projects/linux·blob master · refs

1162 lines · 31.5 KiB

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/text-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
34bool 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
46static 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
61static 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
70static func_desc_t func_desc(unsigned long addr)
71{
72 return *(struct func_desc *)addr;
73}
74static unsigned int local_entry_offset(const Elf64_Sym *sym)
75{
76 return 0;
77}
78
79void *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
89static unsigned long func_addr(unsigned long addr)
90{
91 return func_desc(addr).addr;
92}
93
94static 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). */
104struct 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
117struct 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 */
137static 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 */
161static 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
183static 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. */
207static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
208 const Elf64_Shdr *sechdrs,
209 char *secstrings,
210 struct module *me)
211{
212 unsigned long relocs = 0;
213 unsigned i;
214
215 /* Every relocated section... */
216 for (i = 1; i < hdr->e_shnum; i++) {
217 if (sechdrs[i].sh_type == SHT_RELA) {
218 pr_debug("Found relocations in section %u\n", i);
219 pr_debug("Ptr: %p. Number: %Lu\n",
220 (void *)sechdrs[i].sh_addr,
221 sechdrs[i].sh_size / sizeof(Elf64_Rela));
222
223 /* Sort the relocation information based on a symbol and
224 * addend key. This is a stable O(n*log n) complexity
225 * algorithm but it will reduce the complexity of
226 * count_relocs() to linear complexity O(n)
227 */
228 sort((void *)sechdrs[i].sh_addr,
229 sechdrs[i].sh_size / sizeof(Elf64_Rela),
230 sizeof(Elf64_Rela), relacmp, NULL);
231
232 relocs += count_relocs((void *)sechdrs[i].sh_addr,
233 sechdrs[i].sh_size
234 / sizeof(Elf64_Rela),
235 R_PPC_REL24);
236#ifdef CONFIG_PPC_KERNEL_PCREL
237 relocs += count_relocs((void *)sechdrs[i].sh_addr,
238 sechdrs[i].sh_size
239 / sizeof(Elf64_Rela),
240 R_PPC64_REL24_NOTOC);
241#endif
242 }
243 }
244
245 /* stubs for ftrace_caller and ftrace_regs_caller */
246 relocs += IS_ENABLED(CONFIG_DYNAMIC_FTRACE) + IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_REGS);
247
248#ifdef CONFIG_PPC_FTRACE_OUT_OF_LINE
249 /* stubs for the function tracer */
250 for (i = 1; i < hdr->e_shnum; i++) {
251 if (!strcmp(secstrings + sechdrs[i].sh_name, "__patchable_function_entries")) {
252 me->arch.ool_stub_count = sechdrs[i].sh_size / sizeof(unsigned long);
253 me->arch.ool_stub_index = 0;
254 relocs += roundup(me->arch.ool_stub_count * sizeof(struct ftrace_ool_stub),
255 sizeof(struct ppc64_stub_entry)) /
256 sizeof(struct ppc64_stub_entry);
257 break;
258 }
259 }
260#endif
261
262 pr_debug("Looks like a total of %lu stubs, max\n", relocs);
263 return relocs * sizeof(struct ppc64_stub_entry);
264}
265
266#ifdef CONFIG_PPC_KERNEL_PCREL
267static int count_pcpu_relocs(const Elf64_Shdr *sechdrs,
268 const Elf64_Rela *rela, unsigned int num,
269 unsigned int symindex, unsigned int pcpu)
270{
271 unsigned int i, r_info, r_addend, _count_relocs;
272
