ELF64_R_SYM(rela[i].r_info); | |
196 | r_addend = rela[i].r_addend; | |
1da177e4 | 197 | } |
eda09fbd EM |
198 | |
199 | return _count_relocs; | |
1da177e4 LT |
200 | } |
201 | ||
eda09fbd EM |
202 | static int relacmp(const void *_x, const void *_y) |
203 | { | |
204 | const Elf64_Rela *x, *y; | |
205 | ||
206 | y = (Elf64_Rela *)_x; | |
207 | x = (Elf64_Rela *)_y; | |
208 | ||
209 | /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to | |
210 | * make the comparison cheaper/faster. It won't affect the sorting or | |
211 | * the counting algorithms' performance | |
212 | */ | |
213 | if (x->r_info < y->r_info) | |
214 | return -1; | |
215 | else if (x->r_info > y->r_info) | |
216 | return 1; | |
217 | else if (x->r_addend < y->r_addend) | |
218 | return -1; | |
219 | else if (x->r_addend > y->r_addend) | |
220 | return 1; | |
221 | else | |
222 | return 0; | |
223 | } | |
224 | ||
225 | static void relaswap(void *_x, void *_y, int size) | |
226 | { | |
227 | uint64_t *x, *y, tmp; | |
228 | int i; | |
229 | ||
230 | y = (uint64_t *)_x; | |
231 | x = (uint64_t *)_y; | |
232 | ||
233 | for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) { | |
234 | tmp = x[i]; | |
235 | x[i] = y[i]; | |
236 | y[i] = tmp; | |
237 | } | |
238 | } | |
239 | ||
1da177e4 LT |
240 | /* Get size of potential trampolines required. */ |
241 | static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, | |
242 | const Elf64_Shdr *sechdrs) | |
243 | { | |
244 | /* One extra reloc so it's always 0-funcaddr terminated */ | |
245 | unsigned long relocs = 1; | |
246 | unsigned i; | |
247 | ||
248 | /* Every relocated section... */ | |
249 | for (i = 1; i < hdr->e_shnum; i++) { | |
250 | if (sechdrs[i].sh_type == SHT_RELA) { | |
c7d1f6af AB |
251 | pr_debug("Found relocations in section %u\n", i); |
252 | pr_debug("Ptr: %p. Number: %Lu\n", | |
1da177e4 LT |
253 | (void *)sechdrs[i].sh_addr, |
254 | sechdrs[i].sh_size / sizeof(Elf64_Rela)); | |
eda09fbd EM |
255 | |
256 | /* Sort the relocation information based on a symbol and | |
257 | * addend key. This is a stable O(n*log n) complexity | |
258 | * alogrithm but it will reduce the complexity of | |
259 | * count_relocs() to linear complexity O(n) | |
260 | */ | |
261 | sort((void *)sechdrs[i].sh_addr, | |
262 | sechdrs[i].sh_size / sizeof(Elf64_Rela), | |
263 | sizeof(Elf64_Rela), relacmp, relaswap); | |
264 | ||
1da177e4 LT |
265 | relocs += count_relocs((void *)sechdrs[i].sh_addr, |
266 | sechdrs[i].sh_size | |
267 | / sizeof(Elf64_Rela)); | |
268 | } | |
269 | } | |
270 | ||
f48cb8b4 SR |
271 | #ifdef CONFIG_DYNAMIC_FTRACE |
272 | /* make the trampoline to the ftrace_caller */ | |
273 | relocs++; | |
274 | #endif | |
275 | ||
c7d1f6af | 276 | pr_debug("Looks like a total of %lu stubs, max\n", relocs); |
1da177e4 LT |
277 | return relocs * sizeof(struct ppc64_stub_entry); |
278 | } | |
279 | ||
5b12c5c6 | 280 | /* Still needed for ELFv2, for .TOC. */ |
1da177e4 LT |
281 | static void dedotify_versions(struct modversion_info *vers, |
282 | unsigned long size) | |
283 | { | |
284 | struct modversion_info *end; | |
285 | ||
286 | for (end = (void *)vers + size; vers < end; vers++) | |
c2cbcf53 | 287 | if (vers->name[0] == '.') { |
