Loading...
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
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). We only
107 * need 6 instructions on ABIv2 but we always allocate 7 so
108 * so we don't have to modify the trampoline load instruction. */
109 u32 jump[7];
110 /* Used by ftrace to identify stubs */
111 u32 magic;
112 /* Data for the above code */
113 func_desc_t funcdata;
114};
115
116/*
117 * PPC64 uses 24 bit jumps, but we need to jump into other modules or
118 * the kernel which may be further. So we jump to a stub.
119 *
120 * For ELFv1 we need to use this to set up the new r2 value (aka TOC
121 * pointer). For ELFv2 it's the callee's responsibility to set up the
122 * new r2, but for both we need to save the old r2.
123 *
124 * We could simply patch the new r2 value and function pointer into
125 * the stub, but it's significantly shorter to put these values at the
126 * end of the stub code, and patch the stub address (32-bits relative
127 * to the TOC ptr, r2) into the stub.
128 */
129static u32 ppc64_stub_insns[] = {
130 PPC_RAW_ADDIS(_R11, _R2, 0),
131 PPC_RAW_ADDI(_R11, _R11, 0),
132 /* Save current r2 value in magic place on the stack. */
133 PPC_RAW_STD(_R2, _R1, R2_STACK_OFFSET),
134 PPC_RAW_LD(_R12, _R11, 32),
135#ifdef CONFIG_PPC64_ELF_ABI_V1
136 /* Set up new r2 from function descriptor */
137 PPC_RAW_LD(_R2, _R11, 40),
138#endif
139 PPC_RAW_MTCTR(_R12),
140 PPC_RAW_BCTR(),
141};
142
143/* Count how many different 24-bit relocations (different symbol,
144 different addend) */
145static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num)
146{
147 unsigned int i, r_info, r_addend, _count_relocs;
148
149 /* FIXME: Only count external ones --RR */
150 _count_relocs = 0;
151 r_info = 0;
152 r_addend = 0;
153 for (i = 0; i < num; i++)
154 /* Only count 24-bit relocs, others don't need stubs */
155 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
156 (r_info != ELF64_R_SYM(rela[i].r_info) ||
157 r_addend != rela[i].r_addend)) {
158 _count_relocs++;
159 r_info = ELF64_R_SYM(rela[i].r_info);
160 r_addend = rela[i].r_addend;
161 }
162
163 return _count_relocs;
164}
165
166static int relacmp(const void *_x, const void *_y)
167{
168 const Elf64_Rela *x, *y;
169
170 y = (Elf64_Rela *)_x;
171 x = (Elf64_Rela *)_y;
172
173 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to
174 * make the comparison cheaper/faster. It won't affect the sorting or
175 * the counting algorithms' performance
176 */
177 if (x->r_info < y->r_info)
178 return -1;
179 else if (x->r_info > y->r_info)
180 return 1;
181 else if (x->r_addend < y->r_addend)
182 return -1;
183 else if (x->r_addend > y->r_addend)
184 return 1;
185 else
186 return 0;
187}
188
189/* Get size of potential trampolines required. */
190static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
191 const Elf64_Shdr *sechdrs)
192{
193 /* One extra reloc so it's always 0-addr terminated */
194 unsigned long relocs = 1;
195 unsigned i;
196
197 /* Every relocated section... */
198 for (i = 1; i < hdr->e_shnum; i++) {
199 if (sechdrs[i].sh_type == SHT_RELA) {
200 pr_debug("Found relocations in section %u\n", i);
201 pr_debug("Ptr: %p. Number: %Lu\n",
202 (void *)sechdrs[i].sh_addr,
203 sechdrs[i].sh_size / sizeof(Elf64_Rela));
204
205 /* Sort the relocation information based on a symbol and
206 * addend key. This is a stable O(n*log n) complexity
207 * algorithm but it will reduce the complexity of
208 * count_relocs() to linear complexity O(n)
209 */
210 sort((void *)sechdrs[i].sh_addr,
211 sechdrs[i].sh_size / sizeof(Elf64_Rela),
212 sizeof(Elf64_Rela), relacmp, NULL);
213
214 relocs += count_relocs((void *)sechdrs[i].sh_addr,
215 sechdrs[i].sh_size
216 / sizeof(Elf64_Rela));
217 }
218 }
219
220#ifdef CONFIG_DYNAMIC_FTRACE
221 /* make the trampoline to the ftrace_caller */
222 relocs++;
223#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
224 /* an additional one for ftrace_regs_caller */
225 relocs++;
226#endif
227#endif
228
229 pr_debug("Looks like a total of %lu stubs, max\n", relocs);
230 return relocs * sizeof(struct ppc64_stub_entry);
231}
232
233/* Still needed for ELFv2, for .TOC. */
234static void dedotify_versions(struct modversion_info *vers,
235 unsigned long size)
236{
237 struct modversion_info *end;
238
239 for (end = (void *)vers + size; vers < end; vers++)
240 if (vers->name[0] == '.') {
241 memmove(vers->name, vers->name+1, strlen(vers->name));
242 }
243}
244
245/*
246 * Undefined symbols which refer to .funcname, hack to funcname. Make .TOC.