273 _count_relocs = 0;
274 r_info = 0;
275 r_addend = 0;
276
277 for (i = 0; i < num; i++) {
278 Elf64_Sym *sym;
279
280 /* This is the symbol it is referring to */
281 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
282 + ELF64_R_SYM(rela[i].r_info);
283
284 if (sym->st_shndx == pcpu &&
285 (r_info != ELF64_R_SYM(rela[i].r_info) ||
286 r_addend != rela[i].r_addend)) {
287 _count_relocs++;
288 r_info = ELF64_R_SYM(rela[i].r_info);
289 r_addend = rela[i].r_addend;
290 }
291 }
292
293 return _count_relocs;
294}
295
296/* Get size of potential GOT required. */
297static unsigned long get_got_size(const Elf64_Ehdr *hdr,
298 const Elf64_Shdr *sechdrs,
299 struct module *me)
300{
301 /* One extra reloc so it's always 0-addr terminated */
302 unsigned long relocs = 1;
303 unsigned int i, symindex = 0;
304
305 for (i = 1; i < hdr->e_shnum; i++) {
306 if (sechdrs[i].sh_type == SHT_SYMTAB) {
307 symindex = i;
308 break;
309 }
310 }
311 WARN_ON_ONCE(!symindex);
312
313 /* Every relocated section... */
314 for (i = 1; i < hdr->e_shnum; i++) {
315 if (sechdrs[i].sh_type == SHT_RELA) {
316 pr_debug("Found relocations in section %u\n", i);
317 pr_debug("Ptr: %p. Number: %llu\n", (void *)sechdrs[i].sh_addr,
318 sechdrs[i].sh_size / sizeof(Elf64_Rela));
319
320 /*
321 * Sort the relocation information based on a symbol and
322 * addend key. This is a stable O(n*log n) complexity
323 * algorithm but it will reduce the complexity of
324 * count_relocs() to linear complexity O(n)
325 */
326 sort((void *)sechdrs[i].sh_addr,
327 sechdrs[i].sh_size / sizeof(Elf64_Rela),
328 sizeof(Elf64_Rela), relacmp, NULL);
329
330 relocs += count_relocs((void *)sechdrs[i].sh_addr,
331 sechdrs[i].sh_size
332 / sizeof(Elf64_Rela),
333 R_PPC64_GOT_PCREL34);
334
335 /*
336 * Percpu data access typically gets linked with
337 * REL34 relocations, but the percpu section gets
338 * moved at load time and requires that to be
339 * converted to GOT linkage.
340 */
341 if (IS_ENABLED(CONFIG_SMP) && symindex)
342 relocs += count_pcpu_relocs(sechdrs,
343 (void *)sechdrs[i].sh_addr,
344 sechdrs[i].sh_size
345 / sizeof(Elf64_Rela),
346 symindex, me->arch.pcpu_section);
347 }
348 }
349
350 pr_debug("Looks like a total of %lu GOT entries, max\n", relocs);
351 return relocs * sizeof(struct ppc64_got_entry);
352}
353#else /* CONFIG_PPC_KERNEL_PCREL */
354
355/* Still needed for ELFv2, for .TOC. */
356static void dedotify_versions(struct modversion_info *vers,
357 unsigned long size)
358{
359 struct modversion_info *end;
360
361 for (end = (void *)vers + size; vers < end; vers++)
362 if (vers->name[0] == '.') {
363 memmove(vers->name, vers->name+1, strlen(vers->name));
364 }
365}
366
367/* Same as normal versions, remove a leading dot if present. */
368static void dedotify_ext_version_names(char *str_seq, unsigned long size)
369{
370 unsigned long out = 0;
371 unsigned long in;
372 char last = '\0';
373
374 for (in = 0; in < size; in++) {
375 /* Skip one leading dot */
376 if (last == '\0' && str_seq[in] == '.')
377 in++;
378 last = str_seq[in];
379 str_seq[out++] = last;
380 }
381 /* Zero the trailing portion of the names table for robustness */
382 memset(&str_seq[out], 0, size - out);
383}
384
385/*
386 * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC.
387 * seem to be defined (value set later).
