]>
Commit | Line | Data |
---|---|---|
1a59d1b8 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* Kernel module help for PPC64. |
3 | Copyright (C) 2001, 2003 Rusty Russell IBM Corporation. | |
4 | ||
1da177e4 | 5 | */ |
c7d1f6af AB |
6 | |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
8 | ||
1da177e4 LT |
9 | #include <linux/module.h> |
10 | #include <linux/elf.h> | |
11 | #include <linux/moduleloader.h> | |
12 | #include <linux/err.h> | |
13 | #include <linux/vmalloc.h> | |
f48cb8b4 | 14 | #include <linux/ftrace.h> |
73c9ceab | 15 | #include <linux/bug.h> |
83775b85 | 16 | #include <linux/uaccess.h> |
1da177e4 | 17 | #include <asm/module.h> |
21c4ff80 | 18 | #include <asm/firmware.h> |
b7bcda63 | 19 | #include <asm/code-patching.h> |
eda09fbd | 20 | #include <linux/sort.h> |
b88c4767 | 21 | #include <asm/setup.h> |
336a7b5d | 22 | #include <asm/sections.h> |
1da177e4 LT |
23 | |
24 | /* FIXME: We don't do .init separately. To do this, we'd need to have | |
25 | a separate r2 value in the init and core section, and stub between | |
26 | them, too. | |
27 | ||
28 | Using a magic allocator which places modules within 32MB solves | |
29 | this, and makes other things simpler. Anton? | |
30 | --RR. */ | |
1da177e4 | 31 | |
f55d9665 | 32 | #ifdef PPC64_ELF_ABI_v2 |
008d7a91 RR |
33 | |
34 | /* An address is simply the address of the function. */ | |
35 | typedef unsigned long func_desc_t; | |
36 | ||
37 | static func_desc_t func_desc(unsigned long addr) | |
38 | { | |
39 | return addr; | |
40 | } | |
41 | static unsigned long func_addr(unsigned long addr) | |
42 | { | |
43 | return addr; | |
44 | } | |
45 | static unsigned long stub_func_addr(func_desc_t func) | |
46 | { | |
47 | return func; | |
48 | } | |
49 | ||
50 | /* PowerPC64 specific values for the Elf64_Sym st_other field. */ | |
51 | #define STO_PPC64_LOCAL_BIT 5 | |
52 | #define STO_PPC64_LOCAL_MASK (7 << STO_PPC64_LOCAL_BIT) | |
53 | #define PPC64_LOCAL_ENTRY_OFFSET(other) \ | |
54 | (((1 << (((other) & STO_PPC64_LOCAL_MASK) >> STO_PPC64_LOCAL_BIT)) >> 2) << 2) | |
55 | ||
56 | static unsigned int local_entry_offset(const Elf64_Sym *sym) | |
57 | { | |
58 | /* sym->st_other indicates offset to local entry point | |
59 | * (otherwise it will assume r12 is the address of the start | |
60 | * of function and try to derive r2 from it). */ | |
61 | return PPC64_LOCAL_ENTRY_OFFSET(sym->st_other); | |
62 | } | |
d2fae548 | 63 | #else |
008d7a91 RR |
64 | |
65 | /* An address is address of the OPD entry, which contains address of fn. */ | |
66 | typedef struct ppc64_opd_entry func_desc_t; | |
67 | ||
68 | static func_desc_t func_desc(unsigned long addr) | |
69 | { | |
70 | return *(struct ppc64_opd_entry *)addr; | |
71 | } | |
72 | static unsigned long func_addr(unsigned long addr) | |
73 | { | |
74 | return func_desc(addr).funcaddr; | |
75 | } | |
76 | static unsigned long stub_func_addr(func_desc_t func) | |
77 | { | |
78 | return func.funcaddr; | |
79 | } | |
80 | static unsigned int local_entry_offset(const Elf64_Sym *sym) | |
81 | { | |
82 | return 0; | |
83 | } | |
5633e85b SS |
84 | |
85 | void *dereference_module_function_descriptor(struct module *mod, void *ptr) | |
86 | { | |
87 | if (ptr < (void *)mod->arch.start_opd || | |
88 | ptr >= (void *)mod->arch.end_opd) | |
89 | return ptr; | |
90 | ||
91 | return dereference_function_descriptor(ptr); | |
92 | } | |
d2fae548 RR |
93 | #endif |
94 | ||
f17c4e01 ME |
95 | #define STUB_MAGIC 0x73747562 /* stub */ |
96 | ||
1da177e4 LT |
97 | /* Like PPC32, we need little trampolines to do > 24-bit jumps (into |
98 | the kernel itself). But on PPC64, these need to be used for every | |
99 | jump, actually, to reset r2 (TOC+0x8000). */ | |
100 | struct ppc64_stub_entry | |
101 | { | |
83775b85 AB |
102 | /* 28 byte jump instruction sequence (7 instructions). We only |
103 | * need 6 instructions on ABIv2 but we always allocate 7 so | |
