xref: /openbmc/linux/arch/arm/kernel/module.c (revision 08b7cf13)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/arch/arm/kernel/module.c
4  *
5  *  Copyright (C) 2002 Russell King.
6  *  Modified for nommu by Hyok S. Choi
7  *
8  * Module allocation method suggested by Andi Kleen.
9  */
10 #include <linux/module.h>
11 #include <linux/moduleloader.h>
12 #include <linux/kernel.h>
13 #include <linux/mm.h>
14 #include <linux/elf.h>
15 #include <linux/vmalloc.h>
16 #include <linux/fs.h>
17 #include <linux/string.h>
18 #include <linux/gfp.h>
19 
20 #include <asm/sections.h>
21 #include <asm/smp_plat.h>
22 #include <asm/unwind.h>
23 #include <asm/opcodes.h>
24 
25 #ifdef CONFIG_XIP_KERNEL
26 /*
27  * The XIP kernel text is mapped in the module area for modules and
28  * some other stuff to work without any indirect relocations.
29  * MODULES_VADDR is redefined here and not in asm/memory.h to avoid
30  * recompiling the whole kernel when CONFIG_XIP_KERNEL is turned on/off.
31  */
32 #undef MODULES_VADDR
33 #define MODULES_VADDR	(((unsigned long)_exiprom + ~PMD_MASK) & PMD_MASK)
34 #endif
35 
36 #ifdef CONFIG_MMU
37 void *module_alloc(unsigned long size)
38 {
39 	gfp_t gfp_mask = GFP_KERNEL;
40 	void *p;
41 
42 	/* Silence the initial allocation */
43 	if (IS_ENABLED(CONFIG_ARM_MODULE_PLTS))
44 		gfp_mask |= __GFP_NOWARN;
45 
46 	p = __vmalloc_node_range(size, 1, MODULES_VADDR, MODULES_END,
47 				gfp_mask, PAGE_KERNEL_EXEC, 0, NUMA_NO_NODE,
48 				__builtin_return_address(0));
49 	if (!IS_ENABLED(CONFIG_ARM_MODULE_PLTS) || p)
50 		return p;
51 	return __vmalloc_node_range(size, 1,  VMALLOC_START, VMALLOC_END,
52 				GFP_KERNEL, PAGE_KERNEL_EXEC, 0, NUMA_NO_NODE,
53 				__builtin_return_address(0));
54 }
55 #endif
56 
57 bool module_init_section(const char *name)
58 {
59 	return strstarts(name, ".init") ||
60 		strstarts(name, ".ARM.extab.init") ||
61 		strstarts(name, ".ARM.exidx.init");
62 }
63 
64 bool module_exit_section(const char *name)
65 {
66 	return strstarts(name, ".exit") ||
67 		strstarts(name, ".ARM.extab.exit") ||
68 		strstarts(name, ".ARM.exidx.exit");
69 }
70 
71 #ifdef CONFIG_ARM_HAS_GROUP_RELOCS
72 /*
73  * This implements the partitioning algorithm for group relocations as
74  * documented in the ARM AArch32 ELF psABI (IHI 0044).
75  *
76  * A single PC-relative symbol reference is divided in up to 3 add or subtract
77  * operations, where the final one could be incorporated into a load/store
78  * instruction with immediate offset. E.g.,
79  *
80  *   ADD	Rd, PC, #...		or	ADD	Rd, PC, #...
81  *   ADD	Rd, Rd, #...			ADD	Rd, Rd, #...
82  *   LDR	Rd, [Rd, #...]			ADD	Rd, Rd, #...
83  *
84  * The latter has a guaranteed range of only 16 MiB (3x8 == 24 bits), so it is
85  * of limited use in the kernel. However, the ADD/ADD/LDR combo has a range of
86  * -/+ 256 MiB, (2x8 + 12 == 28 bits), which means it has sufficient range for
87  * any in-kernel symbol reference (unless module PLTs are being used).
88  *
89  * The main advantage of this approach over the typical pattern using a literal
90  * load is that literal loads may miss in the D-cache, and generally lead to
91  * lower cache efficiency for variables that are referenced often from many
92  * different places in the code.
