1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Kernel module help for PPC64.
3 Copyright (C) 2001, 2003 Rusty Russell IBM Corporation.
4
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
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>
14 #include <linux/ftrace.h>
15 #include <linux/bug.h>
16 #include <linux/uaccess.h>
17 #include <linux/kernel.h>
18 #include <asm/module.h>
19 #include <asm/firmware.h>
20 #include <asm/code-patching.h>
21 #include <linux/sort.h>
22 #include <asm/setup.h>
23 #include <asm/sections.h>
24 #include <asm/inst.h>
25
26 /* FIXME: We don't do .init separately. To do this, we'd need to have
27 a separate r2 value in the init and core section, and stub between
28 them, too.
29
30 Using a magic allocator which places modules within 32MB solves
31 this, and makes other things simpler. Anton?
32 --RR. */
33
module_elf_check_arch(Elf_Ehdr * hdr)34 bool module_elf_check_arch(Elf_Ehdr *hdr)
35 {
36 unsigned long abi_level = hdr->e_flags & 0x3;
37
38 if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
39 return abi_level == 2;
40 else
41 return abi_level < 2;
42 }
43
44 #ifdef CONFIG_PPC64_ELF_ABI_V2
45
func_desc(unsigned long addr)46 static func_desc_t func_desc(unsigned long addr)
47 {
48 func_desc_t desc = {
49 .addr = addr,
50 };
51
52 return desc;
53 }
54
55 /* PowerPC64 specific values for the Elf64_Sym st_other field. */
56 #define STO_PPC64_LOCAL_BIT 5
57 #define STO_PPC64_LOCAL_MASK (7 << STO_PPC64_LOCAL_BIT)
58 #define PPC64_LOCAL_ENTRY_OFFSET(other) \
59 (((1 << (((other) & STO_PPC64_LOCAL_MASK) >> STO_PPC64_LOCAL_BIT)) >> 2) << 2)
60
local_entry_offset(const Elf64_Sym * sym)61 static unsigned int local_entry_offset(const Elf64_Sym *sym)
62 {
63 /* sym->st_other indicates offset to local entry point
64 * (otherwise it will assume r12 is the address of the start
65 * of function and try to derive r2 from it). */
66 return PPC64_LOCAL_ENTRY_OFFSET(sym->st_other);
67 }
68 #else
69
func_desc(unsigned long addr)70 static func_desc_t func_desc(unsigned long addr)
71 {
72 return *(struct func_desc *)addr;
73 }
local_entry_offset(const Elf64_Sym * sym)74 static unsigned int local_entry_offset(const Elf64_Sym *sym)
75 {
76 return 0;
77 }
78
dereference_module_function_descriptor(struct module * mod,void * ptr)79 void *dereference_module_function_descriptor(struct module *mod, void *ptr)
80 {
81 if (ptr < (void *)mod->arch.start_opd ||
82 ptr >= (void *)mod->arch.end_opd)
83 return ptr;
84
85 return dereference_function_descriptor(ptr);
86 }
87 #endif
88
func_addr(unsigned long addr)89 static unsigned long func_addr(unsigned long addr)
90 {
91 return func_desc(addr).addr;
92 }
93
stub_func_addr(func_desc_t func)94 static unsigned long stub_func_addr(func_desc_t func)
95 {
96 return func.addr;
97 }
98
99 #define STUB_MAGIC 0x73747562 /* stub */
100
101 /* Like PPC32, we need little trampolines to do > 24-bit jumps (into
102 the kernel itself). But on PPC64, these need to be used for every
103 jump, actually, to reset r2 (TOC+0x8000). */
104 struct ppc64_stub_entry {
105 /*
106 * 28 byte jump instruction sequence (7 instructions) that can
107 * hold ppc64_stub_insns or stub_insns. Must be 8-byte aligned
108 * with PCREL kernels that use prefix instructions in the stub.
109 */
110 u32 jump[7];
111 /* Used by ftrace to identify stubs */
112 u32 magic;
113 /* Data for the above code */
114 func_desc_t funcdata;
115 } __aligned(8);
116
117 struct ppc64_got_entry {
118 u64 addr;
119 };
120
121 /*
122 * PPC64 uses 24 bit jumps, but we need to jump into other modules or
123 * the kernel which may be further. So we jump to a stub.
124 *
125 * Target address and TOC are loaded from function descriptor in the
126 * ppc64_stub_entry.
127 *
128 * r12 is used to generate the target address, which is required for the
129 * ELFv2 global entry point calling convention.
130 *
131 * TOC handling:
132 * - PCREL does not have a TOC.
133 * - ELFv2 non-PCREL just has to save r2, the callee is responsible for
134 * setting its own TOC pointer at the global entry address.
135 * - ELFv1 must load the new TOC pointer from the function descriptor.