388 */
389static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
390{
391 unsigned int i;
392
393 for (i = 1; i < numsyms; i++) {
394 if (syms[i].st_shndx == SHN_UNDEF) {
395 char *name = strtab + syms[i].st_name;
396 if (name[0] == '.') {
397 if (strcmp(name+1, "TOC.") == 0)
398 syms[i].st_shndx = SHN_ABS;
399 syms[i].st_name++;
400 }
401 }
402 }
403}
404
405static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs,
406 const char *strtab,
407 unsigned int symindex)
408{
409 unsigned int i, numsyms;
410 Elf64_Sym *syms;
411
412 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr;
413 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym);
414
415 for (i = 1; i < numsyms; i++) {
416 if (syms[i].st_shndx == SHN_ABS
417 && strcmp(strtab + syms[i].st_name, "TOC.") == 0)
418 return &syms[i];
419 }
420 return NULL;
421}
422#endif /* CONFIG_PPC_KERNEL_PCREL */
423
424bool module_init_section(const char *name)
425{
426 /* We don't handle .init for the moment: always return false. */
427 return false;
428}
429
430int module_frob_arch_sections(Elf64_Ehdr *hdr,
431 Elf64_Shdr *sechdrs,
432 char *secstrings,
433 struct module *me)
434{
435 unsigned int i;
436
437 /* Find .toc and .stubs sections, symtab and strtab */
438 for (i = 1; i < hdr->e_shnum; i++) {
439 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
440 me->arch.stubs_section = i;
441#ifdef CONFIG_PPC_KERNEL_PCREL
442 else if (strcmp(secstrings + sechdrs[i].sh_name, ".data..percpu") == 0)
443 me->arch.pcpu_section = i;
444 else if (strcmp(secstrings + sechdrs[i].sh_name, ".mygot") == 0) {
445 me->arch.got_section = i;
446 if (sechdrs[i].sh_addralign < 8)
447 sechdrs[i].sh_addralign = 8;
448 }
449#else
450 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) {
451 me->arch.toc_section = i;
452 if (sechdrs[i].sh_addralign < 8)
453 sechdrs[i].sh_addralign = 8;
454 } else if (strcmp(secstrings + sechdrs[i].sh_name, "__versions") == 0)
455 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
456 sechdrs[i].sh_size);
457 else if (strcmp(secstrings + sechdrs[i].sh_name, "__version_ext_names") == 0)
458 dedotify_ext_version_names((void *)hdr + sechdrs[i].sh_offset,
459 sechdrs[i].sh_size);
460
461 if (sechdrs[i].sh_type == SHT_SYMTAB)
462 dedotify((void *)hdr + sechdrs[i].sh_offset,
463 sechdrs[i].sh_size / sizeof(Elf64_Sym),
464 (void *)hdr
465 + sechdrs[sechdrs[i].sh_link].sh_offset);
466#endif
467 }
468
469 if (!me->arch.stubs_section) {
470 pr_err("%s: doesn't contain .stubs.\n", me->name);
471 return -ENOEXEC;
472 }
473
474#ifdef CONFIG_PPC_KERNEL_PCREL
475 if (!me->arch.got_section) {
476 pr_err("%s: doesn't contain .mygot.\n", me->name);
477 return -ENOEXEC;
478 }
479
480 /* Override the got size */
481 sechdrs[me->arch.got_section].sh_size = get_got_size(hdr, sechdrs, me);
482#else
483 /* If we don't have a .toc, just use .stubs. We need to set r2
484 to some reasonable value in case the module calls out to
485 other functions via a stub, or if a function pointer escapes
486 the module by some means. */
487 if (!me->arch.toc_section)
488 me->arch.toc_section = me->arch.stubs_section;
489#endif
490
491 /* Override the stubs size */
492 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs, secstrings, me);
493
494 return 0;
495}
496
497#if defined(CONFIG_MPROFILE_KERNEL) || defined(CONFIG_ARCH_USING_PATCHABLE_FUNCTION_ENTRY)
498
499static u32 stub_insns[] = {
500#ifdef CONFIG_PPC_KERNEL_PCREL
501 PPC_RAW_LD(_R12, _R13, offsetof(struct paca_struct, kernelbase)),
502 PPC_RAW_NOP(), /* align the prefix insn */
503 /* paddi r12,r12,addr */
504 PPC_PREFIX_MLS | __PPC_PRFX_R(0),
505 PPC_INST_PADDI | ___PPC_RT(_R12) | ___PPC_RA(_R12),
506 PPC_RAW_MTCTR(_R12),
507 PPC_RAW_BCTR(),
508#else
509 PPC_RAW_LD(_R12, _R13, offsetof(struct paca_struct, kernel_toc)),
510 PPC_RAW_ADDIS(_R12, _R12, 0),
511 PPC_RAW_ADDI(_R12, _R12, 0),
512 PPC_RAW_MTCTR(_R12),
513 PPC_RAW_BCTR(),
514#endif
515};
516
517/*
518 * For mprofile-kernel we use a special stub for ftrace_caller() because we
519 * can't rely on r2 containing this module's TOC when we enter the stub.
520 *
521 * That can happen if the function calling us didn't need to use the toc. In
522 * that case it won't have setup r2, and the r2 value will be either the
523 * kernel's toc, or possibly another modules toc.
524 *
525 * To deal with that this stub uses the kernel toc, which is always accessible
526 * via the paca (in r13). The target (ftrace_caller()) is responsible for
527 * saving and restoring the toc before returning.