1da177e4 | 288 | memmove(vers->name, vers->name+1, strlen(vers->name)); |
c2cbcf53 LD |
289 | #ifdef ARCH_RELOCATES_KCRCTAB |
290 | /* The TOC symbol has no CRC computed. To avoid CRC | |
291 | * check failing, we must force it to the expected | |
292 | * value (see CRC check in module.c). | |
293 | */ | |
294 | if (!strcmp(vers->name, "TOC.")) | |
295 | vers->crc = -(unsigned long)reloc_start; | |
296 | #endif | |
297 | } | |
1da177e4 LT |
298 | } |
299 | ||
c153693d AM |
300 | /* |
301 | * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC. | |
302 | * seem to be defined (value set later). | |
303 | */ | |
1da177e4 LT |
304 | static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) |
305 | { | |
306 | unsigned int i; | |
307 | ||
308 | for (i = 1; i < numsyms; i++) { | |
309 | if (syms[i].st_shndx == SHN_UNDEF) { | |
310 | char *name = strtab + syms[i].st_name; | |
c153693d AM |
311 | if (name[0] == '.') { |
312 | if (strcmp(name+1, "TOC.") == 0) | |
313 | syms[i].st_shndx = SHN_ABS; | |
f15838e9 | 314 | syms[i].st_name++; |
c153693d | 315 | } |
1da177e4 LT |
316 | } |
317 | } | |
318 | } | |
319 | ||
4edebbea RR |
320 | static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs, |
321 | const char *strtab, | |
322 | unsigned int symindex) | |
323 | { | |
324 | unsigned int i, numsyms; | |
325 | Elf64_Sym *syms; | |
326 | ||
327 | syms = (Elf64_Sym *)sechdrs[symindex].sh_addr; | |
328 | numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym); | |
329 | ||
330 | for (i = 1; i < numsyms; i++) { | |
c153693d | 331 | if (syms[i].st_shndx == SHN_ABS |
008d7a91 | 332 | && strcmp(strtab + syms[i].st_name, "TOC.") == 0) |
4edebbea RR |
333 | return &syms[i]; |
334 | } | |
335 | return NULL; | |
336 | } | |
337 | ||
1da177e4 LT |
338 | int module_frob_arch_sections(Elf64_Ehdr *hdr, |
339 | Elf64_Shdr *sechdrs, | |
340 | char *secstrings, | |
341 | struct module *me) | |
342 | { | |
343 | unsigned int i; | |
344 | ||
345 | /* Find .toc and .stubs sections, symtab and strtab */ | |
346 | for (i = 1; i < hdr->e_shnum; i++) { | |
347 | char *p; | |
348 | if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) | |
349 | me->arch.stubs_section = i; | |
350 | else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) | |
351 | me->arch.toc_section = i; | |
352 | else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) | |
353 | dedotify_versions((void *)hdr + sechdrs[i].sh_offset, | |
354 | sechdrs[i].sh_size); | |
355 | ||
356 | /* We don't handle .init for the moment: rename to _init */ | |
357 | while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) | |
358 | p[0] = '_'; | |
359 | ||
360 | if (sechdrs[i].sh_type == SHT_SYMTAB) | |
361 | dedotify((void *)hdr + sechdrs[i].sh_offset, | |
362 | sechdrs[i].sh_size / sizeof(Elf64_Sym), | |
363 | (void *)hdr | |
364 | + sechdrs[sechdrs[i].sh_link].sh_offset); | |
365 | } | |
f749edae AM |
366 | |
367 | if (!me->arch.stubs_section) { | |
c7d1f6af | 368 | pr_err("%s: doesn't contain .stubs.\n", me->name); |
1da177e4 LT |
369 | return -ENOEXEC; |
370 | } | |
371 | ||
f749edae AM |
372 | /* If we don't have a .toc, just use .stubs. We need to set r2 |