136 */
137 static u32 ppc64_stub_insns[] = {
138 #ifdef CONFIG_PPC_KERNEL_PCREL
139 /* pld r12,addr */
140 PPC_PREFIX_8LS | __PPC_PRFX_R(1),
141 PPC_INST_PLD | ___PPC_RT(_R12),
142 #else
143 PPC_RAW_ADDIS(_R11, _R2, 0),
144 PPC_RAW_ADDI(_R11, _R11, 0),
145 /* Save current r2 value in magic place on the stack. */
146 PPC_RAW_STD(_R2, _R1, R2_STACK_OFFSET),
147 PPC_RAW_LD(_R12, _R11, 32),
148 #ifdef CONFIG_PPC64_ELF_ABI_V1
149 /* Set up new r2 from function descriptor */
150 PPC_RAW_LD(_R2, _R11, 40),
151 #endif
152 #endif
153 PPC_RAW_MTCTR(_R12),
154 PPC_RAW_BCTR(),
155 };
156
157 /*
158 * Count how many different r_type relocations (different symbol,
159 * different addend).
160 */
count_relocs(const Elf64_Rela * rela,unsigned int num,unsigned long r_type)161 static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num,
162 unsigned long r_type)
163 {
164 unsigned int i, r_info, r_addend, _count_relocs;
165
166 /* FIXME: Only count external ones --RR */
167 _count_relocs = 0;
168 r_info = 0;
169 r_addend = 0;
170 for (i = 0; i < num; i++)
171 /* Only count r_type relocs, others don't need stubs */
172 if (ELF64_R_TYPE(rela[i].r_info) == r_type &&
173 (r_info != ELF64_R_SYM(rela[i].r_info) ||
174 r_addend != rela[i].r_addend)) {
175 _count_relocs++;
176 r_info = ELF64_R_SYM(rela[i].r_info);
177 r_addend = rela[i].r_addend;
178 }
179
180 return _count_relocs;
181 }
182
relacmp(const void * _x,const void * _y)183 static int relacmp(const void *_x, const void *_y)
184 {
185 const Elf64_Rela *x, *y;
186
187 y = (Elf64_Rela *)_x;
188 x = (Elf64_Rela *)_y;
189
190 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to
191 * make the comparison cheaper/faster. It won't affect the sorting or
192 * the counting algorithms' performance
193 */
194 if (x->r_info < y->r_info)
195 return -1;
196 else if (x->r_info > y->r_info)
197 return 1;
198 else if (x->r_addend < y->r_addend)
199 return -1;
200 else if (x->r_addend > y->r_addend)
201 return 1;
202 else
203 return 0;
204 }
205
206 /* Get size of potential trampolines required. */
get_stubs_size(const Elf64_Ehdr * hdr,const Elf64_Shdr * sechdrs)207 static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
208 const Elf64_Shdr *sechdrs)
209 {
210 /* One extra reloc so it's always 0-addr terminated */
211 unsigned long relocs = 1;
212 unsigned i;
213
214 /* Every relocated section... */
215 for (i = 1; i < hdr->e_shnum; i++) {
216 if (sechdrs[i].sh_type == SHT_RELA) {
217 pr_debug("Found relocations in section %u\n", i);
218 pr_debug("Ptr: %p. Number: %Lu\n",
219 (void *)sechdrs[i].sh_addr,
220 sechdrs[i].sh_size / sizeof(Elf64_Rela));
221
222 /* Sort the relocation information based on a symbol and
223 * addend key. This is a stable O(n*log n) complexity
224 * algorithm but it will reduce the complexity of
225 * count_relocs() to linear complexity O(n)
226 */
227 sort((void *)sechdrs[i].sh_addr,
228 sechdrs[i].sh_size / sizeof(Elf64_Rela),
229 sizeof(Elf64_Rela), relacmp, NULL);
230
231 relocs += count_relocs((void *)sechdrs[i].sh_addr,
232 sechdrs[i].sh_size
233 / sizeof(Elf64_Rela),
234 R_PPC_REL24);
235 #ifdef CONFIG_PPC_KERNEL_PCREL
236 relocs += count_relocs((void *)sechdrs[i].sh_addr,
237 sechdrs[i].sh_size
238 / sizeof(Elf64_Rela),
239 R_PPC64_REL24_NOTOC);
240 #endif
241 }
242 }
243
244 #ifdef CONFIG_DYNAMIC_FTRACE
245 /* make the trampoline to the ftrace_caller */
246 relocs++;
247 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
248 /* an additional one for ftrace_regs_caller */
249 relocs++;
250 #endif
251 #endif
252
253 pr_debug("Looks like a total of %lu stubs, max\n", relocs);
254 return relocs * sizeof(struct ppc64_stub_entry);
255 }
256
257 #ifdef CONFIG_PPC_KERNEL_PCREL