528 */
529static inline int create_ftrace_stub(struct ppc64_stub_entry *entry,
530 unsigned long addr,
531 struct module *me)
532{
533 long reladdr;
534
535 if ((unsigned long)entry->jump % 8 != 0) {
536 pr_err("%s: Address of stub entry is not 8-byte aligned\n", me->name);
537 return 0;
538 }
539
540 BUILD_BUG_ON(sizeof(stub_insns) > sizeof(entry->jump));
541 memcpy(entry->jump, stub_insns, sizeof(stub_insns));
542
543 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) {
544 /* Stub uses address relative to kernel base (from the paca) */
545 reladdr = addr - local_paca->kernelbase;
546 if (reladdr > 0x1FFFFFFFFL || reladdr < -0x200000000L) {
547 pr_err("%s: Address of %ps out of range of 34-bit relative address.\n",
548 me->name, (void *)addr);
549 return 0;
550 }
551
552 entry->jump[2] |= IMM_H18(reladdr);
553 entry->jump[3] |= IMM_L(reladdr);
554 } else {
555 /* Stub uses address relative to kernel toc (from the paca) */
556 reladdr = addr - kernel_toc_addr();
557 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
558 pr_err("%s: Address of %ps out of range of kernel_toc.\n",
559 me->name, (void *)addr);
560 return 0;
561 }
562
563 entry->jump[1] |= PPC_HA(reladdr);
564 entry->jump[2] |= PPC_LO(reladdr);
565 }
566
567 /* Even though we don't use funcdata in the stub, it's needed elsewhere. */
568 entry->funcdata = func_desc(addr);
569 entry->magic = STUB_MAGIC;
570
571 return 1;
572}
573
574static bool is_mprofile_ftrace_call(const char *name)
575{
576 if (!strcmp("_mcount", name))
577 return true;
578#ifdef CONFIG_DYNAMIC_FTRACE
579 if (!strcmp("ftrace_caller", name))
580 return true;
581#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
582 if (!strcmp("ftrace_regs_caller", name))
583 return true;
584#endif
585#endif
586
587 return false;
588}
589#else
590static inline int create_ftrace_stub(struct ppc64_stub_entry *entry,
591 unsigned long addr,
592 struct module *me)
593{
594 return 0;
595}
596
597static bool is_mprofile_ftrace_call(const char *name)
598{
599 return false;
600}
601#endif
602
603/*
604 * r2 is the TOC pointer: it actually points 0x8000 into the TOC (this gives the
605 * value maximum span in an instruction which uses a signed offset). Round down
606 * to a 256 byte boundary for the odd case where we are setting up r2 without a
607 * .toc section.
608 */
609static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me)
610{
611#ifndef CONFIG_PPC_KERNEL_PCREL
612 return (sechdrs[me->arch.toc_section].sh_addr & ~0xfful) + 0x8000;
613#else
614 return -1;
615#endif
616}
617
618/* Patch stub to reference function and correct r2 value. */
619static inline int create_stub(const Elf64_Shdr *sechdrs,
620 struct ppc64_stub_entry *entry,
621 unsigned long addr,
622 struct module *me,
623 const char *name)
624{
625 long reladdr;
626 func_desc_t desc;
627 int i;
628
629 if (is_mprofile_ftrace_call(name))
630 return create_ftrace_stub(entry, addr, me);
631
632 if ((unsigned long)entry->jump % 8 != 0) {
633 pr_err("%s: Address of stub entry is not 8-byte aligned\n", me->name);
634 return 0;
635 }
636
637 BUILD_BUG_ON(sizeof(ppc64_stub_insns) > sizeof(entry->jump));
638 for (i = 0; i < ARRAY_SIZE(ppc64_stub_insns); i++) {
639 if (patch_instruction(&entry->jump[i],
640 ppc_inst(ppc64_stub_insns[i])))
641 return 0;
642 }
643
644 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) {
645 /* Stub uses address relative to itself! */
646 reladdr = 0 + offsetof(struct ppc64_stub_entry, funcdata);
647 BUILD_BUG_ON(reladdr != 32);
648 if (reladdr > 0x1FFFFFFFFL || reladdr < -0x200000000L) {
649 pr_err("%s: Address of %p out of range of 34-bit relative address.\n",
650 me->name, (void *)reladdr);
651 return 0;
652 }
653 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr);