247 * seem to be defined (value set later).
248 */
249static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
250{
251 unsigned int i;
252
253 for (i = 1; i < numsyms; i++) {
254 if (syms[i].st_shndx == SHN_UNDEF) {
255 char *name = strtab + syms[i].st_name;
256 if (name[0] == '.') {
257 if (strcmp(name+1, "TOC.") == 0)
258 syms[i].st_shndx = SHN_ABS;
259 syms[i].st_name++;
260 }
261 }
262 }
263}
264
265static Elf64_Sym *find_dot_toc(Elf64_Shdr *sechdrs,
266 const char *strtab,
267 unsigned int symindex)
268{
269 unsigned int i, numsyms;
270 Elf64_Sym *syms;
271
272 syms = (Elf64_Sym *)sechdrs[symindex].sh_addr;
273 numsyms = sechdrs[symindex].sh_size / sizeof(Elf64_Sym);
274
275 for (i = 1; i < numsyms; i++) {
276 if (syms[i].st_shndx == SHN_ABS
277 && strcmp(strtab + syms[i].st_name, "TOC.") == 0)
278 return &syms[i];
279 }
280 return NULL;
281}
282
283bool module_init_section(const char *name)
284{
285 /* We don't handle .init for the moment: always return false. */
286 return false;
287}
288
289int module_frob_arch_sections(Elf64_Ehdr *hdr,
290 Elf64_Shdr *sechdrs,
291 char *secstrings,
292 struct module *me)
293{
294 unsigned int i;
295
296 /* Find .toc and .stubs sections, symtab and strtab */
297 for (i = 1; i < hdr->e_shnum; i++) {
298 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
299 me->arch.stubs_section = i;
300 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0) {
301 me->arch.toc_section = i;
302 if (sechdrs[i].sh_addralign < 8)
303 sechdrs[i].sh_addralign = 8;
304 }
305 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0)
306 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
307 sechdrs[i].sh_size);
308
309 if (sechdrs[i].sh_type == SHT_SYMTAB)
310 dedotify((void *)hdr + sechdrs[i].sh_offset,
311 sechdrs[i].sh_size / sizeof(Elf64_Sym),
312 (void *)hdr
313 + sechdrs[sechdrs[i].sh_link].sh_offset);
314 }
315
316 if (!me->arch.stubs_section) {
317 pr_err("%s: doesn't contain .stubs.\n", me->name);
318 return -ENOEXEC;
319 }
320
321 /* If we don't have a .toc, just use .stubs. We need to set r2
322 to some reasonable value in case the module calls out to
323 other functions via a stub, or if a function pointer escapes
324 the module by some means. */
325 if (!me->arch.toc_section)
326 me->arch.toc_section = me->arch.stubs_section;
327
328 /* Override the stubs size */
329 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs);
330 return 0;
331}
332
333#ifdef CONFIG_MPROFILE_KERNEL
334
335static u32 stub_insns[] = {
336 PPC_RAW_LD(_R12, _R13, offsetof(struct paca_struct, kernel_toc)),
337 PPC_RAW_ADDIS(_R12, _R12, 0),
338 PPC_RAW_ADDI(_R12, _R12, 0),
339 PPC_RAW_MTCTR(_R12),
340 PPC_RAW_BCTR(),
341};
342
343/*
344 * For mprofile-kernel we use a special stub for ftrace_caller() because we
345 * can't rely on r2 containing this module's TOC when we enter the stub.
346 *
347 * That can happen if the function calling us didn't need to use the toc. In
348 * that case it won't have setup r2, and the r2 value will be either the
349 * kernel's toc, or possibly another modules toc.
350 *
351 * To deal with that this stub uses the kernel toc, which is always accessible
352 * via the paca (in r13). The target (ftrace_caller()) is responsible for
353 * saving and restoring the toc before returning.
354 */
355static inline int create_ftrace_stub(struct ppc64_stub_entry *entry,
356 unsigned long addr,
357 struct module *me)
358{
359 long reladdr;
360
361 memcpy(entry->jump, stub_insns, sizeof(stub_insns));
362
363 /* Stub uses address relative to kernel toc (from the paca) */
364 reladdr = addr - kernel_toc_addr();
365 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
366 pr_err("%s: Address of %ps out of range of kernel_toc.\n",
367 me->name, (void *)addr);
368 return 0;
369 }
370
371 entry->jump[1] |= PPC_HA(reladdr);
372 entry->jump[2] |= PPC_LO(reladdr);
373
374 /* Even though we don't use funcdata in the stub, it's needed elsewhere. */
375 entry->funcdata = func_desc(addr);
376 entry->magic = STUB_MAGIC;
377
378 return 1;
379}
380
381static bool is_mprofile_ftrace_call(const char *name)
382{
383 if (!strcmp("_mcount", name))
384 return true;
385#ifdef CONFIG_DYNAMIC_FTRACE
386 if (!strcmp("ftrace_caller", name))
387 return true;
388#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
389 if (!strcmp("ftrace_regs_caller", name))
390 return true;
391#endif
392#endif
393
394 return false;
395}
396#else
397static inline int create_ftrace_stub(struct ppc64_stub_entry *entry,
398 unsigned long addr,
399 struct module *me)
400{
401 return 0;
402}
403
404static bool is_mprofile_ftrace_call(const char *name)
405{
406 return false;
407}
408#endif
409
410/*
411 * r2 is the TOC pointer: it actually points 0x8000 into the TOC (this gives the
412 * value maximum span in an instruction which uses a signed offset). Round down
413 * to a 256 byte boundary for the odd case where we are setting up r2 without a
414 * .toc section.