count_pcpu_relocs(const Elf64_Shdr * sechdrs,const Elf64_Rela * rela,unsigned int num,unsigned int symindex,unsigned int pcpu)258 static int count_pcpu_relocs(const Elf64_Shdr *sechdrs,
259 const Elf64_Rela *rela, unsigned int num,
260 unsigned int symindex, unsigned int pcpu)
261 {
262 unsigned int i, r_info, r_addend, _count_relocs;
263
264 _count_relocs = 0;
265 r_info = 0;
266 r_addend = 0;
267
268 for (i = 0; i < num; i++) {
269 Elf64_Sym *sym;
270
271 /* This is the symbol it is referring to */
272 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
273 + ELF64_R_SYM(rela[i].r_info);
274
275 if (sym->st_shndx == pcpu &&
276 (r_info != ELF64_R_SYM(rela[i].r_info) ||
277 r_addend != rela[i].r_addend)) {
278 _count_relocs++;
279 r_info = ELF64_R_SYM(rela[i].r_info);
280 r_addend = rela[i].r_addend;
281 }
282 }
283
284 return _count_relocs;
285 }
286
287 /* Get size of potential GOT required. */
get_got_size(const Elf64_Ehdr * hdr,const Elf64_Shdr * sechdrs,struct module * me)288 static unsigned long get_got_size(const Elf64_Ehdr *hdr,
289 const Elf64_Shdr *sechdrs,
290 struct module *me)
291 {
292 /* One extra reloc so it's always 0-addr terminated */
293 unsigned long relocs = 1;
294 unsigned int i, symindex = 0;
295
296 for (i = 1; i < hdr->e_shnum; i++) {
297 if (sechdrs[i].sh_type == SHT_SYMTAB) {
298 symindex = i;
299 break;
300 }
301 }
302 WARN_ON_ONCE(!symindex);
303
304 /* Every relocated section... */
305 for (i = 1; i < hdr->e_shnum; i++) {
306 if (sechdrs[i].sh_type == SHT_RELA) {
307 pr_debug("Found relocations in section %u\n", i);
308 pr_debug("Ptr: %p. Number: %llu\n", (void *)sechdrs[i].sh_addr,
309 sechdrs[i].sh_size / sizeof(Elf64_Rela));
310
311 /*
312 * Sort the relocation information based on a symbol and
313 * addend key. This is a stable O(n*log n) complexity
314 * algorithm but it will reduce the complexity of
315 * count_relocs() to linear complexity O(n)
316 */
317 sort((void *)sechdrs[i].sh_addr,
318 sechdrs[i].sh_size / sizeof(Elf64_Rela),
319 sizeof(Elf64_Rela), relacmp, NULL);
320
321 relocs += count_relocs((void *)sechdrs[i].sh_addr,
322 sechdrs[i].sh_size
323 / sizeof(Elf64_Rela),
324 R_PPC64_GOT_PCREL34);
325
326 /*
327 * Percpu data access typically gets linked with
328 * REL34 relocations, but the percpu section gets
329 * moved at load time and requires that to be
330 * converted to GOT linkage.
331 */
332 if (IS_ENABLED(CONFIG_SMP) && symindex)
333 relocs += count_pcpu_relocs(sechdrs,
334 (void *)sechdrs[i].sh_addr,
335 sechdrs[i].sh_size
336 / sizeof(Elf64_Rela),
337 symindex, me->arch.pcpu_section);
338 }
339 }
340
341 pr_debug("Looks like a total of %lu GOT entries, max\n", relocs);
342 return relocs * sizeof(struct ppc64_got_entry);
343 }
344 #else /* CONFIG_PPC_KERNEL_PCREL */
345
346 /* Still needed for ELFv2, for .TOC. */
dedotify_versions(struct modversion_info * vers,unsigned long size)347 static void dedotify_versions(struct modversion_info *vers,
348 unsigned long size)
349 {
350 struct modversion_info *end;
351
352 for (end = (void *)vers + size; vers < end; vers++)
353 if (vers->name[0] == '.') {
354 memmove(vers->name, vers->name+1, strlen(vers->name));
355 }
356 }
357
358 /* Same as normal versions, remove a leading dot if present. */
dedotify_ext_version_names(char * str_seq,unsigned long size)359 static void dedotify_ext_version_names(char *str_seq, unsigned long size)
360 {
361 unsigned long out = 0;
362 unsigned long in;
363 char last = '\0';
364
365 for (in = 0; in < size; in++) {
366 /* Skip one leading dot */
367 if (last == '\0' && str_seq[in] == '.')
368 in++;
369 last = str_seq[in];
370 str_seq[out++] = last;
371 }
372 /* Zero the trailing portion of the names table for robustness */
373 memset(&str_seq[out], 0, size - out);
374 }
375
376 /*
377 * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC.
378 * seem to be defined (value set later).