654
655 /* May not even need this if we're relative to 0 */
656 if (patch_instruction(&entry->jump[0],
657 ppc_inst_prefix(entry->jump[0] | IMM_H18(reladdr),
658 entry->jump[1] | IMM_L(reladdr))))
659 return 0;
660
661 } else {
662 /* Stub uses address relative to r2. */
663 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
664 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
665 pr_err("%s: Address %p of stub out of range of %p.\n",
666 me->name, (void *)reladdr, (void *)my_r2);
667 return 0;
668 }
669 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr);
670
671 if (patch_instruction(&entry->jump[0],
672 ppc_inst(entry->jump[0] | PPC_HA(reladdr))))
673 return 0;
674
675 if (patch_instruction(&entry->jump[1],
676 ppc_inst(entry->jump[1] | PPC_LO(reladdr))))
677 return 0;
678 }
679
680 // func_desc_t is 8 bytes if ABIv2, else 16 bytes
681 desc = func_desc(addr);
682 for (i = 0; i < sizeof(func_desc_t) / sizeof(u32); i++) {
683 if (patch_u32(((u32 *)&entry->funcdata) + i, ((u32 *)&desc)[i]))
684 return 0;
685 }
686
687 if (patch_u32(&entry->magic, STUB_MAGIC))
688 return 0;
689
690 return 1;
691}
692
693/* Create stub to jump to function described in this OPD/ptr: we need the
694 stub to set up the TOC ptr (r2) for the function. */
695static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs,
696 unsigned long addr,
697 struct module *me,
698 const char *name)
699{
700 struct ppc64_stub_entry *stubs;
701 unsigned int i, num_stubs;
702
703 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
704
705 /* Find this stub, or if that fails, the next avail. entry */
706 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
707 for (i = 0; i < me->arch.stub_count; i++) {
708 if (WARN_ON(i >= num_stubs))
709 return 0;
710
711 if (stub_func_addr(stubs[i].funcdata) == func_addr(addr))
712 return (unsigned long)&stubs[i];
713 }
714
715 if (!create_stub(sechdrs, &stubs[i], addr, me, name))
716 return 0;
717
718 me->arch.stub_count++;
719 return (unsigned long)&stubs[i];
720}
721
722#ifdef CONFIG_PPC_KERNEL_PCREL
723/* Create GOT to load the location described in this ptr */
724static unsigned long got_for_addr(const Elf64_Shdr *sechdrs,
725 unsigned long addr,
726 struct module *me,
727 const char *name)
728{
729 struct ppc64_got_entry *got;
730 unsigned int i, num_got;
731
732 if (!IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
733 return addr;
734
735 num_got = sechdrs[me->arch.got_section].sh_size / sizeof(*got);
736
737 /* Find this stub, or if that fails, the next avail. entry */
738 got = (void *)sechdrs[me->arch.got_section].sh_addr;
739 for (i = 0; got[i].addr; i++) {
740 if (WARN_ON(i >= num_got))
741 return 0;
742
743 if (got[i].addr == addr)
744 return (unsigned long)&got[i];
745 }
746
747 got[i].addr = addr;
748
749 return (unsigned long)&got[i];
750}
751#endif
752
753/* We expect a noop next: if it is, replace it with instruction to
754 restore r2. */
755static int restore_r2(const char *name, u32 *instruction, struct module *me)
756{
757 u32 *prev_insn = instruction - 1;
758 u32 insn_val = *instruction;
759
760 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
761 return 0;
762
763 if (is_mprofile_ftrace_call(name))
764 return 0;
765
766 /*
767 * Make sure the branch isn't a sibling call. Sibling calls aren't
768 * "link" branches and they don't return, so they don't need the r2
769 * restore afterwards.
770 */
771 if (!instr_is_relative_link_branch(ppc_inst(*prev_insn)))
772 return 0;
773
774 /*
775 * For livepatch, the restore r2 instruction might have already been
776 * written previously, if the referenced symbol is in a previously
777 * unloaded module which is now being loaded again. In that case, skip
778 * the warning and the instruction write.