104 | * so we don't have to modify the trampoline load instruction. */ | |
0e60e46e | 105 | u32 jump[7]; |
f17c4e01 ME |
106 | /* Used by ftrace to identify stubs */ |
107 | u32 magic; | |
1da177e4 | 108 | /* Data for the above code */ |
008d7a91 | 109 | func_desc_t funcdata; |
1da177e4 LT |
110 | }; |
111 | ||
5c729a11 RR |
112 | /* |
113 | * PPC64 uses 24 bit jumps, but we need to jump into other modules or | |
114 | * the kernel which may be further. So we jump to a stub. | |
115 | * | |
116 | * For ELFv1 we need to use this to set up the new r2 value (aka TOC | |
117 | * pointer). For ELFv2 it's the callee's responsibility to set up the | |
118 | * new r2, but for both we need to save the old r2. | |
119 | * | |
120 | * We could simply patch the new r2 value and function pointer into | |
121 | * the stub, but it's significantly shorter to put these values at the | |
122 | * end of the stub code, and patch the stub address (32-bits relative | |
123 | * to the TOC ptr, r2) into the stub. | |
124 | */ | |
83775b85 AB |
125 | |
126 | static u32 ppc64_stub_insns[] = { | |
b1ce369e RR |
127 | 0x3d620000, /* addis r11,r2, <high> */ |
128 | 0x396b0000, /* addi r11,r11, <low> */ | |
fed8393e | 129 | /* Save current r2 value in magic place on the stack. */ |
d2fae548 | 130 | 0xf8410000|R2_STACK_OFFSET, /* std r2,R2_STACK_OFFSET(r1) */ |
b1ce369e | 131 | 0xe98b0020, /* ld r12,32(r11) */ |
f55d9665 | 132 | #ifdef PPC64_ELF_ABI_v1 |
5c729a11 | 133 | /* Set up new r2 from function descriptor */ |
872aa779 | 134 | 0xe84b0028, /* ld r2,40(r11) */ |
5c729a11 | 135 | #endif |
b1ce369e | 136 | 0x7d8903a6, /* mtctr r12 */ |
d2fae548 | 137 | 0x4e800420 /* bctr */ |
83775b85 AB |
138 | }; |
139 | ||
140 | #ifdef CONFIG_DYNAMIC_FTRACE | |
f17c4e01 ME |
141 | int module_trampoline_target(struct module *mod, unsigned long addr, |
142 | unsigned long *target) | |
83775b85 | 143 | { |
f17c4e01 ME |
144 | struct ppc64_stub_entry *stub; |
145 | func_desc_t funcdata; | |
146 | u32 magic; | |
83775b85 | 147 | |
f17c4e01 ME |
148 | if (!within_module_core(addr, mod)) { |
149 | pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name); | |
83775b85 | 150 | return -EFAULT; |
83775b85 AB |
151 | } |
152 | ||
f17c4e01 | 153 | stub = (struct ppc64_stub_entry *)addr; |
dd9fa162 | 154 | |
f17c4e01 ME |
155 | if (probe_kernel_read(&magic, &stub->magic, sizeof(magic))) { |
156 | pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name); | |
dd9fa162 | 157 | return -EFAULT; |
f17c4e01 | 158 | } |
dd9fa162 | 159 | |
f17c4e01 ME |
160 | if (magic != STUB_MAGIC) { |
161 | pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name); | |
162 | return -EFAULT; | |
163 | } | |
dd9fa162 | 164 | |
f17c4e01 ME |
165 | if (probe_kernel_read(&funcdata, &stub->funcdata, sizeof(funcdata))) { |
166 | pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name); | |
167 | return -EFAULT; | |
168 | } | |
dd9fa162 | 169 | |
f17c4e01 | 170 | *target = stub_func_addr(funcdata); |
dd9fa162 AB |
171 | |
172 | return 0; | |
173 | } | |
83775b85 | 174 | #endif |
1da177e4 LT |
175 | |
176 | /* Count how many different 24-bit relocations (different symbol, | |
177 | different addend) */ | |
178 | static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num) | |
179 | { | |
eda09fbd | 180 | unsigned int i, r_info, r_addend, _count_relocs; |
1da177e4 LT |
181 | |
182 | /* FIXME: Only count external ones --RR */ | |
eda09fbd EM |
183 | _count_relocs = 0; |
184 | r_info = 0; | |
185 | r_addend = 0; | |
186 | for (i = 0; i < num; i++) | |
1da177e4 | 187 | /* Only count 24-bit relocs, others don't need stubs */ |
eda09fbd EM |
188 | if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 && |
189 | (r_info != ELF64_R_SYM(rela[i].r_info) || | |
190 | r_addend != rela[i].r_addend)) { | |
191 | _count_relocs++; | |
192 | r_info = ELF64_R_SYM(rela[i].r_info); | |
193 | r_addend = rela[i].r_addend; | |
1da177e4 | 194 | } |
eda09fbd EM |
195 | |
196 | return _count_relocs; | |