379 */
dedotify(Elf64_Sym * syms,unsigned int numsyms,char * strtab)380 static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
381 {
382 unsigned int i;
383
384 for (i = 1; i < numsyms; i++) {
385 if (syms[i].st_shndx == SHN_UNDEF) {
386 char *name = strtab + syms[i].st_name;
387 if (name[0] == '.') {
388 if (strcmp(name+1, "TOC.") == 0)
389 syms[i].st_shndx = SHN_ABS;
390 syms[i].st_name++;
391 }
392 }
393 }
394 }
395
find_dot_toc(Elf64_Shdr * sechdrs,const char * strtab,unsigned int symindex)396 static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs,
397 const char *strtab,
398 unsigned int symindex)
399 {
400 unsigned int i, numsyms;
401 Elf64_Sym *syms;
402
403 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr;
404 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym);
405
406 for (i = 1; i < numsyms; i++) {
407 if (syms[i].st_shndx == SHN_ABS
408 && strcmp(strtab + syms[i].st_name, "TOC.") == 0)
409 return &syms[i];
410 }
411 return NULL;
412 }
413 #endif /* CONFIG_PPC_KERNEL_PCREL */
414
module_init_section(const char * name)415 bool module_init_section(const char *name)
416 {
417 /* We don't handle .init for the moment: always return false. */
418 return false;
419 }
420
module_frob_arch_sections(Elf64_Ehdr * hdr,Elf64_Shdr * sechdrs,char * secstrings,struct module * me)421 int module_frob_arch_sections(Elf64_Ehdr *hdr,
422 Elf64_Shdr *sechdrs,
423 char *secstrings,
424 struct module *me)
425 {
426 unsigned int i;
427
428 /* Find .toc and .stubs sections, symtab and strtab */
429 for (i = 1; i < hdr->e_shnum; i++) {
430 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
431 me->arch.stubs_section = i;
432 #ifdef CONFIG_PPC_KERNEL_PCREL
433 else if (strcmp(secstrings + sechdrs[i].sh_name, ".data..percpu") == 0)
434 me->arch.pcpu_section = i;
435 else if (strcmp(secstrings + sechdrs[i].sh_name, ".mygot") == 0) {
436 me->arch.got_section = i;
437 if (sechdrs[i].sh_addralign < 8)
438 sechdrs[i].sh_addralign = 8;
439 }
440 #else
441 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) {
442 me->arch.toc_section = i;
443 if (sechdrs[i].sh_addralign < 8)
444 sechdrs[i].sh_addralign = 8;
445 } else if (strcmp(secstrings + sechdrs[i].sh_name, "__versions") == 0)
446 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
447 sechdrs[i].sh_size);
448 else if (strcmp(secstrings + sechdrs[i].sh_name, "__version_ext_names") == 0)
449 dedotify_ext_version_names((void *)hdr + sechdrs[i].sh_offset,
450 sechdrs[i].sh_size);
451
452 if (sechdrs[i].sh_type == SHT_SYMTAB)
453 dedotify((void *)hdr + sechdrs[i].sh_offset,
454 sechdrs[i].sh_size / sizeof(Elf64_Sym),
455 (void *)hdr
456 + sechdrs[sechdrs[i].sh_link].sh_offset);
457 #endif
458 }
459
460 if (!me->arch.stubs_section) {
461 pr_err("%s: doesn't contain .stubs.\n", me->name);
462 return -ENOEXEC;
463 }
464
465 #ifdef CONFIG_PPC_KERNEL_PCREL
466 if (!me->arch.got_section) {
467 pr_err("%s: doesn't contain .mygot.\n", me->name);
468 return -ENOEXEC;
469 }
470
471 /* Override the got size */
472 sechdrs[me->arch.got_section].sh_size = get_got_size(hdr, sechdrs, me);
473 #else
474 /* If we don't have a .toc, just use .stubs. We need to set r2
475 to some reasonable value in case the module calls out to
476 other functions via a stub, or if a function pointer escapes
477 the module by some means. */
478 if (!me->arch.toc_section)
479 me->arch.toc_section = me->arch.stubs_section;
480 #endif
481
482 /* Override the stubs size */
483 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs);
484
485 return 0;
486 }
487
488 #if defined(CONFIG_MPROFILE_KERNEL) || defined(CONFIG_ARCH_USING_PATCHABLE_FUNCTION_ENTRY)
489
490 static u32 stub_insns[] = {
491 #ifdef CONFIG_PPC_KERNEL_PCREL
492 PPC_RAW_LD(_R12, _R13, offsetof(struct paca_struct, kernelbase)),
493 PPC_RAW_NOP(), /* align the prefix insn */
494 /* paddi r12,r12,addr */
495 PPC_PREFIX_MLS | __PPC_PRFX_R(0),
496 PPC_INST_PADDI | ___PPC_RT(_R12) | ___PPC_RA(_R12),
497 PPC_RAW_MTCTR(_R12),
498 PPC_RAW_BCTR(),
499 #else
500 PPC_RAW_LD(_R12, _R13, offsetof(struct paca_struct, kernel_toc)),
501 PPC_RAW_ADDIS(_R12, _R12, 0),
502 PPC_RAW_ADDI(_R12, _R12, 0),
503 PPC_RAW_MTCTR(_R12),
504 PPC_RAW_BCTR(),
505 #endif
506 };
507
508 /*
509 * For mprofile-kernel we use a special stub for ftrace_caller() because we
510 * can't rely on r2 containing this module's TOC when we enter the stub.
511 *
512 * That can happen if the function calling us didn't need to use the toc. In
513 * that case it won't have setup r2, and the r2 value will be either the
514 * kernel's toc, or possibly another modules toc.
515 *
516 * To deal with that this stub uses the kernel toc, which is always accessible
517 * via the paca (in r13). The target (ftrace_caller()) is responsible for
518 * saving and restoring the toc before returning.