415 */
416static inline unsigned long my_r2(const Elf64_Shdr *sechdrs, struct module *me)
417{
418 return (sechdrs[me->arch.toc_section].sh_addr & ~0xfful) + 0x8000;
419}
420
421/* Patch stub to reference function and correct r2 value. */
422static inline int create_stub(const Elf64_Shdr *sechdrs,
423 struct ppc64_stub_entry *entry,
424 unsigned long addr,
425 struct module *me,
426 const char *name)
427{
428 long reladdr;
429 func_desc_t desc;
430 int i;
431
432 if (is_mprofile_ftrace_call(name))
433 return create_ftrace_stub(entry, addr, me);
434
435 for (i = 0; i < ARRAY_SIZE(ppc64_stub_insns); i++) {
436 if (patch_instruction(&entry->jump[i],
437 ppc_inst(ppc64_stub_insns[i])))
438 return 0;
439 }
440
441 /* Stub uses address relative to r2. */
442 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
443 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
444 pr_err("%s: Address %p of stub out of range of %p.\n",
445 me->name, (void *)reladdr, (void *)my_r2);
446 return 0;
447 }
448 pr_debug("Stub %p get data from reladdr %li\n", entry, reladdr);
449
450 if (patch_instruction(&entry->jump[0],
451 ppc_inst(entry->jump[0] | PPC_HA(reladdr))))
452 return 0;
453
454 if (patch_instruction(&entry->jump[1],
455 ppc_inst(entry->jump[1] | PPC_LO(reladdr))))
456 return 0;
457
458 // func_desc_t is 8 bytes if ABIv2, else 16 bytes
459 desc = func_desc(addr);
460 for (i = 0; i < sizeof(func_desc_t) / sizeof(u32); i++) {
461 if (patch_instruction(((u32 *)&entry->funcdata) + i,
462 ppc_inst(((u32 *)(&desc))[i])))
463 return 0;
464 }
465
466 if (patch_instruction(&entry->magic, ppc_inst(STUB_MAGIC)))
467 return 0;
468
469 return 1;
470}
471
472/* Create stub to jump to function described in this OPD/ptr: we need the
473 stub to set up the TOC ptr (r2) for the function. */
474static unsigned long stub_for_addr(const Elf64_Shdr *sechdrs,
475 unsigned long addr,
476 struct module *me,
477 const char *name)
478{
479 struct ppc64_stub_entry *stubs;
480 unsigned int i, num_stubs;
481
482 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
483
484 /* Find this stub, or if that fails, the next avail. entry */
485 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
486 for (i = 0; stub_func_addr(stubs[i].funcdata); i++) {
487 if (WARN_ON(i >= num_stubs))
488 return 0;
489
490 if (stub_func_addr(stubs[i].funcdata) == func_addr(addr))
491 return (unsigned long)&stubs[i];
492 }
493
494 if (!create_stub(sechdrs, &stubs[i], addr, me, name))
495 return 0;
496
497 return (unsigned long)&stubs[i];
498}
499
500/* We expect a noop next: if it is, replace it with instruction to
501 restore r2. */
502static int restore_r2(const char *name, u32 *instruction, struct module *me)
503{
504 u32 *prev_insn = instruction - 1;
505
506 if (is_mprofile_ftrace_call(name))
507 return 1;
508
509 /*
510 * Make sure the branch isn't a sibling call. Sibling calls aren't
511 * "link" branches and they don't return, so they don't need the r2
512 * restore afterwards.