1da177e4 LT |
197 | } |
198 | ||
eda09fbd EM |
199 | static int relacmp(const void *_x, const void *_y) |
200 | { | |
201 | const Elf64_Rela *x, *y; | |
202 | ||
203 | y = (Elf64_Rela *)_x; | |
204 | x = (Elf64_Rela *)_y; | |
205 | ||
206 | /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to | |
207 | * make the comparison cheaper/faster. It won't affect the sorting or | |
208 | * the counting algorithms' performance | |
209 | */ | |
210 | if (x->r_info < y->r_info) | |
211 | return -1; | |
212 | else if (x->r_info > y->r_info) | |
213 | return 1; | |
214 | else if (x->r_addend < y->r_addend) | |
215 | return -1; | |
216 | else if (x->r_addend > y->r_addend) | |
217 | return 1; | |
218 | else | |
219 | return 0; | |
220 | } | |
221 | ||
222 | static void relaswap(void *_x, void *_y, int size) | |
223 | { | |
224 | uint64_t *x, *y, tmp; | |
225 | int i; | |
226 | ||
227 | y = (uint64_t *)_x; | |
228 | x = (uint64_t *)_y; | |
229 | ||
230 | for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) { | |
231 | tmp = x[i]; | |
232 | x[i] = y[i]; | |
233 | y[i] = tmp; | |
234 | } | |
235 | } | |
236 | ||
1da177e4 LT |
237 | /* Get size of potential trampolines required. */ |
238 | static unsigned long get_stubs_size(const Elf64_Ehdr *hdr, | |
239 | const Elf64_Shdr *sechdrs) | |
240 | { | |
241 | /* One extra reloc so it's always 0-funcaddr terminated */ | |
242 | unsigned long relocs = 1; | |
243 | unsigned i; | |
244 | ||
245 | /* Every relocated section... */ | |
246 | for (i = 1; i < hdr->e_shnum; i++) { | |
247 | if (sechdrs[i].sh_type == SHT_RELA) { | |
c7d1f6af AB |
248 | pr_debug("Found relocations in section %u\n", i); |
249 | pr_debug("Ptr: %p. Number: %Lu\n", | |
1da177e4 LT |
250 | (void *)sechdrs[i].sh_addr, |
251 | sechdrs[i].sh_size / sizeof(Elf64_Rela)); | |
eda09fbd EM |
252 | |
253 | /* Sort the relocation information based on a symbol and | |
254 | * addend key. This is a stable O(n*log n) complexity | |
255 | * alogrithm but it will reduce the complexity of | |
256 | * count_relocs() to linear complexity O(n) | |
257 | */ | |
258 | sort((void *)sechdrs[i].sh_addr, | |
259 | sechdrs[i].sh_size / sizeof(Elf64_Rela), | |
260 | sizeof(Elf64_Rela), relacmp, relaswap); | |
261 | ||
1da177e4 LT |
262 | relocs += count_relocs((void *)sechdrs[i].sh_addr, |
263 | sechdrs[i].sh_size | |
264 | / sizeof(Elf64_Rela)); | |
265 | } | |
266 | } | |
267 | ||
f48cb8b4 SR |
268 | #ifdef CONFIG_DYNAMIC_FTRACE |
269 | /* make the trampoline to the ftrace_caller */ | |
270 | relocs++; | |
ae30cc05 NR |
271 | #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS |
272 | /* an additional one for ftrace_regs_caller */ | |
273 | relocs++; | |
274 | #endif | |
f48cb8b4 SR |
275 | #endif |
276 | ||
c7d1f6af | 277 | pr_debug("Looks like a total of %lu stubs, max\n", relocs); |
1da177e4 LT |
278 | return relocs * sizeof(struct ppc64_stub_entry); |
279 | } | |
280 | ||
5b12c5c6 | 281 | /* Still needed for ELFv2, for .TOC. */ |
1da177e4 LT |
282 | static void dedotify_versions(struct modversion_info *vers, |
283 | unsigned long size) | |
284 | { | |
285 | struct modversion_info *end; | |
286 | ||
287 | for (end = (void *)vers + size; vers < end; vers++) | |
c2cbcf53 | 288 | if (vers->name[0] == '.') { |
1da177e4 | 289 | memmove(vers->name, vers->name+1, strlen(vers->name)); |
c2cbcf53 | 290 | } |
1da177e4 LT |
291 | } |
292 | ||
c153693d AM |
293 | /* |
294 | * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC. | |
295 | * seem to be defined (value set later). | |
296 | */ | |
1da177e4 LT |
297 | static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab) |
298 | { | |
299 | unsigned int i; | |
300 | ||
301 | for (i = 1; i < numsyms; i++) { | |
302 | if (syms[i].st_shndx == SHN_UNDEF) { | |
303 | char *name = strtab + syms[i].st_name; | |
c153693d AM |
304 | if (name[0] == '.') { |
305 | if (strcmp(name+1, "TOC.") == 0) | |
306 | syms[i].st_shndx = SHN_ABS; | |
f15838e9 | 307 | syms[i].st_name++; |