93  */
94 static u32 get_group_rem(u32 group, u32 *offset)
95 {
96 	u32 val = *offset;
97 	u32 shift;
98 	do {
99 		shift = val ? (31 - __fls(val)) & ~1 : 32;
100 		*offset = val;
101 		if (!val)
102 			break;
103 		val &= 0xffffff >> shift;
104 	} while (group--);
105 	return shift;
106 }
107 #endif
108 
109 int
110 apply_relocate(Elf32_Shdr *sechdrs, const char *strtab, unsigned int symindex,
111 	       unsigned int relindex, struct module *module)
112 {
113 	Elf32_Shdr *symsec = sechdrs + symindex;
114 	Elf32_Shdr *relsec = sechdrs + relindex;
115 	Elf32_Shdr *dstsec = sechdrs + relsec->sh_info;
116 	Elf32_Rel *rel = (void *)relsec->sh_addr;
117 	unsigned int i;
118 
119 	for (i = 0; i < relsec->sh_size / sizeof(Elf32_Rel); i++, rel++) {
120 		unsigned long loc;
121 		Elf32_Sym *sym;
122 		const char *symname;
123 #ifdef CONFIG_ARM_HAS_GROUP_RELOCS
124 		u32 shift, group = 1;
125 #endif
126 		s32 offset;
127 		u32 tmp;
128 #ifdef CONFIG_THUMB2_KERNEL
129 		u32 upper, lower, sign, j1, j2;
130 #endif
131 
132 		offset = ELF32_R_SYM(rel->r_info);
133 		if (offset < 0 || offset > (symsec->sh_size / sizeof(Elf32_Sym))) {
134 			pr_err("%s: section %u reloc %u: bad relocation sym offset\n",
135 				module->name, relindex, i);
136 			return -ENOEXEC;
137 		}
138 
139 		sym = ((Elf32_Sym *)symsec->sh_addr) + offset;
140 		symname = strtab + sym->st_name;
141 
142 		if (rel->r_offset < 0 || rel->r_offset > dstsec->sh_size - sizeof(u32)) {
143 			pr_err("%s: section %u reloc %u sym '%s': out of bounds relocation, offset %d size %u\n",
144 			       module->name, relindex, i, symname,
145 			       rel->r_offset, dstsec->sh_size);
146 			return -ENOEXEC;
147 		}
148 
149 		loc = dstsec->sh_addr + rel->r_offset;
150 
151 		switch (ELF32_R_TYPE(rel->r_info)) {
152 		case R_ARM_NONE:
153 			/* ignore */
154 			break;
155 
156 		case R_ARM_ABS32:
157 		case R_ARM_TARGET1:
158 			*(u32 *)loc += sym->st_value;
159 			break;
160 
161 		case R_ARM_PC24:
162 		case R_ARM_CALL:
163 		case R_ARM_JUMP24:
164 			if (sym->st_value & 3) {
165 				pr_err("%s: section %u reloc %u sym '%s': unsupported interworking call (ARM -> Thumb)\n",
166 				       module->name, relindex, i, symname);
167 				return -ENOEXEC;
168 			}
169 
170 			offset = __mem_to_opcode_arm(*(u32 *)loc);
171 			offset = (offset & 0x00ffffff) << 2;
172 			if (offset & 0x02000000)
173 				offset -= 0x04000000;
174 
175 			offset += sym->st_value - loc;
176 
177 			/*
178 			 * Route through a PLT entry if 'offset' exceeds the
179 			 * supported range. Note that 'offset + loc + 8'
180 			 * contains the absolute jump target, i.e.,
181 			 * @sym + addend, corrected for the +8 PC bias.
182 			 */
183 			if (IS_ENABLED(CONFIG_ARM_MODULE_PLTS) &&
184 			    (offset <= (s32)0xfe000000 ||
185 			     offset >= (s32)0x02000000))
186 				offset = get_module_plt(module, loc,
187 							offset + loc + 8)
188 					 - loc - 8;
189 
190 			if (offset <= (s32)0xfe000000 ||
191 			    offset >= (s32)0x02000000) {
192 				pr_err("%s: section %u reloc %u sym '%s': relocation %u out of range (%#lx -> %#x)\n",
193 				       module->name, relindex, i, symname,
194 				       ELF32_R_TYPE(rel->r_info), loc,
195 				       sym->st_value);
196 				return -ENOEXEC;
197 			}
198 
199 			offset >>= 2;
200 			offset &= 0x00ffffff;
201 
202 			*(u32 *)loc &= __opcode_to_mem_arm(0xff000000);
203 			*(u32 *)loc |= __opcode_to_mem_arm(offset);
204 			break;
205 
206 	       case R_ARM_V4BX:
207 		       /* Preserve Rm and the condition code. Alter
208 			* other bits to re-code instruction as
209 			* MOV PC,Rm.