779 */
780 if (insn_val == PPC_INST_LD_TOC)
781 return 0;
782
783 if (insn_val != PPC_RAW_NOP()) {
784 pr_err("%s: Expected nop after call, got %08x at %pS\n",
785 me->name, insn_val, instruction);
786 return -ENOEXEC;
787 }
788
789 /* ld r2,R2_STACK_OFFSET(r1) */
790 return patch_instruction(instruction, ppc_inst(PPC_INST_LD_TOC));
791}
792
793int apply_relocate_add(Elf64_Shdr *sechdrs,
794 const char *strtab,
795 unsigned int symindex,
796 unsigned int relsec,
797 struct module *me)
798{
799 unsigned int i;
800 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
801 Elf64_Sym *sym;
802 unsigned long *location;
803 unsigned long value;
804
805 pr_debug("Applying ADD relocate section %u to %u\n", relsec,
806 sechdrs[relsec].sh_info);
807
808#ifndef CONFIG_PPC_KERNEL_PCREL
809 /* First time we're called, we can fix up .TOC. */
810 if (!me->arch.toc_fixed) {
811 sym = find_dot_toc(sechdrs, strtab, symindex);
812 /* It's theoretically possible that a module doesn't want a
813 * .TOC. so don't fail it just for that. */
814 if (sym)
815 sym->st_value = my_r2(sechdrs, me);
816 me->arch.toc_fixed = true;
817 }
818#endif
819 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
820 /* This is where to make the change */
821 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
822 + rela[i].r_offset;
823 /* This is the symbol it is referring to */
824 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
825 + ELF64_R_SYM(rela[i].r_info);
826
827 pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n",
828 location, (long)ELF64_R_TYPE(rela[i].r_info),
829 strtab + sym->st_name, (unsigned long)sym->st_value,
830 (long)rela[i].r_addend);
831
832 /* `Everything is relative'. */
833 value = sym->st_value + rela[i].r_addend;
834
835 switch (ELF64_R_TYPE(rela[i].r_info)) {
836 case R_PPC64_ADDR32:
837 /* Simply set it */
838 *(u32 *)location = value;
839 break;
840
841 case R_PPC64_ADDR64:
842 /* Simply set it */
843 *(unsigned long *)location = value;
844 break;
845
846#ifndef CONFIG_PPC_KERNEL_PCREL
847 case R_PPC64_TOC:
848 *(unsigned long *)location = my_r2(sechdrs, me);
849 break;
850
851 case R_PPC64_TOC16:
852 /* Subtract TOC pointer */
853 value -= my_r2(sechdrs, me);
854 if (value + 0x8000 > 0xffff) {
855 pr_err("%s: bad TOC16 relocation (0x%lx)\n",
856 me->name, value);
857 return -ENOEXEC;
858 }
859 *((uint16_t *) location)
860 = (*((uint16_t *) location) & ~0xffff)
861 | (value & 0xffff);
862 break;
863
864 case R_PPC64_TOC16_LO:
865 /* Subtract TOC pointer */
866 value -= my_r2(sechdrs, me);
867 *((uint16_t *) location)
868 = (*((uint16_t *) location) & ~0xffff)
869 | (value & 0xffff);
870 break;
871
872 case R_PPC64_TOC16_DS:
873 /* Subtract TOC pointer */
874 value -= my_r2(sechdrs, me);
875 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
876 pr_err("%s: bad TOC16_DS relocation (0x%lx)\n",
877 me->name, value);
878 return -ENOEXEC;
879 }
880 *((uint16_t *) location)
881 = (*((uint16_t *) location) & ~0xfffc)
882 | (value & 0xfffc);
883 break;
884
885 case R_PPC64_TOC16_LO_DS:
886 /* Subtract TOC pointer */
887 value -= my_r2(sechdrs, me);
888 if ((value & 3) != 0) {
889 pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n",
890 me->name, value);
891 return -ENOEXEC;
892 }
893 *((uint16_t *) location)
894 = (*((uint16_t *) location) & ~0xfffc)