373 | to some reasonable value in case the module calls out to | |
374 | other functions via a stub, or if a function pointer escapes | |
375 | the module by some means. */ | |
376 | if (!me->arch.toc_section) | |
377 | me->arch.toc_section = me->arch.stubs_section; | |
378 | ||
1da177e4 LT |
379 | /* Override the stubs size */ |
380 | sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); | |
381 | return 0; | |
382 | } | |
383 | ||
1da177e4 LT |
384 | /* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this |
385 | gives the value maximum span in an instruction which uses a signed | |
386 | offset) */ | |
136cd345 | 387 | static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me) |
1da177e4 LT |
388 | { |
389 | return sechdrs[me->arch.toc_section].sh_addr + 0x8000; | |
390 | } | |
391 | ||
392 | /* Both low and high 16 bits are added as SIGNED additions, so if low | |
393 | 16 bits has high bit set, high 16 bits must be adjusted. These | |
394 | macros do that (stolen from binutils). */ | |
395 | #define PPC_LO(v) ((v) & 0xffff) | |
396 | #define PPC_HI(v) (((v) >> 16) & 0xffff) | |
397 | #define PPC_HA(v) PPC_HI ((v) + 0x8000) | |
398 | ||
399 | /* Patch stub to reference function and correct r2 value. */ | |
136cd345 | 400 | static inline int create_stub(const Elf64_Shdr *sechdrs, |
1da177e4 | 401 | struct ppc64_stub_entry *entry, |
008d7a91 | 402 | unsigned long addr, |
1da177e4 LT |
403 | struct module *me) |
404 | { | |
1da177e4 LT |
405 | long reladdr; |
406 | ||
83775b85 | 407 | memcpy(entry->jump, ppc64_stub_insns, sizeof(ppc64_stub_insns)); |
1da177e4 | 408 | |
1da177e4 LT |
409 | /* Stub uses address relative to r2. */ |
410 | reladdr = (unsigned long)entry - my_r2(sechdrs, me); | |
411 | if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { | |
c7d1f6af | 412 | pr_err("%s: Address %p of stub out of range of %p.\n", |
1da177e4 LT |
413 | me->name, (void *)reladdr, (void *)my_r2); |
414 | return 0; | |
415 | } | |
c7d1f6af | 416 | pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr); |
1da177e4 | 417 | |
0e60e46e RR |
418 | entry->jump[0] |= PPC_HA(reladdr); |
419 | entry->jump[1] |= PPC_LO(reladdr); | |
008d7a91 | 420 | entry->funcdata = func_desc(addr); |
f17c4e01 ME |
421 | entry->magic = STUB_MAGIC; |
422 | ||
1da177e4 LT |
423 | return 1; |
424 | } | |
425 | ||
008d7a91 | 426 | /* Create stub to jump to function described in this OPD/ptr: we need the |
1da177e4 | 427 | stub to set up the TOC ptr (r2) for the function. */ |
136cd345 | 428 | static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs, |
008d7a91 | 429 | unsigned long addr, |
1da177e4 LT |
430 | struct module *me) |
431 | { | |
432 | struct ppc64_stub_entry *stubs; | |
1da177e4 LT |
433 | unsigned int i, num_stubs; |
434 | ||
435 | num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); | |
436 | ||
437 | /* Find this stub, or if that fails, the next avail. entry */ | |
438 | stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; | |
008d7a91 | 439 | for (i = 0; stub_func_addr(stubs[i].funcdata); i++) { |
1da177e4 LT |
440 | BUG_ON(i >= num_stubs); |
441 | ||
008d7a91 | 442 | if (stub_func_addr(stubs[i].funcdata) == func_addr(addr)) |
1da177e4 LT |