519 */
create_ftrace_stub(struct ppc64_stub_entry * entry,unsigned long addr,struct module * me)520 static inline int create_ftrace_stub(struct ppc64_stub_entry *entry,
521 unsigned long addr,
522 struct module *me)
523 {
524 long reladdr;
525
526 if ((unsigned long)entry->jump % 8 != 0) {
527 pr_err("%s: Address of stub entry is not 8-byte aligned\n", me->name);
528 return 0;
529 }
530
531 BUILD_BUG_ON(sizeof(stub_insns) > sizeof(entry->jump));
532 memcpy(entry->jump, stub_insns, sizeof(stub_insns));
533
534 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) {
535 /* Stub uses address relative to kernel base (from the paca) */
536 reladdr = addr - local_paca->kernelbase;
537 if (reladdr > 0x1FFFFFFFFL || reladdr < -0x200000000L) {
538 pr_err("%s: Address of %ps out of range of 34-bit relative address.\n",
539 me->name, (void *)addr);
540 return 0;
541 }
542
543 entry->jump[2] |= IMM_H18(reladdr);
544 entry->jump[3] |= IMM_L(reladdr);
545 } else {
546 /* Stub uses address relative to kernel toc (from the paca) */
547 reladdr = addr - kernel_toc_addr();
548 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
549 pr_err("%s: Address of %ps out of range of kernel_toc.\n",
550 me->name, (void *)addr);
551 return 0;
552 }
553
554 entry->jump[1] |= PPC_HA(reladdr);
555 entry->jump[2] |= PPC_LO(reladdr);
556 }
557
558 /* Even though we don't use funcdata in the stub, it's needed elsewhere. */
559 entry->funcdata = func_desc(addr);
560 entry->magic = STUB_MAGIC;
561
562 return 1;
563 }
564
is_mprofile_ftrace_call(const char * name)565 static bool is_mprofile_ftrace_call(const char *name)
566 {
567 if (!strcmp("_mcount", name))
568 return true;
569 #ifdef CONFIG_DYNAMIC_FTRACE
570 if (!strcmp("ftrace_caller", name))
571 return true;
572 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
573 if (!strcmp("ftrace_regs_caller", name))
574 return true;
575 #endif
576 #endif
577
578 return false;
579 }
580 #else
create_ftrace_stub(struct ppc64_stub_entry * entry,unsigned long addr,struct module * me)581 static inline int create_ftrace_stub(struct ppc64_stub_entry *entry,
582 unsigned long addr,
583 struct module *me)
584 {
585 return 0;
586 }
587
is_mprofile_ftrace_call(const char * name)588 static bool is_mprofile_ftrace_call(const char *name)
589 {
590 return false;
591 }
592 #endif
593
594 /*
595 * r2 is the TOC pointer: it actually points 0x8000 into the TOC (this gives the
596 * value maximum span in an instruction which uses a signed offset). Round down
597 * to a 256 byte boundary for the odd case where we are setting up r2 without a
598 * .toc section.
599 */
my_r2(const Elf64_Shdr * sechdrs,struct module * me)600 static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me)
601 {
602 #ifndef CONFIG_PPC_KERNEL_PCREL
603 return (sechdrs[me->arch.toc_section].sh_addr & ~0xfful) + 0x8000;
604 #else
605 return -1;
606 #endif
607 }
608
609 /* Patch stub to reference function and correct r2 value. */
create_stub(const Elf64_Shdr * sechdrs,struct ppc64_stub_entry * entry,unsigned long addr,struct module * me,const char * name)610 static inline int create_stub(const Elf64_Shdr *sechdrs,
611 struct ppc64_stub_entry *entry,
612 unsigned long addr,
613 struct module *me,
614 const char *name)
615 {
616 long reladdr;
617 func_desc_t desc;
618 int i;
619
620 if (is_mprofile_ftrace_call(name))
621 return create_ftrace_stub(entry, addr, me);
622
623 if ((unsigned long)entry->jump % 8 != 0) {
624 pr_err("%s: Address of stub entry is not 8-byte aligned\n", me->name);
625 return 0;
626 }
627
628 BUILD_BUG_ON(sizeof(ppc64_stub_insns) > sizeof(entry->jump));
629 for (i = 0; i < ARRAY_SIZE(ppc64_stub_insns); i++) {
630 if (patch_instruction(&entry->jump[i],
631 ppc_inst(ppc64_stub_insns[i])))
632 return 0;
633 }
634
635 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) {
636 /* Stub uses address relative to itself! */
637 reladdr = 0 + offsetof(struct ppc64_stub_entry, funcdata);
638 BUILD_BUG_ON(reladdr != 32);
639 if (reladdr > 0x1FFFFFFFFL || reladdr < -0x200000000L) {
640 pr_err("%s: Address of %p out of range of 34-bit relative address.\n",
641 me->name, (void *)reladdr);
642 return 0;
643 }
644 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr);
645
646 /* May not even need this if we're relative to 0 */
647 if (patch_instruction(&entry->jump[0],
648 ppc_inst_prefix(entry->jump[0] | IMM_H18(reladdr),
649 entry->jump[1] | IMM_L(reladdr))))
650 return 0;
651
652 } else {
653 /* Stub uses address relative to r2. */
654 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
655 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
656 pr_err("%s: Address %p of stub out of range of %p.\n",