513 */
514 if (!instr_is_relative_link_branch(ppc_inst(*prev_insn)))
515 return 1;
516
517 if (*instruction != PPC_RAW_NOP()) {
518 pr_err("%s: Expected nop after call, got %08x at %pS\n",
519 me->name, *instruction, instruction);
520 return 0;
521 }
522
523 /* ld r2,R2_STACK_OFFSET(r1) */
524 if (patch_instruction(instruction, ppc_inst(PPC_INST_LD_TOC)))
525 return 0;
526
527 return 1;
528}
529
530int apply_relocate_add(Elf64_Shdr *sechdrs,
531 const char *strtab,
532 unsigned int symindex,
533 unsigned int relsec,
534 struct module *me)
535{
536 unsigned int i;
537 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
538 Elf64_Sym *sym;
539 unsigned long *location;
540 unsigned long value;
541
542 pr_debug("Applying ADD relocate section %u to %u\n", relsec,
543 sechdrs[relsec].sh_info);
544
545 /* First time we're called, we can fix up .TOC. */
546 if (!me->arch.toc_fixed) {
547 sym = find_dot_toc(sechdrs, strtab, symindex);
548 /* It's theoretically possible that a module doesn't want a
549 * .TOC. so don't fail it just for that. */
550 if (sym)
551 sym->st_value = my_r2(sechdrs, me);
552 me->arch.toc_fixed = true;
553 }
554
555 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
556 /* This is where to make the change */
557 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
558 + rela[i].r_offset;
559 /* This is the symbol it is referring to */
560 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
561 + ELF64_R_SYM(rela[i].r_info);
562
563 pr_debug("RELOC at %p: %li-type as %s (0x%lx) + %li\n",
564 location, (long)ELF64_R_TYPE(rela[i].r_info),
565 strtab + sym->st_name, (unsigned long)sym->st_value,
566 (long)rela[i].r_addend);
567
568 /* `Everything is relative'. */
569 value = sym->st_value + rela[i].r_addend;
570
571 switch (ELF64_R_TYPE(rela[i].r_info)) {
572 case R_PPC64_ADDR32:
573 /* Simply set it */
574 *(u32 *)location = value;
575 break;
576
577 case R_PPC64_ADDR64:
578 /* Simply set it */
579 *(unsigned long *)location = value;
580 break;
581
582 case R_PPC64_TOC:
583 *(unsigned long *)location = my_r2(sechdrs, me);
584 break;
585
586 case R_PPC64_TOC16:
587 /* Subtract TOC pointer */
588 value -= my_r2(sechdrs, me);
589 if (value + 0x8000 > 0xffff) {
590 pr_err("%s: bad TOC16 relocation (0x%lx)\n",
591 me->name, value);
592 return -ENOEXEC;
593 }
594 *((uint16_t *) location)
595 = (*((uint16_t *) location) & ~0xffff)
596 | (value & 0xffff);
597 break;
598
599 case R_PPC64_TOC16_LO:
600 /* Subtract TOC pointer */
601 value -= my_r2(sechdrs, me);
602 *((uint16_t *) location)
603 = (*((uint16_t *) location) & ~0xffff)
604 | (value & 0xffff);
605 break;
606
607 case R_PPC64_TOC16_DS:
608 /* Subtract TOC pointer */
609 value -= my_r2(sechdrs, me);
610 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
611 pr_err("%s: bad TOC16_DS relocation (0x%lx)\n",
612 me->name, value);
613 return -ENOEXEC;
614 }
615 *((uint16_t *) location)
616 = (*((uint16_t *) location) & ~0xfffc)
617 | (value & 0xfffc);
618 break;
619
620 case R_PPC64_TOC16_LO_DS:
621 /* Subtract TOC pointer */
622 value -= my_r2(sechdrs, me);
623 if ((value & 3) != 0) {
624 pr_err("%s: bad TOC16_LO_DS relocation (0x%lx)\n",
625 me->name, value);
626 return -ENOEXEC;
627 }
628 *((uint16_t *) location)
629 = (*((uint16_t *) location) & ~0xfffc)
630 | (value & 0xfffc);
631 break;
632
633 case R_PPC64_TOC16_HA:
634 /* Subtract TOC pointer */
635 value -= my_r2(sechdrs, me);
636 value = ((value + 0x8000) >> 16);
637 *((uint16_t *) location)
638 = (*((uint16_t *) location) & ~0xffff)
639 | (value & 0xffff);
640 break;
641
642 case R_PPC_REL24:
643 /* FIXME: Handle weak symbols here --RR */
644 if (sym->st_shndx == SHN_UNDEF ||
645 sym->st_shndx == SHN_LIVEPATCH) {
646 /* External: go via stub */
647 value = stub_for_addr(sechdrs, value, me,
648 strtab + sym->st_name);
649 if (!value)
650 return -ENOENT;
651 if (!restore_r2(strtab + sym->st_name,
652 (u32 *)location + 1, me))
653 return -ENOEXEC;
654 } else
655 value += local_entry_offset(sym);
656
657 /* Convert value to relative */
658 value -= (unsigned long)location;
659 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
660 pr_err("%s: REL24 %li out of range!\n",
661 me->name, (long int)value);
662 return -ENOEXEC;
663 }
664
665 /* Only replace bits 2 through 26 */
666 value = (*(uint32_t *)location & ~PPC_LI_MASK) | PPC_LI(value);
667
668 if (patch_instruction((u32 *)location, ppc_inst(value)))
669 return -EFAULT;
670
671 break;
672
673 case R_PPC64_REL64:
674 /* 64 bits relative (used by features fixups) */
675 *location = value - (unsigned long)location;
676 break;
677
678 case R_PPC64_REL32:
679 /* 32 bits relative (used by relative exception tables) */
680 /* Convert value to relative */
681 value -= (unsigned long)location;
682 if (value + 0x80000000 > 0xffffffff) {
683 pr_err("%s: REL32 %li out of range!\n",
684 me->name, (long int)value);
685 return -ENOEXEC;
686 }
687 *(u32 *)location = value;
688 break;
689
690 case R_PPC64_TOCSAVE:
691 /*
692 * Marker reloc indicates we don't have to save r2.
693 * That would only save us one instruction, so ignore
694 * it.