c153693d | 308 | } |
1da177e4 LT |
309 | } |
310 | } | |
311 | } | |
312 | ||
4edebbea RR |
313 | static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs, |
314 | const char *strtab, | |
315 | unsigned int symindex) | |
316 | { | |
317 | unsigned int i, numsyms; | |
318 | Elf64_Sym *syms; | |
319 | ||
320 | syms = (Elf64_Sym *)sechdrs[symindex].sh_addr; | |
321 | numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym); | |
322 | ||
323 | for (i = 1; i < numsyms; i++) { | |
c153693d | 324 | if (syms[i].st_shndx == SHN_ABS |
008d7a91 | 325 | && strcmp(strtab + syms[i].st_name, "TOC.") == 0) |
4edebbea RR |
326 | return &syms[i]; |
327 | } | |
328 | return NULL; | |
329 | } | |
330 | ||
1da177e4 LT |
331 | int module_frob_arch_sections(Elf64_Ehdr *hdr, |
332 | Elf64_Shdr *sechdrs, | |
333 | char *secstrings, | |
334 | struct module *me) | |
335 | { | |
336 | unsigned int i; | |
337 | ||
338 | /* Find .toc and .stubs sections, symtab and strtab */ | |
339 | for (i = 1; i < hdr->e_shnum; i++) { | |
340 | char *p; | |
341 | if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0) | |
342 | me->arch.stubs_section = i; | |
5c45b528 | 343 | else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) { |
1da177e4 | 344 | me->arch.toc_section = i; |
5c45b528 AM |
345 | if (sechdrs[i].sh_addralign < 8) |
346 | sechdrs[i].sh_addralign = 8; | |
347 | } | |
1da177e4 LT |
348 | else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0) |
349 | dedotify_versions((void *)hdr + sechdrs[i].sh_offset, | |
350 | sechdrs[i].sh_size); | |
351 | ||
352 | /* We don't handle .init for the moment: rename to _init */ | |
353 | while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init"))) | |
354 | p[0] = '_'; | |
355 | ||
356 | if (sechdrs[i].sh_type == SHT_SYMTAB) | |
357 | dedotify((void *)hdr + sechdrs[i].sh_offset, | |
358 | sechdrs[i].sh_size / sizeof(Elf64_Sym), | |
359 | (void *)hdr | |
360 | + sechdrs[sechdrs[i].sh_link].sh_offset); | |
361 | } | |
f749edae AM |
362 | |
363 | if (!me->arch.stubs_section) { | |
c7d1f6af | 364 | pr_err("%s: doesn't contain .stubs.\n", me->name); |
1da177e4 LT |
365 | return -ENOEXEC; |
366 | } | |
367 | ||
f749edae AM |
368 | /* If we don't have a .toc, just use .stubs. We need to set r2 |
369 | to some reasonable value in case the module calls out to | |
370 | other functions via a stub, or if a function pointer escapes | |
371 | the module by some means. */ | |
372 | if (!me->arch.toc_section) | |
373 | me->arch.toc_section = me->arch.stubs_section; | |
374 | ||
1da177e4 LT |
375 | /* Override the stubs size */ |
376 | sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs); | |
377 | return 0; | |
378 | } | |
379 | ||
5c45b528 AM |
380 | /* |
381 | * r2 is the TOC pointer: it actually points 0x8000 into the TOC (this gives the | |
382 | * value maximum span in an instruction which uses a signed offset). Round down | |
383 | * to a 256 byte boundary for the odd case where we are setting up r2 without a | |
384 | * .toc section. | |
385 | */ | |
136cd345 | 386 | static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me) |
1da177e4 | 387 | { |
5c45b528 | 388 | return (sechdrs[me->arch.toc_section].sh_addr & ~0xfful) + 0x8000; |
1da177e4 LT |
389 | } |
390 | ||
391 | /* Both low and high 16 bits are added as SIGNED additions, so if low | |
392 | 16 bits has high bit set, high 16 bits must be adjusted. These | |
393 | macros do that (stolen from binutils). */ | |
394 | #define PPC_LO(v) ((v) & 0xffff) | |
395 | #define PPC_HI(v) (((v) >> 16) & 0xffff) | |
396 | #define PPC_HA(v) PPC_HI ((v) + 0x8000) | |
397 | ||
398 | /* Patch stub to reference function and correct r2 value. */ | |
136cd345 | 399 | static inline int create_stub(const Elf64_Shdr *sechdrs, |
1da177e4 | 400 | struct ppc64_stub_entry *entry, |
008d7a91 | 401 | unsigned long addr, |
1da177e4 LT |
402 | struct module *me) |
403 | { | |
1da177e4 LT |
404 | long reladdr; |
405 | ||