657 me->name, (void *)reladdr, (void *)my_r2);
658 return 0;
659 }
660 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr);
661
662 if (patch_instruction(&entry->jump[0],
663 ppc_inst(entry->jump[0] | PPC_HA(reladdr))))
664 return 0;
665
666 if (patch_instruction(&entry->jump[1],
667 ppc_inst(entry->jump[1] | PPC_LO(reladdr))))
668 return 0;
669 }
670
671 // func_desc_t is 8 bytes if ABIv2, else 16 bytes
672 desc = func_desc(addr);
673 for (i = 0; i < sizeof(func_desc_t) / sizeof(u32); i++) {
674 if (patch_u32(((u32 *)&entry->funcdata) + i, ((u32 *)&desc)[i]))
675 return 0;
676 }
677
678 if (patch_u32(&entry->magic, STUB_MAGIC))
679 return 0;
680
681 return 1;
682 }
683
684 /* Create stub to jump to function described in this OPD/ptr: we need the
685 stub to set up the TOC ptr (r2) for the function. */
stub_for_addr(const Elf64_Shdr * sechdrs,unsigned long addr,struct module * me,const char * name)686 static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs,
687 unsigned long addr,
688 struct module *me,
689 const char *name)
690 {
691 struct ppc64_stub_entry *stubs;
692 unsigned int i, num_stubs;
693
694 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
695
696 /* Find this stub, or if that fails, the next avail. entry */
697 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
698 for (i = 0; stub_func_addr(stubs[i].funcdata); i++) {
699 if (WARN_ON(i >= num_stubs))
700 return 0;
701
702 if (stub_func_addr(stubs[i].funcdata) == func_addr(addr))
703 return (unsigned long)&stubs[i];
704 }
705
706 if (!create_stub(sechdrs, &stubs[i], addr, me, name))
707 return 0;
708
709 return (unsigned long)&stubs[i];
710 }
711
712 #ifdef CONFIG_PPC_KERNEL_PCREL
713 /* Create GOT to load the location described in this ptr */
got_for_addr(const Elf64_Shdr * sechdrs,unsigned long addr,struct module * me,const char * name)714 static unsigned long got_for_addr(const Elf64_Shdr *sechdrs,
715 unsigned long addr,
716 struct module *me,
717 const char *name)
718 {
719 struct ppc64_got_entry *got;
720 unsigned int i, num_got;
721
722 if (!IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
723 return addr;
724
725 num_got = sechdrs[me->arch.got_section].sh_size / sizeof(*got);
726
727 /* Find this stub, or if that fails, the next avail. entry */
728 got = (void *)sechdrs[me->arch.got_section].sh_addr;
729 for (i = 0; got[i].addr; i++) {
730 if (WARN_ON(i >= num_got))
731 return 0;
732
733 if (got[i].addr == addr)
734 return (unsigned long)&got[i];
735 }
736
737 got[i].addr = addr;
738
739 return (unsigned long)&got[i];
740 }
741 #endif
742
743 /* We expect a noop next: if it is, replace it with instruction to
744 restore r2. */
restore_r2(const char * name,u32 * instruction,struct module * me)745 static int restore_r2(const char *name, u32 *instruction, struct module *me)
746 {
747 u32 *prev_insn = instruction - 1;
748 u32 insn_val = *instruction;
749
750 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
751 return 0;
752
753 if (is_mprofile_ftrace_call(name))
754 return 0;
755
756 /*
757 * Make sure the branch isn't a sibling call. Sibling calls aren't
758 * "link" branches and they don't return, so they don't need the r2
759 * restore afterwards.
760 */
761 if (!instr_is_relative_link_branch(ppc_inst(*prev_insn)))
762 return 0;
763
764 /*
765 * For livepatch, the restore r2 instruction might have already been
766 * written previously, if the referenced symbol is in a previously
767 * unloaded module which is now being loaded again. In that case, skip
768 * the warning and the instruction write.
769 */
770 if (insn_val == PPC_INST_LD_TOC)
771 return 0;
772
773 if (insn_val != PPC_RAW_NOP()) {
774 pr_err("%s: Expected nop after call, got %08x at %pS\n",
775 me->name, insn_val, instruction);
776 return -ENOEXEC;
777 }
778
779 /* ld r2,R2_STACK_OFFSET(r1) */
780 return patch_instruction(instruction, ppc_inst(PPC_INST_LD_TOC));
781 }
782
apply_relocate_add(Elf64_Shdr * sechdrs,const char * strtab,unsigned int symindex,unsigned int relsec,struct module * me)783 int apply_relocate_add(Elf64_Shdr *sechdrs,
784 const char *strtab,
785 unsigned int symindex,
786 unsigned int relsec,
787 struct module *me)
788 {
789 unsigned int i;
790 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
791 Elf64_Sym *sym;
792 unsigned long *location;
793 unsigned long value;
794
795 pr_debug("Applying ADD relocate section %u to %u\n", relsec,
796 sechdrs[relsec].sh_info);
797
798 #ifndef CONFIG_PPC_KERNEL_PCREL
799 /* First time we're called, we can fix up .TOC. */
800 if (!me->arch.toc_fixed) {
801 sym = find_dot_toc(sechdrs, strtab, symindex);
802 /* It's theoretically possible that a module doesn't want a