443 | return (unsigned long)&stubs[i]; |
444 | } | |
445 | ||
008d7a91 | 446 | if (!create_stub(sechdrs, &stubs[i], addr, me)) |
1da177e4 LT |
447 | return 0; |
448 | ||
449 | return (unsigned long)&stubs[i]; | |
450 | } | |
451 | ||
15308664 TD |
452 | #ifdef CC_USING_MPROFILE_KERNEL |
453 | static bool is_early_mcount_callsite(u32 *instruction) | |
454 | { | |
455 | /* | |
456 | * Check if this is one of the -mprofile-kernel sequences. | |
457 | */ | |
458 | if (instruction[-1] == PPC_INST_STD_LR && | |
459 | instruction[-2] == PPC_INST_MFLR) | |
460 | return true; | |
461 | ||
462 | if (instruction[-1] == PPC_INST_MFLR) | |
463 | return true; | |
464 | ||
465 | return false; | |
466 | } | |
467 | ||
468 | /* | |
469 | * In case of _mcount calls, do not save the current callee's TOC (in r2) into | |
470 | * the original caller's stack frame. If we did we would clobber the saved TOC | |
471 | * value of the original caller. | |
472 | */ | |
473 | static void squash_toc_save_inst(const char *name, unsigned long addr) | |
474 | { | |
475 | struct ppc64_stub_entry *stub = (struct ppc64_stub_entry *)addr; | |
476 | ||
477 | /* Only for calls to _mcount */ | |
478 | if (strcmp("_mcount", name) != 0) | |
479 | return; | |
480 | ||
481 | stub->jump[2] = PPC_INST_NOP; | |
482 | } | |
483 | #else | |
484 | static void squash_toc_save_inst(const char *name, unsigned long addr) { } | |
485 | ||
486 | /* without -mprofile-kernel, mcount calls are never early */ | |
487 | static bool is_early_mcount_callsite(u32 *instruction) | |
488 | { | |
489 | return false; | |
490 | } | |
491 | #endif | |
492 | ||
1da177e4 LT |
493 | /* We expect a noop next: if it is, replace it with instruction to |
494 | restore r2. */ | |
495 | static int restore_r2(u32 *instruction, struct module *me) | |
496 | { | |
16c57b36 | 497 | if (*instruction != PPC_INST_NOP) { |
15308664 TD |
498 | if (is_early_mcount_callsite(instruction - 1)) |
499 | return 1; | |
c7d1f6af | 500 | pr_err("%s: Expect noop after relocate, got %08x\n", |
1da177e4 LT |
501 | me->name, *instruction); |
502 | return 0; | |
503 | } | |
d2fae548 | 504 | /* ld r2,R2_STACK_OFFSET(r1) */ |
15308664 | 505 | *instruction = PPC_INST_LD_TOC; |
1da177e4 LT |
506 | return 1; |
507 | } | |
508 | ||
509 | int apply_relocate_add(Elf64_Shdr *sechdrs, | |
510 | const char *strtab, | |
511 | unsigned int symindex, | |
512 | unsigned int relsec, | |
513 | struct module *me) | |
514 | { | |
515 | unsigned int i; | |
516 | Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; | |
517 | Elf64_Sym *sym; | |
518 | unsigned long *location; | |
519 | unsigned long value; | |
520 | ||
c7d1f6af | 521 | pr_debug("Applying ADD relocate section %u to %u\n", relsec, |
1da177e4 | 522 | sechdrs[relsec].sh_info); |
4edebbea RR |
523 | |
524 | /* First time we're called, we can fix up .TOC. */ | |
525 | if (!me->arch.toc_fixed) { | |
526 | sym = find_dot_toc(sechdrs, strtab, symindex); | |
527 | /* It's theoretically possible that a module doesn't want a | |
528 | * .TOC. so don't fail it just for that. */ | |
529 | if (sym) | |
530 | sym->st_value = my_r2(sechdrs, me); | |
531 | me->arch.toc_fixed = true; | |
532 | } | |
533 | ||
1da177e4 LT |