210 			*/
211 		       *(u32 *)loc &= __opcode_to_mem_arm(0xf000000f);
212 		       *(u32 *)loc |= __opcode_to_mem_arm(0x01a0f000);
213 		       break;
214 
215 		case R_ARM_PREL31:
216 			offset = (*(s32 *)loc << 1) >> 1; /* sign extend */
217 			offset += sym->st_value - loc;
218 			if (offset >= 0x40000000 || offset < -0x40000000) {
219 				pr_err("%s: section %u reloc %u sym '%s': relocation %u out of range (%#lx -> %#x)\n",
220 				       module->name, relindex, i, symname,
221 				       ELF32_R_TYPE(rel->r_info), loc,
222 				       sym->st_value);
223 				return -ENOEXEC;
224 			}
225 			*(u32 *)loc &= 0x80000000;
226 			*(u32 *)loc |= offset & 0x7fffffff;
227 			break;
228 
229 		case R_ARM_REL32:
230 			*(u32 *)loc += sym->st_value - loc;
231 			break;
232 
233 		case R_ARM_MOVW_ABS_NC:
234 		case R_ARM_MOVT_ABS:
235 		case R_ARM_MOVW_PREL_NC:
236 		case R_ARM_MOVT_PREL:
237 			offset = tmp = __mem_to_opcode_arm(*(u32 *)loc);
238 			offset = ((offset & 0xf0000) >> 4) | (offset & 0xfff);
239 			offset = (offset ^ 0x8000) - 0x8000;
240 
241 			offset += sym->st_value;
242 			if (ELF32_R_TYPE(rel->r_info) == R_ARM_MOVT_PREL ||
243 			    ELF32_R_TYPE(rel->r_info) == R_ARM_MOVW_PREL_NC)
244 				offset -= loc;
245 			if (ELF32_R_TYPE(rel->r_info) == R_ARM_MOVT_ABS ||
246 			    ELF32_R_TYPE(rel->r_info) == R_ARM_MOVT_PREL)
247 				offset >>= 16;
248 
249 			tmp &= 0xfff0f000;
250 			tmp |= ((offset & 0xf000) << 4) |
251 				(offset & 0x0fff);
252 
253 			*(u32 *)loc = __opcode_to_mem_arm(tmp);
254 			break;
255 
256 #ifdef CONFIG_ARM_HAS_GROUP_RELOCS
257 		case R_ARM_ALU_PC_G0_NC:
258 			group = 0;
259 			fallthrough;
260 		case R_ARM_ALU_PC_G1_NC:
261 			tmp = __mem_to_opcode_arm(*(u32 *)loc);
262 			offset = ror32(tmp & 0xff, (tmp & 0xf00) >> 7);
263 			if (tmp & BIT(22))
264 				offset = -offset;
265 			offset += sym->st_value - loc;
266 			if (offset < 0) {
267 				offset = -offset;
268 				tmp = (tmp & ~BIT(23)) | BIT(22); // SUB opcode
269 			} else {
270 				tmp = (tmp & ~BIT(22)) | BIT(23); // ADD opcode
271 			}
272 
273 			shift = get_group_rem(group, &offset);
274 			if (shift < 24) {
275 				offset >>= 24 - shift;
276 				offset |= (shift + 8) << 7;
277 			}
278 			*(u32 *)loc = __opcode_to_mem_arm((tmp & ~0xfff) | offset);
279 			break;
280 
281 		case R_ARM_LDR_PC_G2:
282 			tmp = __mem_to_opcode_arm(*(u32 *)loc);
283 			offset = tmp & 0xfff;
284 			if (~tmp & BIT(23))		// U bit cleared?
285 				offset = -offset;
286 			offset += sym->st_value - loc;
287 			if (offset < 0) {
288 				offset = -offset;
289 				tmp &= ~BIT(23);	// clear U bit
290 			} else {
291 				tmp |= BIT(23);		// set U bit
292 			}
293 			get_group_rem(2, &offset);
294 
295 			if (offset > 0xfff) {
296 				pr_err("%s: section %u reloc %u sym '%s': relocation %u out of range (%#lx -> %#x)\n",
297 				       module->name, relindex, i, symname,
298 				       ELF32_R_TYPE(rel->r_info), loc,
299 				       sym->st_value);
300 				return -ENOEXEC;
301 			}
302 			*(u32 *)loc = __opcode_to_mem_arm((tmp & ~0xfff) | offset);
303 			break;
304 #endif
305 #ifdef CONFIG_THUMB2_KERNEL
306 		case R_ARM_THM_CALL:
307 		case R_ARM_THM_JUMP24:
308 			/*
309 			 * For function symbols, only Thumb addresses are
310 			 * allowed (no interworking).
311 			 *
312 			 * For non-function symbols, the destination
313 			 * has no specific ARM/Thumb disposition, so
314 			 * the branch is resolved under the assumption
315 			 * that interworking is not required.