695 */
696 break;
697
698 case R_PPC64_ENTRY:
699 /*
700 * Optimize ELFv2 large code model entry point if
701 * the TOC is within 2GB range of current location.
702 */
703 value = my_r2(sechdrs, me) - (unsigned long)location;
704 if (value + 0x80008000 > 0xffffffff)
705 break;
706 /*
707 * Check for the large code model prolog sequence:
708 * ld r2, ...(r12)
709 * add r2, r2, r12
710 */
711 if ((((uint32_t *)location)[0] & ~0xfffc) != PPC_RAW_LD(_R2, _R12, 0))
712 break;
713 if (((uint32_t *)location)[1] != PPC_RAW_ADD(_R2, _R2, _R12))
714 break;
715 /*
716 * If found, replace it with:
717 * addis r2, r12, (.TOC.-func)@ha
718 * addi r2, r2, (.TOC.-func)@l
719 */
720 ((uint32_t *)location)[0] = PPC_RAW_ADDIS(_R2, _R12, PPC_HA(value));
721 ((uint32_t *)location)[1] = PPC_RAW_ADDI(_R2, _R2, PPC_LO(value));
722 break;
723
724 case R_PPC64_REL16_HA:
725 /* Subtract location pointer */
726 value -= (unsigned long)location;
727 value = ((value + 0x8000) >> 16);
728 *((uint16_t *) location)
729 = (*((uint16_t *) location) & ~0xffff)
730 | (value & 0xffff);
731 break;
732
733 case R_PPC64_REL16_LO:
734 /* Subtract location pointer */
735 value -= (unsigned long)location;
736 *((uint16_t *) location)
737 = (*((uint16_t *) location) & ~0xffff)
738 | (value & 0xffff);
739 break;
740
741 default:
742 pr_err("%s: Unknown ADD relocation: %lu\n",
743 me->name,
744 (unsigned long)ELF64_R_TYPE(rela[i].r_info));
745 return -ENOEXEC;
746 }
747 }
748
749 return 0;
750}
751
752#ifdef CONFIG_DYNAMIC_FTRACE
753int module_trampoline_target(struct module *mod, unsigned long addr,
754 unsigned long *target)
755{
756 struct ppc64_stub_entry *stub;
757 func_desc_t funcdata;
758 u32 magic;
759
760 if (!within_module_core(addr, mod)) {
761 pr_err("%s: stub %lx not in module %s\n", __func__, addr, mod->name);
762 return -EFAULT;
763 }
764
765 stub = (struct ppc64_stub_entry *)addr;
766
767 if (copy_from_kernel_nofault(&magic, &stub->magic,
768 sizeof(magic))) {
769 pr_err("%s: fault reading magic for stub %lx for %s\n", __func__, addr, mod->name);
770 return -EFAULT;
771 }
772
773 if (magic != STUB_MAGIC) {
774 pr_err("%s: bad magic for stub %lx for %s\n", __func__, addr, mod->name);
775 return -EFAULT;
776 }
777
778 if (copy_from_kernel_nofault(&funcdata, &stub->funcdata,
779 sizeof(funcdata))) {
780 pr_err("%s: fault reading funcdata for stub %lx for %s\n", __func__, addr, mod->name);
781 return -EFAULT;
782 }
783
784 *target = stub_func_addr(funcdata);
785
786 return 0;
787}
788
789int module_finalize_ftrace(struct module *mod, const Elf_Shdr *sechdrs)
790{
791 mod->arch.tramp = stub_for_addr(sechdrs,
792 (unsigned long)ftrace_caller,
793 mod,
794 "ftrace_caller");
795#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
796 mod->arch.tramp_regs = stub_for_addr(sechdrs,
797 (unsigned long)ftrace_regs_caller,
798 mod,
799 "ftrace_regs_caller");
800 if (!mod->arch.tramp_regs)
801 return -ENOENT;
802#endif
803
804 if (!mod->arch.tramp)
805 return -ENOENT;
806
807 return 0;
808}
809#endif
1/* Kernel module help for PPC64.
2 Copyright (C) 2001, 2003 Rusty Russell IBM Corporation.
3
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
8
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17*/
18#include <linux/module.h>
19#include <linux/elf.h>
20#include <linux/moduleloader.h>
21#include <linux/err.h>
22#include <linux/vmalloc.h>
23#include <linux/ftrace.h>
24#include <linux/bug.h>
25#include <asm/module.h>
26#include <asm/firmware.h>
27#include <asm/code-patching.h>
28#include <linux/sort.h>
29
30#include "setup.h"
31
32/* FIXME: We don't do .init separately. To do this, we'd need to have
33 a separate r2 value in the init and core section, and stub between
34 them, too.
35
36 Using a magic allocator which places modules within 32MB solves
37 this, and makes other things simpler. Anton?
38 --RR. */
39#if 0
40#define DEBUGP printk
41#else
42#define DEBUGP(fmt , ...)