895 | (value & 0xfffc);
896 break;
897
898 case R_PPC64_TOC16_HA:
899 /* Subtract TOC pointer */
900 value -= my_r2(sechdrs, me);
901 value = ((value + 0x8000) >> 16);
902 *((uint16_t *) location)
903 = (*((uint16_t *) location) & ~0xffff)
904 | (value & 0xffff);
905 break;
906#endif
907
908 case R_PPC_REL24:
909#ifdef CONFIG_PPC_KERNEL_PCREL
910 /* PCREL still generates REL24 for mcount */
911 case R_PPC64_REL24_NOTOC:
912#endif
913 /* FIXME: Handle weak symbols here --RR */
914 if (sym->st_shndx == SHN_UNDEF ||
915 sym->st_shndx == SHN_LIVEPATCH) {
916 /* External: go via stub */
917 value = stub_for_addr(sechdrs, value, me,
918 strtab + sym->st_name);
919 if (!value)
920 return -ENOENT;
921 if (restore_r2(strtab + sym->st_name,
922 (u32 *)location + 1, me))
923 return -ENOEXEC;
924 } else
925 value += local_entry_offset(sym);
926
927 /* Convert value to relative */
928 value -= (unsigned long)location;
929 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
930 pr_err("%s: REL24 %li out of range!\n",
931 me->name, (long int)value);
932 return -ENOEXEC;
933 }
934
935 /* Only replace bits 2 through 26 */
936 value = (*(uint32_t *)location & ~PPC_LI_MASK) | PPC_LI(value);
937
938 if (patch_instruction((u32 *)location, ppc_inst(value)))
939 return -EFAULT;
940
941 break;
942
943 case R_PPC64_REL64:
944 /* 64 bits relative (used by features fixups) */
945 *location = value - (unsigned long)location;
946 break;
947
948 case R_PPC64_REL32:
949 /* 32 bits relative (used by relative exception tables) */
950 /* Convert value to relative */
951 value -= (unsigned long)location;
952 if (value + 0x80000000 > 0xffffffff) {
953 pr_err("%s: REL32 %li out of range!\n",
954 me->name, (long int)value);
955 return -ENOEXEC;
956 }
957 *(u32 *)location = value;
958 break;
959
960#ifdef CONFIG_PPC_KERNEL_PCREL
961 case R_PPC64_PCREL34: {
962 unsigned long absvalue = value;
963
964 /* Convert value to relative */
965 value -= (unsigned long)location;
966
967 if (value + 0x200000000 > 0x3ffffffff) {
968 if (sym->st_shndx != me->arch.pcpu_section) {
969 pr_err("%s: REL34 %li out of range!\n",
970 me->name, (long)value);
971 return -ENOEXEC;
972 }
973
974 /*
975 * per-cpu section is special cased because
976 * it is moved during loading, so has to be
977 * converted to use GOT.
978 */
979 value = got_for_addr(sechdrs, absvalue, me,
980 strtab + sym->st_name);
981 if (!value)
982 return -ENOENT;
983 value -= (unsigned long)location;
984
985 /* Turn pla into pld */
986 if (patch_instruction((u32 *)location,
987 ppc_inst_prefix((*(u32 *)location & ~0x02000000),
988 (*((u32 *)location + 1) & ~0xf8000000) | 0xe4000000)))
989 return -EFAULT;
990 }
991
992 if (patch_instruction((u32 *)location,
993 ppc_inst_prefix((*(u32 *)location & ~0x3ffff) | IMM_H18(value),
994 (*((u32 *)location + 1) & ~0xffff) | IMM_L(value))))
995 return -EFAULT;
996
997 break;
998 }
999
1000#else
1001 case R_PPC64_TOCSAVE:
1002 /*
1003 * Marker reloc indicates we don't have to save r2.
1004 * That would only save us one instruction, so ignore
1005 * it.
1006 */
1007 break;
1008#endif
1009
1010 case R_PPC64_ENTRY:
1011 if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
1012 break;
1013
1014 /*
1015 * Optimize ELFv2 large code model entry point if
1016 * the TOC is within 2GB range of current location.