316 			 */
317 			if (ELF32_ST_TYPE(sym->st_info) == STT_FUNC &&
318 			    !(sym->st_value & 1)) {
319 				pr_err("%s: section %u reloc %u sym '%s': unsupported interworking call (Thumb -> ARM)\n",
320 				       module->name, relindex, i, symname);
321 				return -ENOEXEC;
322 			}
323 
324 			upper = __mem_to_opcode_thumb16(*(u16 *)loc);
325 			lower = __mem_to_opcode_thumb16(*(u16 *)(loc + 2));
326 
327 			/*
328 			 * 25 bit signed address range (Thumb-2 BL and B.W
329 			 * instructions):
330 			 *   S:I1:I2:imm10:imm11:0
331 			 * where:
332 			 *   S     = upper[10]   = offset[24]
333 			 *   I1    = ~(J1 ^ S)   = offset[23]
334 			 *   I2    = ~(J2 ^ S)   = offset[22]
335 			 *   imm10 = upper[9:0]  = offset[21:12]
336 			 *   imm11 = lower[10:0] = offset[11:1]
337 			 *   J1    = lower[13]
338 			 *   J2    = lower[11]
339 			 */
340 			sign = (upper >> 10) & 1;
341 			j1 = (lower >> 13) & 1;
342 			j2 = (lower >> 11) & 1;
343 			offset = (sign << 24) | ((~(j1 ^ sign) & 1) << 23) |
344 				((~(j2 ^ sign) & 1) << 22) |
345 				((upper & 0x03ff) << 12) |
346 				((lower & 0x07ff) << 1);
347 			if (offset & 0x01000000)
348 				offset -= 0x02000000;
349 			offset += sym->st_value - loc;
350 
351 			/*
352 			 * Route through a PLT entry if 'offset' exceeds the
353 			 * supported range.
354 			 */
355 			if (IS_ENABLED(CONFIG_ARM_MODULE_PLTS) &&
356 			    (offset <= (s32)0xff000000 ||
357 			     offset >= (s32)0x01000000))
358 				offset = get_module_plt(module, loc,
359 							offset + loc + 4)
360 					 - loc - 4;
361 
362 			if (offset <= (s32)0xff000000 ||
363 			    offset >= (s32)0x01000000) {
364 				pr_err("%s: section %u reloc %u sym '%s': relocation %u out of range (%#lx -> %#x)\n",
365 				       module->name, relindex, i, symname,
366 				       ELF32_R_TYPE(rel->r_info), loc,
367 				       sym->st_value);
368 				return -ENOEXEC;
369 			}
370 
371 			sign = (offset >> 24) & 1;
372 			j1 = sign ^ (~(offset >> 23) & 1);
373 			j2 = sign ^ (~(offset >> 22) & 1);
374 			upper = (u16)((upper & 0xf800) | (sign << 10) |
375 					    ((offset >> 12) & 0x03ff));
376 			lower = (u16)((lower & 0xd000) |
377 				      (j1 << 13) | (j2 << 11) |
378 				      ((offset >> 1) & 0x07ff));
379 
380 			*(u16 *)loc = __opcode_to_mem_thumb16(upper);
381 			*(u16 *)(loc + 2) = __opcode_to_mem_thumb16(lower);
382 			break;
383 
384 		case R_ARM_THM_MOVW_ABS_NC:
385 		case R_ARM_THM_MOVT_ABS:
386 		case R_ARM_THM_MOVW_PREL_NC:
387 		case R_ARM_THM_MOVT_PREL:
388 			upper = __mem_to_opcode_thumb16(*(u16 *)loc);
389 			lower = __mem_to_opcode_thumb16(*(u16 *)(loc + 2));
390 
391 			/*
392 			 * MOVT/MOVW instructions encoding in Thumb-2:
393 			 *
394 			 * i	= upper[10]
395 			 * imm4	= upper[3:0]
396 			 * imm3	= lower[14:12]
397 			 * imm8	= lower[7:0]
398 			 *
399 			 * imm16 = imm4:i:imm3:imm8
400 			 */
401 			offset = ((upper & 0x000f) << 12) |
402 				((upper & 0x0400) << 1) |
403 				((lower & 0x7000) >> 4) | (lower & 0x00ff);
404 			offset = (offset ^ 0x8000) - 0x8000;
405 			offset += sym->st_value;
406 
407 			if (ELF32_R_TYPE(rel->r_info) == R_ARM_THM_MOVT_PREL ||
408 			    ELF32_R_TYPE(rel->r_info) == R_ARM_THM_MOVW_PREL_NC)
409 				offset -= loc;