43#endif
44
45/* Like PPC32, we need little trampolines to do > 24-bit jumps (into
46 the kernel itself). But on PPC64, these need to be used for every
47 jump, actually, to reset r2 (TOC+0x8000). */
48struct ppc64_stub_entry
49{
50 /* 28 byte jump instruction sequence (7 instructions) */
51 unsigned char jump[28];
52 unsigned char unused[4];
53 /* Data for the above code */
54 struct ppc64_opd_entry opd;
55};
56
57/* We use a stub to fix up r2 (TOC ptr) and to jump to the (external)
58 function which may be more than 24-bits away. We could simply
59 patch the new r2 value and function pointer into the stub, but it's
60 significantly shorter to put these values at the end of the stub
61 code, and patch the stub address (32-bits relative to the TOC ptr,
62 r2) into the stub. */
63static struct ppc64_stub_entry ppc64_stub =
64{ .jump = {
65 0x3d, 0x82, 0x00, 0x00, /* addis r12,r2, <high> */
66 0x39, 0x8c, 0x00, 0x00, /* addi r12,r12, <low> */
67 /* Save current r2 value in magic place on the stack. */
68 0xf8, 0x41, 0x00, 0x28, /* std r2,40(r1) */
69 0xe9, 0x6c, 0x00, 0x20, /* ld r11,32(r12) */
70 0xe8, 0x4c, 0x00, 0x28, /* ld r2,40(r12) */
71 0x7d, 0x69, 0x03, 0xa6, /* mtctr r11 */
72 0x4e, 0x80, 0x04, 0x20 /* bctr */
73} };
74
75/* Count how many different 24-bit relocations (different symbol,
76 different addend) */
77static unsigned int count_relocs(const Elf64_Rela *rela, unsigned int num)
78{
79 unsigned int i, r_info, r_addend, _count_relocs;
80
81 /* FIXME: Only count external ones --RR */
82 _count_relocs = 0;
83 r_info = 0;
84 r_addend = 0;
85 for (i = 0; i < num; i++)
86 /* Only count 24-bit relocs, others don't need stubs */
87 if (ELF64_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
88 (r_info != ELF64_R_SYM(rela[i].r_info) ||
89 r_addend != rela[i].r_addend)) {
90 _count_relocs++;
91 r_info = ELF64_R_SYM(rela[i].r_info);
92 r_addend = rela[i].r_addend;
93 }
94
95 return _count_relocs;
96}
97
98static int relacmp(const void *_x, const void *_y)
99{
100 const Elf64_Rela *x, *y;
101
102 y = (Elf64_Rela *)_x;
103 x = (Elf64_Rela *)_y;
104
105 /* Compare the entire r_info (as opposed to ELF64_R_SYM(r_info) only) to
106 * make the comparison cheaper/faster. It won't affect the sorting or
107 * the counting algorithms' performance
108 */
109 if (x->r_info < y->r_info)
110 return -1;
111 else if (x->r_info > y->r_info)
112 return 1;
113 else if (x->r_addend < y->r_addend)
114 return -1;
115 else if (x->r_addend > y->r_addend)
116 return 1;
117 else
118 return 0;
119}
120
121static void relaswap(void *_x, void *_y, int size)
122{
123 uint64_t *x, *y, tmp;
124 int i;
125
126 y = (uint64_t *)_x;
127 x = (uint64_t *)_y;
128
129 for (i = 0; i < sizeof(Elf64_Rela) / sizeof(uint64_t); i++) {
130 tmp = x[i];
131 x[i] = y[i];
132 y[i] = tmp;
133 }
134}
135
136/* Get size of potential trampolines required. */
137static unsigned long get_stubs_size(const Elf64_Ehdr *hdr,
138 const Elf64_Shdr *sechdrs)
139{
140 /* One extra reloc so it's always 0-funcaddr terminated */
141 unsigned long relocs = 1;
142 unsigned i;
143
144 /* Every relocated section... */
145 for (i = 1; i < hdr->e_shnum; i++) {
146 if (sechdrs[i].sh_type == SHT_RELA) {
147 DEBUGP("Found relocations in section %u\n", i);
148 DEBUGP("Ptr: %p. Number: %lu\n",
149 (void *)sechdrs[i].sh_addr,
150 sechdrs[i].sh_size / sizeof(Elf64_Rela));
151
152 /* Sort the relocation information based on a symbol and
153 * addend key. This is a stable O(n*log n) complexity
154 * alogrithm but it will reduce the complexity of
155 * count_relocs() to linear complexity O(n)
156 */
157 sort((void *)sechdrs[i].sh_addr,
158 sechdrs[i].sh_size / sizeof(Elf64_Rela),
159 sizeof(Elf64_Rela), relacmp, relaswap);
160
161 relocs += count_relocs((void *)sechdrs[i].sh_addr,
162 sechdrs[i].sh_size
163 / sizeof(Elf64_Rela));
164 }
165 }
166
167#ifdef CONFIG_DYNAMIC_FTRACE
168 /* make the trampoline to the ftrace_caller */
169 relocs++;
170#endif
171
172 DEBUGP("Looks like a total of %lu stubs, max\n", relocs);
173 return relocs * sizeof(struct ppc64_stub_entry);
174}
175
176static void dedotify_versions(struct modversion_info *vers,
177 unsigned long size)
178{
179 struct modversion_info *end;
180
181 for (end = (void *)vers + size; vers < end; vers++)
182 if (vers->name[0] == '.')