1017 */
1018 value = my_r2(sechdrs, me) - (unsigned long)location;
1019 if (value + 0x80008000 > 0xffffffff)
1020 break;
1021 /*
1022 * Check for the large code model prolog sequence:
1023 * ld r2, ...(r12)
1024 * add r2, r2, r12
1025 */
1026 if ((((uint32_t *)location)[0] & ~0xfffc) != PPC_RAW_LD(_R2, _R12, 0))
1027 break;
1028 if (((uint32_t *)location)[1] != PPC_RAW_ADD(_R2, _R2, _R12))
1029 break;
1030 /*
1031 * If found, replace it with:
1032 * addis r2, r12, (.TOC.-func)@ha
1033 * addi r2, r2, (.TOC.-func)@l
1034 */
1035 ((uint32_t *)location)[0] = PPC_RAW_ADDIS(_R2, _R12, PPC_HA(value));
1036 ((uint32_t *)location)[1] = PPC_RAW_ADDI(_R2, _R2, PPC_LO(value));
1037 break;
1038
1039 case R_PPC64_REL16_HA:
1040 /* Subtract location pointer */
1041 value -= (unsigned long)location;
1042 value = ((value + 0x8000) >> 16);
1043 *((uint16_t *) location)
1044 = (*((uint16_t *) location) & ~0xffff)
1045 | (value & 0xffff);
1046 break;
1047
1048 case R_PPC64_REL16_LO:
1049 /* Subtract location pointer */
1050 value -= (unsigned long)location;
1051 *((uint16_t *) location)
1052 = (*((uint16_t *) location) & ~0xffff)
1053 | (value & 0xffff);
1054 break;
1055
1056#ifdef CONFIG_PPC_KERNEL_PCREL
1057 case R_PPC64_GOT_PCREL34:
1058 value = got_for_addr(sechdrs, value, me,
1059 strtab + sym->st_name);
1060 if (!value)
1061 return -ENOENT;
1062 value -= (unsigned long)location;
1063 ((uint32_t *)location)[0] = (((uint32_t *)location)[0] & ~0x3ffff) |
1064 ((value >> 16) & 0x3ffff);
1065 ((uint32_t *)location)[1] = (((uint32_t *)location)[1] & ~0xffff) |
1066 (value & 0xffff);
1067 break;
1068#endif
1069
1070 default:
1071 pr_err("%s: Unknown ADD relocation: %lu\n",
1072 me->name,
1073 (unsigned long)ELF64_R_TYPE(rela[i].r_info));
1074 return -ENOEXEC;
1075 }
1076 }
1077
1078 return 0;
1079}
1080
1081#ifdef CONFIG_DYNAMIC_FTRACE
1082int module_trampoline_target(struct module *mod, unsigned long addr,
1083 unsigned long *target)
1084{
1085 struct ppc64_stub_entry *stub;
1086 func_desc_t funcdata;
1087 u32 magic;
1088
1089 if (!within_module_core(addr, mod)) {
1090 pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name);
1091 return -EFAULT;
1092 }
1093
1094 stub = (struct ppc64_stub_entry *)addr;
1095
1096 if (copy_from_kernel_nofault(&magic, &stub->magic,
1097 sizeof(magic))) {
1098 pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name);
1099 return -EFAULT;
1100 }
1101
1102 if (magic != STUB_MAGIC) {
1103 pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name);
1104 return -EFAULT;
1105 }
1106
1107 if (copy_from_kernel_nofault(&funcdata, &stub->funcdata,
1108 sizeof(funcdata))) {
1109 pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name);
1110 return -EFAULT;
1111 }
1112
1113 *target = stub_func_addr(funcdata);
1114
1115 return 0;
1116}
1117
1118static int setup_ftrace_ool_stubs(const Elf64_Shdr *sechdrs, unsigned long addr, struct module *me)
1119{
1120#ifdef CONFIG_PPC_FTRACE_OUT_OF_LINE
1121 unsigned int total_stubs, num_stubs;
1122 struct ppc64_stub_entry *stub;
1123
1124 total_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stub);
1125 num_stubs = roundup(me->arch.ool_stub_count * sizeof(struct ftrace_ool_stub),
1126 sizeof(struct ppc64_stub_entry)) / sizeof(struct ppc64_stub_entry);
1127
1128 if (WARN_ON(me->arch.stub_count + num_stubs > total_stubs))
1129 return -1;
1130
1131 stub = (void *)sechdrs[me->arch.stubs_section].sh_addr;
1132 me->arch.ool_stubs = (struct ftrace_ool_stub *)(stub + me->arch.stub_count);
1133 me->arch.stub_count += num_stubs;
1134#endif
1135
1136 return 0;
1137}
1138
1139int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs)
1140{
1141 mod->arch.tramp = stub_for_addr(sechdrs,
1142 (unsigned long)ftrace_caller,
1143 mod,
1144 "ftrace_caller");
1145#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1146 mod->arch.tramp_regs = stub_for_addr(sechdrs,
1147 (unsigned long)ftrace_regs_caller,
1148 mod,
1149 "ftrace_regs_caller");
1150 if (!mod->arch.tramp_regs)
1151 return -ENOENT;
1152#endif
1153
1154 if (!mod->arch.tramp)
1155 return -ENOENT;
1156
1157 if (setup_ftrace_ool_stubs(sechdrs, mod->arch.tramp, mod))
1158 return -ENOENT;
1159
1160 return 0;
1161}
1162#endif