534 | for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { |
535 | /* This is where to make the change */ | |
536 | location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr | |
537 | + rela[i].r_offset; | |
538 | /* This is the symbol it is referring to */ | |
539 | sym = (Elf64_Sym *)sechdrs[symindex].sh_addr | |
540 | + ELF64_R_SYM(rela[i].r_info); | |
541 | ||
c7d1f6af | 542 | pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n", |
1da177e4 LT |
543 | location, (long)ELF64_R_TYPE(rela[i].r_info), |
544 | strtab + sym->st_name, (unsigned long)sym->st_value, | |
545 | (long)rela[i].r_addend); | |
546 | ||
547 | /* `Everything is relative'. */ | |
548 | value = sym->st_value + rela[i].r_addend; | |
549 | ||
550 | switch (ELF64_R_TYPE(rela[i].r_info)) { | |
551 | case R_PPC64_ADDR32: | |
552 | /* Simply set it */ | |
553 | *(u32 *)location = value; | |
554 | break; | |
eda09fbd | 555 | |
1da177e4 LT |
556 | case R_PPC64_ADDR64: |
557 | /* Simply set it */ | |
558 | *(unsigned long *)location = value; | |
559 | break; | |
560 | ||
561 | case R_PPC64_TOC: | |
562 | *(unsigned long *)location = my_r2(sechdrs, me); | |
563 | break; | |
564 | ||
9149ccfa | 565 | case R_PPC64_TOC16: |
f749edae | 566 | /* Subtract TOC pointer */ |
9149ccfa PB |
567 | value -= my_r2(sechdrs, me); |
568 | if (value + 0x8000 > 0xffff) { | |
c7d1f6af | 569 | pr_err("%s: bad TOC16 relocation (0x%lx)\n", |
9149ccfa PB |
570 | me->name, value); |
571 | return -ENOEXEC; | |
572 | } | |
573 | *((uint16_t *) location) | |
574 | = (*((uint16_t *) location) & ~0xffff) | |
575 | | (value & 0xffff); | |
576 | break; | |
577 | ||
1fbe9cf2 AB |
578 | case R_PPC64_TOC16_LO: |
579 | /* Subtract TOC pointer */ | |
580 | value -= my_r2(sechdrs, me); | |
581 | *((uint16_t *) location) | |
582 | = (*((uint16_t *) location) & ~0xffff) | |
583 | | (value & 0xffff); | |
584 | break; | |
585 | ||
1da177e4 | 586 | case R_PPC64_TOC16_DS: |
f749edae | 587 | /* Subtract TOC pointer */ |
1da177e4 LT |
588 | value -= my_r2(sechdrs, me); |
589 | if ((value & 3) != 0 || value + 0x8000 > 0xffff) { | |
c7d1f6af | 590 | pr_err("%s: bad TOC16_DS relocation (0x%lx)\n", |
1da177e4 LT |
591 | me->name, value); |
592 | return -ENOEXEC; | |
593 | } | |
594 | *((uint16_t *) location) | |
595 | = (*((uint16_t *) location) & ~0xfffc) | |
596 | | (value & 0xfffc); | |
597 | break; | |
598 | ||
1fbe9cf2 AB |
599 | case R_PPC64_TOC16_LO_DS: |
600 | /* Subtract TOC pointer */ | |
601 | value -= my_r2(sechdrs, me); | |
602 | if ((value & 3) != 0) { | |
c7d1f6af | 603 | pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n", |
1fbe9cf2 AB |
604 | me->name, value); |
605 | return -ENOEXEC; | |
606 | } | |
607 | *((uint16_t *) location) | |
608 | = (*((uint16_t *) location) & ~0xfffc) | |
609 | | (value & 0xfffc); | |
610 | break; | |
611 | ||
612 | case R_PPC64_TOC16_HA: | |
613 | /* Subtract TOC pointer */ | |
614 | value -= my_r2(sechdrs, me); | |
615 | value = ((value + 0x8000) >> 16); | |
616 | *((uint16_t *) location) | |
617 | = (*((uint16_t *) location) & ~0xffff) | |
618 | | (value & 0xffff); | |
619 | break; | |
620 | ||
1da177e4 LT |
621 | case R_PPC_REL24: |
622 | /* FIXME: Handle weak symbols here --RR */ | |