183 memmove(vers->name, vers->name+1, strlen(vers->name));
184}
185
186/* Undefined symbols which refer to .funcname, hack to funcname */
187static void dedotify(Elf64_Sym *syms, unsigned int numsyms, char *strtab)
188{
189 unsigned int i;
190
191 for (i = 1; i < numsyms; i++) {
192 if (syms[i].st_shndx == SHN_UNDEF) {
193 char *name = strtab + syms[i].st_name;
194 if (name[0] == '.')
195 memmove(name, name+1, strlen(name));
196 }
197 }
198}
199
200int module_frob_arch_sections(Elf64_Ehdr *hdr,
201 Elf64_Shdr *sechdrs,
202 char *secstrings,
203 struct module *me)
204{
205 unsigned int i;
206
207 /* Find .toc and .stubs sections, symtab and strtab */
208 for (i = 1; i < hdr->e_shnum; i++) {
209 char *p;
210 if (strcmp(secstrings + sechdrs[i].sh_name, ".stubs") == 0)
211 me->arch.stubs_section = i;
212 else if (strcmp(secstrings + sechdrs[i].sh_name, ".toc") == 0)
213 me->arch.toc_section = i;
214 else if (strcmp(secstrings+sechdrs[i].sh_name,"__versions")==0)
215 dedotify_versions((void *)hdr + sechdrs[i].sh_offset,
216 sechdrs[i].sh_size);
217
218 /* We don't handle .init for the moment: rename to _init */
219 while ((p = strstr(secstrings + sechdrs[i].sh_name, ".init")))
220 p[0] = '_';
221
222 if (sechdrs[i].sh_type == SHT_SYMTAB)
223 dedotify((void *)hdr + sechdrs[i].sh_offset,
224 sechdrs[i].sh_size / sizeof(Elf64_Sym),
225 (void *)hdr
226 + sechdrs[sechdrs[i].sh_link].sh_offset);
227 }
228
229 if (!me->arch.stubs_section) {
230 printk("%s: doesn't contain .stubs.\n", me->name);
231 return -ENOEXEC;
232 }
233
234 /* If we don't have a .toc, just use .stubs. We need to set r2
235 to some reasonable value in case the module calls out to
236 other functions via a stub, or if a function pointer escapes
237 the module by some means. */
238 if (!me->arch.toc_section)
239 me->arch.toc_section = me->arch.stubs_section;
240
241 /* Override the stubs size */
242 sechdrs[me->arch.stubs_section].sh_size = get_stubs_size(hdr, sechdrs);
243 return 0;
244}
245
246/* r2 is the TOC pointer: it actually points 0x8000 into the TOC (this
247 gives the value maximum span in an instruction which uses a signed
248 offset) */
249static inline unsigned long my_r2(Elf64_Shdr *sechdrs, struct module *me)
250{
251 return sechdrs[me->arch.toc_section].sh_addr + 0x8000;
252}
253
254/* Both low and high 16 bits are added as SIGNED additions, so if low
255 16 bits has high bit set, high 16 bits must be adjusted. These
256 macros do that (stolen from binutils). */
257#define PPC_LO(v) ((v) & 0xffff)
258#define PPC_HI(v) (((v) >> 16) & 0xffff)
259#define PPC_HA(v) PPC_HI ((v) + 0x8000)
260
261/* Patch stub to reference function and correct r2 value. */
262static inline int create_stub(Elf64_Shdr *sechdrs,
263 struct ppc64_stub_entry *entry,
264 struct ppc64_opd_entry *opd,
265 struct module *me)
266{
267 Elf64_Half *loc1, *loc2;
268 long reladdr;
269
270 *entry = ppc64_stub;
271
272 loc1 = (Elf64_Half *)&entry->jump[2];
273 loc2 = (Elf64_Half *)&entry->jump[6];
274
275 /* Stub uses address relative to r2. */
276 reladdr = (unsigned long)entry - my_r2(sechdrs, me);
277 if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
278 printk("%s: Address %p of stub out of range of %p.\n",
279 me->name, (void *)reladdr, (void *)my_r2);
280 return 0;
281 }
282 DEBUGP("Stub %p get data from reladdr %li\n", entry, reladdr);
283
284 *loc1 = PPC_HA(reladdr);
285 *loc2 = PPC_LO(reladdr);
286 entry->opd.funcaddr = opd->funcaddr;
287 entry->opd.r2 = opd->r2;
288 return 1;
289}
290
291/* Create stub to jump to function described in this OPD: we need the
292 stub to set up the TOC ptr (r2) for the function. */
293static unsigned long stub_for_addr(Elf64_Shdr *sechdrs,
294 unsigned long opdaddr,
295 struct module *me)
296{
297 struct ppc64_stub_entry *stubs;
298 struct ppc64_opd_entry *opd = (void *)opdaddr;
299 unsigned int i, num_stubs;
300
301 num_stubs = sechdrs[me->arch.stubs_section].sh_size / sizeof(*stubs);
302