83775b85 | 406 | memcpy(entry->jump, ppc64_stub_insns, sizeof(ppc64_stub_insns)); |
1da177e4 | 407 | |
1da177e4 LT |
408 | /* Stub uses address relative to r2. */ |
409 | reladdr = (unsigned long)entry - my_r2(sechdrs, me); | |
410 | if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { | |
c7d1f6af | 411 | pr_err("%s: Address %p of stub out of range of %p.\n", |
1da177e4 LT |
412 | me->name, (void *)reladdr, (void *)my_r2); |
413 | return 0; | |
414 | } | |
c7d1f6af | 415 | pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr); |
1da177e4 | 416 | |
0e60e46e RR |
417 | entry->jump[0] |= PPC_HA(reladdr); |
418 | entry->jump[1] |= PPC_LO(reladdr); | |
008d7a91 | 419 | entry->funcdata = func_desc(addr); |
f17c4e01 ME |
420 | entry->magic = STUB_MAGIC; |
421 | ||
1da177e4 LT |
422 | return 1; |
423 | } | |
424 | ||
008d7a91 | 425 | /* Create stub to jump to function described in this OPD/ptr: we need the |
1da177e4 | 426 | stub to set up the TOC ptr (r2) for the function. */ |
136cd345 | 427 | static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs, |
008d7a91 | 428 | unsigned long addr, |
1da177e4 LT |
429 | struct module *me) |
430 | { | |
431 | struct ppc64_stub_entry *stubs; | |
1da177e4 LT |
432 | unsigned int i, num_stubs; |
433 | ||
434 | num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs); | |
435 | ||
436 | /* Find this stub, or if that fails, the next avail. entry */ | |
437 | stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr; | |
008d7a91 | 438 | for (i = 0; stub_func_addr(stubs[i].funcdata); i++) { |
1c0437af KB |
439 | if (WARN_ON(i >= num_stubs)) |
440 | return 0; | |
1da177e4 | 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 | ||
abba7597 | 452 | #ifdef CONFIG_MPROFILE_KERNEL |
250122ba | 453 | static bool is_mprofile_mcount_callsite(const char *name, u32 *instruction) |
15308664 | 454 | { |
250122ba NR |
455 | if (strcmp("_mcount", name)) |
456 | return false; | |
457 | ||
15308664 TD |
458 | /* |
459 | * Check if this is one of the -mprofile-kernel sequences. | |
460 | */ | |
461 | if (instruction[-1] == PPC_INST_STD_LR && | |
462 | instruction[-2] == PPC_INST_MFLR) | |
463 | return true; | |
464 | ||
465 | if (instruction[-1] == PPC_INST_MFLR) | |
466 | return true; | |
467 | ||
468 | return false; | |
469 | } | |
470 | ||
471 | /* | |
472 | * In case of _mcount calls, do not save the current callee's TOC (in r2) into | |
473 | * the original caller's stack frame. If we did we would clobber the saved TOC | |
474 | * value of the original caller. | |
475 | */ | |
476 | static void squash_toc_save_inst(const char *name, unsigned long addr) | |
477 | { | |
478 | struct ppc64_stub_entry *stub = (struct ppc64_stub_entry *)addr; | |
479 | ||
480 | /* Only for calls to _mcount */ | |
481 | if (strcmp("_mcount", name) != 0) | |
482 | return; | |
483 | ||
484 | stub->jump[2] = PPC_INST_NOP; | |
485 | } | |
486 | #else | |
487 | static void squash_toc_save_inst(const char *name, unsigned long addr) { } | |
488 | ||
250122ba | 489 | static bool is_mprofile_mcount_callsite(const char *name, u32 *instruction) |
15308664 TD |
490 | { |
491 | return false; | |
492 | } | |
493 | #endif | |
494 | ||
1da177e4 LT |
495 | /* We expect a noop next: if it is, replace it with instruction to |
496 | restore r2. */ | |
250122ba | 497 | static int restore_r2(const char *name, u32 *instruction, struct module *me) |
1da177e4 | 498 | { |
b9eab08d JP |
499 | u32 *prev_insn = instruction - 1; |
500 | ||
250122ba | 501 | if (is_mprofile_mcount_callsite(name, prev_insn)) |
b9eab08d JP |
502 | return 1; |
503 | ||
504 | /* | |
505 | * Make sure the branch isn't a sibling call. Sibling calls aren't | |
506 | * "link" branches and they don't return, so they don't need the r2 | |
507 | * restore afterwards. | |
508 | */ | |
509 | if (!instr_is_relative_link_branch(*prev_insn)) | |
31278b17 ME |
510 | return 1; |
511 | ||
16c57b36 | 512 | if (*instruction != PPC_INST_NOP) { |