803 * .TOC. so don't fail it just for that. */
804 if (sym)
805 sym->st_value = my_r2(sechdrs, me);
806 me->arch.toc_fixed = true;
807 }
808 #endif
809 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
810 /* This is where to make the change */
811 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
812 + rela[i].r_offset;
813 /* This is the symbol it is referring to */
814 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
815 + ELF64_R_SYM(rela[i].r_info);
816
817 pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n",
818 location, (long)ELF64_R_TYPE(rela[i].r_info),
819 strtab + sym->st_name, (unsigned long)sym->st_value,
820 (long)rela[i].r_addend);
821
822 /* `Everything is relative'. */
823 value = sym->st_value + rela[i].r_addend;
824
825 switch (ELF64_R_TYPE(rela[i].r_info)) {
826 case R_PPC64_ADDR32:
827 /* Simply set it */
828 *(u32 *)location = value;
829 break;
830
831 case R_PPC64_ADDR64:
832 /* Simply set it */
833 *(unsigned long *)location = value;
834 break;
835
836 #ifndef CONFIG_PPC_KERNEL_PCREL
837 case R_PPC64_TOC:
838 *(unsigned long *)location = my_r2(sechdrs, me);
839 break;
840
841 case R_PPC64_TOC16:
842 /* Subtract TOC pointer */
843 value -= my_r2(sechdrs, me);
844 if (value + 0x8000 > 0xffff) {
845 pr_err("%s: bad TOC16 relocation (0x%lx)\n",
846 me->name, value);
847 return -ENOEXEC;
848 }
849 *((uint16_t *) location)
850 = (*((uint16_t *) location) & ~0xffff)
851 | (value & 0xffff);
852 break;
853
854 case R_PPC64_TOC16_LO:
855 /* Subtract TOC pointer */
856 value -= my_r2(sechdrs, me);
857 *((uint16_t *) location)
858 = (*((uint16_t *) location) & ~0xffff)
859 | (value & 0xffff);
860 break;
861
862 case R_PPC64_TOC16_DS:
863 /* Subtract TOC pointer */
864 value -= my_r2(sechdrs, me);
865 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
866 pr_err("%s: bad TOC16_DS relocation (0x%lx)\n",
867 me->name, value);
868 return -ENOEXEC;
869 }
870 *((uint16_t *) location)
871 = (*((uint16_t *) location) & ~0xfffc)
872 | (value & 0xfffc);
873 break;
874
875 case R_PPC64_TOC16_LO_DS:
876 /* Subtract TOC pointer */
877 value -= my_r2(sechdrs, me);
878 if ((value & 3) != 0) {
879 pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n",
880 me->name, value);
881 return -ENOEXEC;
882 }
883 *((uint16_t *) location)
884 = (*((uint16_t *) location) & ~0xfffc)
885 | (value & 0xfffc);
886 break;
887
888 case R_PPC64_TOC16_HA:
889 /* Subtract TOC pointer */
890 value -= my_r2(sechdrs, me);
891 value = ((value + 0x8000) >> 16);
892 *((uint16_t *) location)
893 = (*((uint16_t *) location) & ~0xffff)
894 | (value & 0xffff);
895 break;
896 #endif
897
898 case R_PPC_REL24:
899 #ifdef CONFIG_PPC_KERNEL_PCREL
900 /* PCREL still generates REL24 for mcount */
901 case R_PPC64_REL24_NOTOC:
902 #endif
903 /* FIXME: Handle weak symbols here --RR */
904 if (sym->st_shndx == SHN_UNDEF ||
905 sym->st_shndx == SHN_LIVEPATCH) {
906 /* External: go via stub */
907 value = stub_for_addr(sechdrs, value, me,
908 strtab + sym->st_name);
909 if (!value)
910 return -ENOENT;
911 if (restore_r2(strtab + sym->st_name,
912 (u32 *)location + 1, me))
913 return -ENOEXEC;
914 } else
915 value += local_entry_offset(sym);
916
917 /* Convert value to relative */
918 value -= (unsigned long)location;
919 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
920 pr_err("%s: REL24 %li out of range!\n",
921 me->name, (long int)value);
922 return -ENOEXEC;
923 }
924
925 /* Only replace bits 2 through 26 */
926 value = (*(uint32_t *)location & ~PPC_LI_MASK) | PPC_LI(value);
927
928 if (patch_instruction((u32 *)location, ppc_inst(value)))
929 return -EFAULT;
930
931 break;
932
933 case R_PPC64_REL64:
934 /* 64 bits relative (used by features fixups) */
935 *location = value - (unsigned long)location;
936 break;
937
938 case R_PPC64_REL32:
939 /* 32 bits relative (used by relative exception tables) */
940 /* Convert value to relative */
941 value -= (unsigned long)location;
942 if (value + 0x80000000 > 0xffffffff) {
943 pr_err("%s: REL32 %li out of range!\n",
944 me->name, (long int)value);
945 return -ENOEXEC;
946 }
947 *(u32 *)location = value;
948 break;
949
950 #ifdef CONFIG_PPC_KERNEL_PCREL
951 case R_PPC64_PCREL34: {
952 unsigned long absvalue = value;
953
954 /* Convert value to relative */
955 value -= (unsigned long)location;
956
957 if (value + 0x200000000 > 0x3ffffffff) {
958 if (sym->st_shndx != me->arch.pcpu_section) {
959 pr_err("%s: REL34 %li out of range!\n",
960 me->name, (long)value);
961 return -ENOEXEC;
962 }
963
964 /*
965 * per-cpu section is special cased because
966 * it is moved during loading, so has to be
967 * converted to use GOT.