623 | if (sym->st_shndx == SHN_UNDEF) { | |
624 | /* External: go via stub */ | |
625 | value = stub_for_addr(sechdrs, value, me); | |
626 | if (!value) | |
627 | return -ENOENT; | |
628 | if (!restore_r2((u32 *)location + 1, me)) | |
629 | return -ENOEXEC; | |
15308664 TD |
630 | |
631 | squash_toc_save_inst(strtab + sym->st_name, value); | |
008d7a91 RR |
632 | } else |
633 | value += local_entry_offset(sym); | |
1da177e4 LT |
634 | |
635 | /* Convert value to relative */ | |
636 | value -= (unsigned long)location; | |
637 | if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ | |
c7d1f6af | 638 | pr_err("%s: REL24 %li out of range!\n", |
1da177e4 LT |
639 | me->name, (long int)value); |
640 | return -ENOEXEC; | |
641 | } | |
642 | ||
643 | /* Only replace bits 2 through 26 */ | |
eda09fbd | 644 | *(uint32_t *)location |
1da177e4 LT |
645 | = (*(uint32_t *)location & ~0x03fffffc) |
646 | | (value & 0x03fffffc); | |
647 | break; | |
648 | ||
21c4ff80 BH |
649 | case R_PPC64_REL64: |
650 | /* 64 bits relative (used by features fixups) */ | |
651 | *location = value - (unsigned long)location; | |
652 | break; | |
653 | ||
d247da0a RR |
654 | case R_PPC64_TOCSAVE: |
655 | /* | |
656 | * Marker reloc indicates we don't have to save r2. | |
657 | * That would only save us one instruction, so ignore | |
658 | * it. | |
659 | */ | |
660 | break; | |
661 | ||
a61674bd UW |
662 | case R_PPC64_ENTRY: |
663 | /* | |
664 | * Optimize ELFv2 large code model entry point if | |
665 | * the TOC is within 2GB range of current location. | |
666 | */ | |
667 | value = my_r2(sechdrs, me) - (unsigned long)location; | |
668 | if (value + 0x80008000 > 0xffffffff) | |
669 | break; | |
670 | /* | |
671 | * Check for the large code model prolog sequence: | |
672 | * ld r2, ...(r12) | |
673 | * add r2, r2, r12 | |
674 | */ | |
675 | if ((((uint32_t *)location)[0] & ~0xfffc) | |
676 | != 0xe84c0000) | |
677 | break; | |
678 | if (((uint32_t *)location)[1] != 0x7c426214) | |
679 | break; | |
680 | /* | |
681 | * If found, replace it with: | |
682 | * addis r2, r12, (.TOC.-func)@ha | |
683 | * addi r2, r12, (.TOC.-func)@l | |
684 | */ | |
685 | ((uint32_t *)location)[0] = 0x3c4c0000 + PPC_HA(value); | |
686 | ((uint32_t *)location)[1] = 0x38420000 + PPC_LO(value); | |
687 | break; | |
688 | ||
0906584a RR |
689 | case R_PPC64_REL16_HA: |
690 | /* Subtract location pointer */ | |
691 | value -= (unsigned long)location; | |
692 | value = ((value + 0x8000) >> 16); | |
693 | *((uint16_t *) location) | |
694 | = (*((uint16_t *) location) & ~0xffff) | |
695 | | (value & 0xffff); | |
696 | break; | |
697 | ||
698 | case R_PPC64_REL16_LO: | |
699 | /* Subtract location pointer */ | |
700 | value -= (unsigned long)location; | |
701 | *((uint16_t *) location) | |
702 | = (*((uint16_t *) location) & ~0xffff) | |
703 | | (value & 0xffff); | |
704 | break; | |
705 | ||
1da177e4 | 706 | default: |
c7d1f6af | 707 | pr_err("%s: Unknown ADD relocation: %lu\n", |
1da177e4 LT |
708 | me->name, |
709 | (unsigned long)ELF64_R_TYPE(rela[i].r_info)); | |
710 | return -ENOEXEC; | |
711 | } | |
712 | } | |
713 | ||
136cd345 ME |
714 | return 0; |
715 | } | |
716 | ||