1ea61ea2 JP |
513 | pr_err("%s: Expected nop after call, got %08x at %pS\n", |
514 | me->name, *instruction, instruction); | |
1da177e4 LT |
515 | return 0; |
516 | } | |
d2fae548 | 517 | /* ld r2,R2_STACK_OFFSET(r1) */ |
15308664 | 518 | *instruction = PPC_INST_LD_TOC; |
1da177e4 LT |
519 | return 1; |
520 | } | |
521 | ||
522 | int apply_relocate_add(Elf64_Shdr *sechdrs, | |
523 | const char *strtab, | |
524 | unsigned int symindex, | |
525 | unsigned int relsec, | |
526 | struct module *me) | |
527 | { | |
528 | unsigned int i; | |
529 | Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr; | |
530 | Elf64_Sym *sym; | |
531 | unsigned long *location; | |
532 | unsigned long value; | |
533 | ||
c7d1f6af | 534 | pr_debug("Applying ADD relocate section %u to %u\n", relsec, |
1da177e4 | 535 | sechdrs[relsec].sh_info); |
4edebbea RR |
536 | |
537 | /* First time we're called, we can fix up .TOC. */ | |
538 | if (!me->arch.toc_fixed) { | |
539 | sym = find_dot_toc(sechdrs, strtab, symindex); | |
540 | /* It's theoretically possible that a module doesn't want a | |
541 | * .TOC. so don't fail it just for that. */ | |
542 | if (sym) | |
543 | sym->st_value = my_r2(sechdrs, me); | |
544 | me->arch.toc_fixed = true; | |
545 | } | |
546 | ||
1da177e4 LT |
547 | for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) { |
548 | /* This is where to make the change */ | |
549 | location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr | |
550 | + rela[i].r_offset; | |
551 | /* This is the symbol it is referring to */ | |
552 | sym = (Elf64_Sym *)sechdrs[symindex].sh_addr | |
553 | + ELF64_R_SYM(rela[i].r_info); | |
554 | ||
c7d1f6af | 555 | pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n", |
1da177e4 LT |
556 | location, (long)ELF64_R_TYPE(rela[i].r_info), |
557 | strtab + sym->st_name, (unsigned long)sym->st_value, | |
558 | (long)rela[i].r_addend); | |
559 | ||
560 | /* `Everything is relative'. */ | |
561 | value = sym->st_value + rela[i].r_addend; | |
562 | ||
563 | switch (ELF64_R_TYPE(rela[i].r_info)) { | |
564 | case R_PPC64_ADDR32: | |
565 | /* Simply set it */ | |
566 | *(u32 *)location = value; | |
567 | break; | |
eda09fbd | 568 | |
1da177e4 LT |
569 | case R_PPC64_ADDR64: |
570 | /* Simply set it */ | |
571 | *(unsigned long *)location = value; | |
572 | break; | |
573 | ||
574 | case R_PPC64_TOC: | |
575 | *(unsigned long *)location = my_r2(sechdrs, me); | |
576 | break; | |
577 | ||
9149ccfa | 578 | case R_PPC64_TOC16: |
f749edae | 579 | /* Subtract TOC pointer */ |
9149ccfa PB |
580 | value -= my_r2(sechdrs, me); |
581 | if (value + 0x8000 > 0xffff) { | |
c7d1f6af | 582 | pr_err("%s: bad TOC16 relocation (0x%lx)\n", |
9149ccfa PB |
583 | me->name, value); |
584 | return -ENOEXEC; | |
585 | } | |
586 | *((uint16_t *) location) | |
587 | = (*((uint16_t *) location) & ~0xffff) | |
588 | | (value & 0xffff); | |
589 | break; | |
590 | ||
1fbe9cf2 AB |
591 | case R_PPC64_TOC16_LO: |
592 | /* Subtract TOC pointer */ | |
593 | value -= my_r2(sechdrs, me); | |
594 | *((uint16_t *) location) | |
595 | = (*((uint16_t *) location) & ~0xffff) | |
596 | | (value & 0xffff); | |
597 | break; | |
598 | ||
1da177e4 | 599 | case R_PPC64_TOC16_DS: |
f749edae | 600 | /* Subtract TOC pointer */ |
1da177e4 LT |
601 | value -= my_r2(sechdrs, me); |
602 | if ((value & 3) != 0 || value + 0x8000 > 0xffff) { | |
c7d1f6af | 603 | pr_err("%s: bad TOC16_DS relocation (0x%lx)\n", |
1da177e4 LT |
604 | me->name, value); |
605 | return -ENOEXEC; | |
606 | } | |
607 | *((uint16_t *) location) | |
608 | = (*((uint16_t *) location) & ~0xfffc) | |
609 | | (value & 0xfffc); | |
610 | break; | |
611 | ||
1fbe9cf2 AB |
612 | case R_PPC64_TOC16_LO_DS: |
613 | /* Subtract TOC pointer */ | |
614 | value -= my_r2(sechdrs, me); | |
615 | if ((value & 3) != 0) { | |
c7d1f6af | 616 | pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n", |
1fbe9cf2 AB |
617 | me->name, value); |
618 | return -ENOEXEC; | |