968 */
969 value = got_for_addr(sechdrs, absvalue, me,
970 strtab + sym->st_name);
971 if (!value)
972 return -ENOENT;
973 value -= (unsigned long)location;
974
975 /* Turn pla into pld */
976 if (patch_instruction((u32 *)location,
977 ppc_inst_prefix((*(u32 *)location & ~0x02000000),
978 (*((u32 *)location + 1) & ~0xf8000000) | 0xe4000000)))
979 return -EFAULT;
980 }
981
982 if (patch_instruction((u32 *)location,
983 ppc_inst_prefix((*(u32 *)location & ~0x3ffff) | IMM_H18(value),
984 (*((u32 *)location + 1) & ~0xffff) | IMM_L(value))))
985 return -EFAULT;
986
987 break;
988 }
989
990 #else
991 case R_PPC64_TOCSAVE:
992 /*
993 * Marker reloc indicates we don't have to save r2.
994 * That would only save us one instruction, so ignore
995 * it.
996 */
997 break;
998 #endif
999
1000 case R_PPC64_ENTRY:
1001 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
1002 break;
1003
1004 /*
1005 * Optimize ELFv2 large code model entry point if
1006 * the TOC is within 2GB range of current location.
1007 */
1008 value = my_r2(sechdrs, me) - (unsigned long)location;
1009 if (value + 0x80008000 > 0xffffffff)
1010 break;
1011 /*
1012 * Check for the large code model prolog sequence:
1013 * ld r2, ...(r12)
1014 * add r2, r2, r12
1015 */
1016 if ((((uint32_t *)location)[0] & ~0xfffc) != PPC_RAW_LD(_R2, _R12, 0))
1017 break;
1018 if (((uint32_t *)location)[1] != PPC_RAW_ADD(_R2, _R2, _R12))
1019 break;
1020 /*
1021 * If found, replace it with:
1022 * addis r2, r12, (.TOC.-func)@ha
1023 * addi r2, r2, (.TOC.-func)@l
1024 */
1025 ((uint32_t *)location)[0] = PPC_RAW_ADDIS(_R2, _R12, PPC_HA(value));
1026 ((uint32_t *)location)[1] = PPC_RAW_ADDI(_R2, _R2, PPC_LO(value));
1027 break;
1028
1029 case R_PPC64_REL16_HA:
1030 /* Subtract location pointer */
1031 value -= (unsigned long)location;
1032 value = ((value + 0x8000) >> 16);
1033 *((uint16_t *) location)
1034 = (*((uint16_t *) location) & ~0xffff)
1035 | (value & 0xffff);
1036 break;
1037
1038 case R_PPC64_REL16_LO:
1039 /* Subtract location pointer */
1040 value -= (unsigned long)location;
1041 *((uint16_t *) location)
1042 = (*((uint16_t *) location) & ~0xffff)
1043 | (value & 0xffff);
1044 break;
1045
1046 #ifdef CONFIG_PPC_KERNEL_PCREL
1047 case R_PPC64_GOT_PCREL34:
1048 value = got_for_addr(sechdrs, value, me,
1049 strtab + sym->st_name);
1050 if (!value)
1051 return -ENOENT;
1052 value -= (unsigned long)location;
1053 ((uint32_t *)location)[0] = (((uint32_t *)location)[0] & ~0x3ffff) |
1054 ((value >> 16) & 0x3ffff);
1055 ((uint32_t *)location)[1] = (((uint32_t *)location)[1] & ~0xffff) |
1056 (value & 0xffff);
1057 break;
1058 #endif
1059
1060 default:
1061 pr_err("%s: Unknown ADD relocation: %lu\n",
1062 me->name,
1063 (unsigned long)ELF64_R_TYPE(rela[i].r_info));
1064 return -ENOEXEC;
1065 }
1066 }
1067
1068 return 0;
1069 }
1070
1071 #ifdef CONFIG_DYNAMIC_FTRACE
module_trampoline_target(struct module * mod,unsigned long addr,unsigned long * target)1072 int module_trampoline_target(struct module *mod, unsigned long addr,
1073 unsigned long *target)
1074 {
1075 struct ppc64_stub_entry *stub;
1076 func_desc_t funcdata;
1077 u32 magic;
1078
1079 if (!within_module_core(addr, mod)) {
1080 pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name);
1081 return -EFAULT;
1082 }
1083
1084 stub = (struct ppc64_stub_entry *)addr;
1085
1086 if (copy_from_kernel_nofault(&magic, &stub->magic,
1087 sizeof(magic))) {
1088 pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name);
1089 return -EFAULT;
1090 }
1091
1092 if (magic != STUB_MAGIC) {
1093 pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name);
1094 return -EFAULT;
1095 }
1096
1097 if (copy_from_kernel_nofault(&funcdata, &stub->funcdata,
1098 sizeof(funcdata))) {
1099 pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name);
1100 return -EFAULT;
1101 }
1102
1103 *target = stub_func_addr(funcdata);
1104
1105 return 0;
1106 }
1107
module_finalize_ftrace(struct module * mod,const Elf_Shdr * sechdrs)1108 int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs)
1109 {
1110 mod->arch.tramp = stub_for_addr(sechdrs,
1111 (unsigned long)ftrace_caller,
1112 mod,
1113 "ftrace_caller");
1114 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1115 mod->arch.tramp_regs = stub_for_addr(sechdrs,
1116 (unsigned long)ftrace_regs_caller,
1117 mod,
1118 "ftrace_regs_caller");
1119 if (!mod->arch.tramp_regs)
1120 return -ENOENT;
1121 #endif
1122
1123 if (!mod->arch.tramp)
1124 return -ENOENT;
1125
1126 return 0;
1127 }
1128 #endif
1129