f48cb8b4 | 717 | #ifdef CONFIG_DYNAMIC_FTRACE |
336a7b5d ME |
718 | |
719 | #ifdef CC_USING_MPROFILE_KERNEL | |
720 | ||
721 | #define PACATOC offsetof(struct paca_struct, kernel_toc) | |
722 | ||
723 | /* | |
724 | * For mprofile-kernel we use a special stub for ftrace_caller() because we | |
725 | * can't rely on r2 containing this module's TOC when we enter the stub. | |
726 | * | |
727 | * That can happen if the function calling us didn't need to use the toc. In | |
728 | * that case it won't have setup r2, and the r2 value will be either the | |
729 | * kernel's toc, or possibly another modules toc. | |
730 | * | |
731 | * To deal with that this stub uses the kernel toc, which is always accessible | |
732 | * via the paca (in r13). The target (ftrace_caller()) is responsible for | |
733 | * saving and restoring the toc before returning. | |
734 | */ | |
735 | static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, struct module *me) | |
736 | { | |
737 | struct ppc64_stub_entry *entry; | |
738 | unsigned int i, num_stubs; | |
739 | static u32 stub_insns[] = { | |
740 | 0xe98d0000 | PACATOC, /* ld r12,PACATOC(r13) */ | |
741 | 0x3d8c0000, /* addis r12,r12,<high> */ | |
742 | 0x398c0000, /* addi r12,r12,<low> */ | |
743 | 0x7d8903a6, /* mtctr r12 */ | |
744 | 0x4e800420, /* bctr */ | |
745 | }; | |
746 | long reladdr; | |
747 | ||
748 | num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*entry); | |
749 | ||
750 | /* Find the next available stub entry */ | |
751 | entry = (void *)sechdrs[me->arch.stubs_section].sh_addr; | |
752 | for (i = 0; i < num_stubs && stub_func_addr(entry->funcdata); i++, entry++); | |
753 | ||
754 | if (i >= num_stubs) { | |
755 | pr_err("%s: Unable to find a free slot for ftrace stub.\n", me->name); | |
756 | return 0; | |
757 | } | |
758 | ||
759 | memcpy(entry->jump, stub_insns, sizeof(stub_insns)); | |
760 | ||
761 | /* Stub uses address relative to kernel toc (from the paca) */ | |
762 | reladdr = (unsigned long)ftrace_caller - kernel_toc_addr(); | |
763 | if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { | |
764 | pr_err("%s: Address of ftrace_caller out of range of kernel_toc.\n", me->name); | |
765 | return 0; | |
766 | } | |
767 | ||
768 | entry->jump[1] |= PPC_HA(reladdr); | |
769 | entry->jump[2] |= PPC_LO(reladdr); | |
770 | ||
771 | /* Eventhough we don't use funcdata in the stub, it's needed elsewhere. */ | |
772 | entry->funcdata = func_desc((unsigned long)ftrace_caller); | |
773 | entry->magic = STUB_MAGIC; | |
774 | ||
775 | return (unsigned long)entry; | |
776 | } | |
777 | #else | |
778 | static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, struct module *me) | |
779 | { | |
780 | return stub_for_addr(sechdrs, (unsigned long)ftrace_caller, me); | |
781 | } | |
f48cb8b4 SR |
782 | #endif |
783 | ||
136cd345 ME |
784 | int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs) |
785 | { | |
786 | mod->arch.toc = my_r2(sechdrs, mod); | |
336a7b5d | 787 | mod->arch.tramp = create_ftrace_stub(sechdrs, mod); |
136cd345 ME |
788 | |
789 | if (!mod->arch.tramp) | |
790 | return -ENOENT; | |
f48cb8b4 | 791 | |
1da177e4 LT |
792 | return 0; |
793 | } | |
136cd345 | 794 | #endif |