303 /* Find this stub, or if that fails, the next avail. entry */
304 stubs = (void *)sechdrs[me->arch.stubs_section].sh_addr;
305 for (i = 0; stubs[i].opd.funcaddr; i++) {
306 BUG_ON(i >= num_stubs);
307
308 if (stubs[i].opd.funcaddr == opd->funcaddr)
309 return (unsigned long)&stubs[i];
310 }
311
312 if (!create_stub(sechdrs, &stubs[i], opd, me))
313 return 0;
314
315 return (unsigned long)&stubs[i];
316}
317
318/* We expect a noop next: if it is, replace it with instruction to
319 restore r2. */
320static int restore_r2(u32 *instruction, struct module *me)
321{
322 if (*instruction != PPC_INST_NOP) {
323 printk("%s: Expect noop after relocate, got %08x\n",
324 me->name, *instruction);
325 return 0;
326 }
327 *instruction = 0xe8410028; /* ld r2,40(r1) */
328 return 1;
329}
330
331int apply_relocate_add(Elf64_Shdr *sechdrs,
332 const char *strtab,
333 unsigned int symindex,
334 unsigned int relsec,
335 struct module *me)
336{
337 unsigned int i;
338 Elf64_Rela *rela = (void *)sechdrs[relsec].sh_addr;
339 Elf64_Sym *sym;
340 unsigned long *location;
341 unsigned long value;
342
343 DEBUGP("Applying ADD relocate section %u to %u\n", relsec,
344 sechdrs[relsec].sh_info);
345 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
346 /* This is where to make the change */
347 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
348 + rela[i].r_offset;
349 /* This is the symbol it is referring to */
350 sym = (Elf64_Sym *)sechdrs[symindex].sh_addr
351 + ELF64_R_SYM(rela[i].r_info);
352
353 DEBUGP("RELOC at %p: %li-type as %s (%lu) + %li\n",
354 location, (long)ELF64_R_TYPE(rela[i].r_info),
355 strtab + sym->st_name, (unsigned long)sym->st_value,
356 (long)rela[i].r_addend);
357
358 /* `Everything is relative'. */
359 value = sym->st_value + rela[i].r_addend;
360
361 switch (ELF64_R_TYPE(rela[i].r_info)) {
362 case R_PPC64_ADDR32:
363 /* Simply set it */
364 *(u32 *)location = value;
365 break;
366
367 case R_PPC64_ADDR64:
368 /* Simply set it */
369 *(unsigned long *)location = value;
370 break;
371
372 case R_PPC64_TOC:
373 *(unsigned long *)location = my_r2(sechdrs, me);
374 break;
375
376 case R_PPC64_TOC16:
377 /* Subtract TOC pointer */
378 value -= my_r2(sechdrs, me);
379 if (value + 0x8000 > 0xffff) {
380 printk("%s: bad TOC16 relocation (%lu)\n",
381 me->name, value);
382 return -ENOEXEC;
383 }
384 *((uint16_t *) location)
385 = (*((uint16_t *) location) & ~0xffff)
386 | (value & 0xffff);
387 break;
388
389 case R_PPC64_TOC16_DS:
390 /* Subtract TOC pointer */
391 value -= my_r2(sechdrs, me);
392 if ((value & 3) != 0 || value + 0x8000 > 0xffff) {
393 printk("%s: bad TOC16_DS relocation (%lu)\n",
394 me->name, value);
395 return -ENOEXEC;
396 }
397 *((uint16_t *) location)
398 = (*((uint16_t *) location) & ~0xfffc)
399 | (value & 0xfffc);
400 break;
401
402 case R_PPC_REL24:
403 /* FIXME: Handle weak symbols here --RR */
404 if (sym->st_shndx == SHN_UNDEF) {
405 /* External: go via stub */
406 value = stub_for_addr(sechdrs, value, me);
407 if (!value)
408 return -ENOENT;
409 if (!restore_r2((u32 *)location + 1, me))
410 return -ENOEXEC;
411 }
412
413 /* Convert value to relative */
414 value -= (unsigned long)location;
415 if (value + 0x2000000 > 0x3ffffff || (value & 3) != 0){
416 printk("%s: REL24 %li out of range!\n",
417 me->name, (long int)value);
418 return -ENOEXEC;
419 }
420
421 /* Only replace bits 2 through 26 */
422 *(uint32_t *)location
423 = (*(uint32_t *)location & ~0x03fffffc)
424 | (value & 0x03fffffc);
425 break;
426
427 case R_PPC64_REL64:
428 /* 64 bits relative (used by features fixups) */
429 *location = value - (unsigned long)location;
430 break;
431
432 default:
433 printk("%s: Unknown ADD relocation: %lu\n",
434 me->name,
435 (unsigned long)ELF64_R_TYPE(rela[i].r_info));
436 return -ENOEXEC;
437 }
438 }
439
440#ifdef CONFIG_DYNAMIC_FTRACE
441 me->arch.toc = my_r2(sechdrs, me);
442 me->arch.tramp = stub_for_addr(sechdrs,
443 (unsigned long)ftrace_caller,
444 me);
445#endif
446
447 return 0;
448}