619 | } | |
620 | *((uint16_t *) location) | |
621 | = (*((uint16_t *) location) & ~0xfffc) | |
622 | | (value & 0xfffc); | |
623 | break; | |
624 | ||
625 | case R_PPC64_TOC16_HA: | |
626 | /* Subtract TOC pointer */ | |
627 | value -= my_r2(sechdrs, me); | |
628 | value = ((value + 0x8000) >> 16); | |
629 | *((uint16_t *) location) | |
630 | = (*((uint16_t *) location) & ~0xffff) | |
631 | | (value & 0xffff); | |
632 | break; | |
633 | ||
1da177e4 LT |
634 | case R_PPC_REL24: |
635 | /* FIXME: Handle weak symbols here --RR */ | |
a443bf6e KB |
636 | if (sym->st_shndx == SHN_UNDEF || |
637 | sym->st_shndx == SHN_LIVEPATCH) { | |
1da177e4 LT |
638 | /* External: go via stub */ |
639 | value = stub_for_addr(sechdrs, value, me); | |
640 | if (!value) | |
641 | return -ENOENT; | |
250122ba NR |
642 | if (!restore_r2(strtab + sym->st_name, |
643 | (u32 *)location + 1, me)) | |
1da177e4 | 644 | return -ENOEXEC; |
15308664 TD |
645 | |
646 | squash_toc_save_inst(strtab + sym->st_name, value); | |
008d7a91 RR |
647 | } else |
648 | value += local_entry_offset(sym); | |
1da177e4 LT |
649 | |
650 | /* Convert value to relative */ | |
651 | value -= (unsigned long)location; | |
652 | if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){ | |
c7d1f6af | 653 | pr_err("%s: REL24 %li out of range!\n", |
1da177e4 LT |
654 | me->name, (long int)value); |
655 | return -ENOEXEC; | |
656 | } | |
657 | ||
658 | /* Only replace bits 2 through 26 */ | |
eda09fbd | 659 | *(uint32_t *)location |
1da177e4 LT |
660 | = (*(uint32_t *)location & ~0x03fffffc) |
661 | | (value & 0x03fffffc); | |
662 | break; | |
663 | ||
21c4ff80 BH |
664 | case R_PPC64_REL64: |
665 | /* 64 bits relative (used by features fixups) */ | |
666 | *location = value - (unsigned long)location; | |
667 | break; | |
668 | ||
9f751b82 ME |
669 | case R_PPC64_REL32: |
670 | /* 32 bits relative (used by relative exception tables) */ | |
b851ba02 NP |
671 | /* Convert value to relative */ |
672 | value -= (unsigned long)location; | |
673 | if (value + 0x80000000 > 0xffffffff) { | |
674 | pr_err("%s: REL32 %li out of range!\n", | |
675 | me->name, (long int)value); | |
676 | return -ENOEXEC; | |
677 | } | |
678 | *(u32 *)location = value; | |
9f751b82 ME |
679 | break; |
680 | ||
d247da0a RR |
681 | case R_PPC64_TOCSAVE: |
682 | /* | |
683 | * Marker reloc indicates we don't have to save r2. | |
684 | * That would only save us one instruction, so ignore | |
685 | * it. | |
686 | */ | |
687 | break; | |
688 | ||
a61674bd UW |
689 | case R_PPC64_ENTRY: |
690 | /* | |
691 | * Optimize ELFv2 large code model entry point if | |
692 | * the TOC is within 2GB range of current location. | |
693 | */ | |
694 | value = my_r2(sechdrs, me) - (unsigned long)location; | |
695 | if (value + 0x80008000 > 0xffffffff) | |
696 | break; | |
697 | /* | |
698 | * Check for the large code model prolog sequence: | |
699 | * ld r2, ...(r12) | |
700 | * add r2, r2, r12 | |
701 | */ | |
702 | if ((((uint32_t *)location)[0] & ~0xfffc) | |
703 | != 0xe84c0000) | |
704 | break; | |
705 | if (((uint32_t *)location)[1] != 0x7c426214) | |
706 | break; | |
707 | /* | |
708 | * If found, replace it with: | |
709 | * addis r2, r12, (.TOC.-func)@ha | |
710 | * addi r2, r12, (.TOC.-func)@l | |
711 | */ | |
712 | ((uint32_t *)location)[0] = 0x3c4c0000 + PPC_HA(value); | |
713 | ((uint32_t *)location)[1] = 0x38420000 + PPC_LO(value); | |
714 | break; | |
715 | ||
0906584a RR |
716 | case R_PPC64_REL16_HA: |
717 | /* Subtract location pointer */ | |
718 | value -= (unsigned long)location; | |
719 | value = ((value + 0x8000) >> 16); | |
720 | *((uint16_t *) location) | |
721 | = (*((uint16_t *) location) & ~0xffff) | |
722 | | (value & 0xffff); | |
723 | break; | |
724 | ||
725 | case R_PPC64_REL16_LO: | |
726 | /* Subtract location pointer */ | |
727 | value -= (unsigned long)location; | |
728 | *((uint16_t *) location) | |
729 | = (*((uint16_t *) location) & ~0xffff) | |
730 | | (value & 0xffff); | |
731 | break; | |
732 | ||
1da177e4 | 733 | default: |
c7d1f6af | 734 | pr_err("%s: Unknown ADD relocation: %lu\n", |
1da177e4 LT |
735 | me->name, |
736 | (unsigned long)ELF64_R_TYPE(rela[i].r_info)); | |
737 | return -ENOEXEC; | |
738 | } | |
739 | } | |
740 | ||
136cd345 ME |
741 | return 0; |
742 | } | |
743 | ||
f48cb8b4 | 744 | #ifdef CONFIG_DYNAMIC_FTRACE |
336a7b5d | 745 | |
abba7597 | 746 | #ifdef CONFIG_MPROFILE_KERNEL |
336a7b5d ME |
747 | |
748 | #define PACATOC offsetof(struct paca_struct, kernel_toc) | |
749 | ||
750 | /* | |
751 | * For mprofile-kernel we use a special stub for ftrace_caller() because we | |
752 | * can't rely on r2 containing this module's TOC when we enter the stub. | |
753 | * | |
754 | * That can happen if the function calling us didn't need to use the toc. In | |
755 | * that case it won't have setup r2, and the r2 value will be either the | |
756 | * kernel's toc, or possibly another modules toc. | |
757 | * | |
758 | * To deal with that this stub uses the kernel toc, which is always accessible | |
759 | * via the paca (in r13). The target (ftrace_caller()) is responsible for | |
760 | * saving and restoring the toc before returning. | |
761 | */ | |
ae30cc05 NR |
762 | static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, |
763 | struct module *me, unsigned long addr) | |
336a7b5d ME |
764 | { |
765 | struct ppc64_stub_entry *entry; | |
766 | unsigned int i, num_stubs; | |
767 | static u32 stub_insns[] = { | |
768 | 0xe98d0000 | PACATOC, /* ld r12,PACATOC(r13) */ | |
769 | 0x3d8c0000, /* addis r12,r12,<high> */ | |
770 | 0x398c0000, /* addi r12,r12,<low> */ | |
771 | 0x7d8903a6, /* mtctr r12 */ | |
772 | 0x4e800420, /* bctr */ | |
773 | }; | |
774 | long reladdr; | |
775 | ||
776 | num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*entry); | |
777 | ||
778 | /* Find the next available stub entry */ | |
779 | entry = (void *)sechdrs[me->arch.stubs_section].sh_addr; | |
780 | for (i = 0; i < num_stubs && stub_func_addr(entry->funcdata); i++, entry++); | |
781 | ||
782 | if (i >= num_stubs) { | |
783 | pr_err("%s: Unable to find a free slot for ftrace stub.\n", me->name); | |
784 | return 0; | |
785 | } | |
786 | ||
787 | memcpy(entry->jump, stub_insns, sizeof(stub_insns)); | |
788 | ||
789 | /* Stub uses address relative to kernel toc (from the paca) */ | |
ae30cc05 | 790 | reladdr = addr - kernel_toc_addr(); |
336a7b5d | 791 | if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) { |
ae30cc05 NR |
792 | pr_err("%s: Address of %ps out of range of kernel_toc.\n", |
793 | me->name, (void *)addr); | |
336a7b5d ME |
794 | return 0; |
795 | } | |
796 | ||
797 | entry->jump[1] |= PPC_HA(reladdr); | |
798 | entry->jump[2] |= PPC_LO(reladdr); | |
799 | ||
800 | /* Eventhough we don't use funcdata in the stub, it's needed elsewhere. */ | |
ae30cc05 | 801 | entry->funcdata = func_desc(addr); |
336a7b5d ME |
802 | entry->magic = STUB_MAGIC; |
803 | ||
804 | return (unsigned long)entry; | |
805 | } | |
806 | #else | |
ae30cc05 NR |
807 | static unsigned long create_ftrace_stub(const Elf64_Shdr *sechdrs, |
808 | struct module *me, unsigned long addr) | |
336a7b5d | 809 | { |
ae30cc05 | 810 | return stub_for_addr(sechdrs, addr, me); |
336a7b5d | 811 | } |
f48cb8b4 SR |
812 | #endif |
813 | ||
136cd345 ME |
814 | int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs) |
815 | { | |
ae30cc05 NR |
816 | mod->arch.tramp = create_ftrace_stub(sechdrs, mod, |
817 | (unsigned long)ftrace_caller); | |
818 | #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS | |
819 | mod->arch.tramp_regs = create_ftrace_stub(sechdrs, mod, | |
820 | (unsigned long)ftrace_regs_caller); | |
821 | if (!mod->arch.tramp_regs) | |
822 | return -ENOENT; | |
823 | #endif | |
136cd345 ME |
824 | |
825 | if (!mod->arch.tramp) | |
826 | return -ENOENT; | |
f48cb8b4 | 827 | |
1da177e4 LT |
828 | return 0; |
829 | } | |
136cd345 | 830 | #endif |