410 			if (ELF32_R_TYPE(rel->r_info) == R_ARM_THM_MOVT_ABS ||
411 			    ELF32_R_TYPE(rel->r_info) == R_ARM_THM_MOVT_PREL)
412 				offset >>= 16;
413 
414 			upper = (u16)((upper & 0xfbf0) |
415 				      ((offset & 0xf000) >> 12) |
416 				      ((offset & 0x0800) >> 1));
417 			lower = (u16)((lower & 0x8f00) |
418 				      ((offset & 0x0700) << 4) |
419 				      (offset & 0x00ff));
420 			*(u16 *)loc = __opcode_to_mem_thumb16(upper);
421 			*(u16 *)(loc + 2) = __opcode_to_mem_thumb16(lower);
422 			break;
423 #endif
424 
425 		default:
426 			pr_err("%s: unknown relocation: %u\n",
427 			       module->name, ELF32_R_TYPE(rel->r_info));
428 			return -ENOEXEC;
429 		}
430 	}
431 	return 0;
432 }
433 
434 struct mod_unwind_map {
435 	const Elf_Shdr *unw_sec;
436 	const Elf_Shdr *txt_sec;
437 };
438 
439 static const Elf_Shdr *find_mod_section(const Elf32_Ehdr *hdr,
440 	const Elf_Shdr *sechdrs, const char *name)
441 {
442 	const Elf_Shdr *s, *se;
443 	const char *secstrs = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
444 
445 	for (s = sechdrs, se = sechdrs + hdr->e_shnum; s < se; s++)
446 		if (strcmp(name, secstrs + s->sh_name) == 0)
447 			return s;
448 
449 	return NULL;
450 }
451 
452 extern void fixup_pv_table(const void *, unsigned long);
453 extern void fixup_smp(const void *, unsigned long);
454 
455 int module_finalize(const Elf32_Ehdr *hdr, const Elf_Shdr *sechdrs,
456 		    struct module *mod)
457 {
458 	const Elf_Shdr *s = NULL;
459 #ifdef CONFIG_ARM_UNWIND
460 	const char *secstrs = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
461 	const Elf_Shdr *sechdrs_end = sechdrs + hdr->e_shnum;
462 	struct mod_unwind_map maps[ARM_SEC_MAX];
463 	int i;
464 
465 	memset(maps, 0, sizeof(maps));
466 
467 	for (s = sechdrs; s < sechdrs_end; s++) {
468 		const char *secname = secstrs + s->sh_name;
469 
470 		if (!(s->sh_flags & SHF_ALLOC))
471 			continue;
472 
473 		if (strcmp(".ARM.exidx.init.text", secname) == 0)
474 			maps[ARM_SEC_INIT].unw_sec = s;
475 		else if (strcmp(".ARM.exidx", secname) == 0)
476 			maps[ARM_SEC_CORE].unw_sec = s;
477 		else if (strcmp(".ARM.exidx.exit.text", secname) == 0)
478 			maps[ARM_SEC_EXIT].unw_sec = s;
479 		else if (strcmp(".ARM.exidx.text.unlikely", secname) == 0)
480 			maps[ARM_SEC_UNLIKELY].unw_sec = s;
481 		else if (strcmp(".ARM.exidx.text.hot", secname) == 0)
482 			maps[ARM_SEC_HOT].unw_sec = s;
483 		else if (strcmp(".init.text", secname) == 0)
484 			maps[ARM_SEC_INIT].txt_sec = s;
485 		else if (strcmp(".text", secname) == 0)
486 			maps[ARM_SEC_CORE].txt_sec = s;
487 		else if (strcmp(".exit.text", secname) == 0)
488 			maps[ARM_SEC_EXIT].txt_sec = s;
489 		else if (strcmp(".text.unlikely", secname) == 0)
490 			maps[ARM_SEC_UNLIKELY].txt_sec = s;
491 		else if (strcmp(".text.hot", secname) == 0)
492 			maps[ARM_SEC_HOT].txt_sec = s;
493 	}
494 
495 	for (i = 0; i < ARM_SEC_MAX; i++)
496 		if (maps[i].unw_sec && maps[i].txt_sec)
497 			mod->arch.unwind[i] =
498 				unwind_table_add(maps[i].unw_sec->sh_addr,
499 					         maps[i].unw_sec->sh_size,
500 					         maps[i].txt_sec->sh_addr,
501 					         maps[i].txt_sec->sh_size);
502 #endif
503 #ifdef CONFIG_ARM_PATCH_PHYS_VIRT
504 	s = find_mod_section(hdr, sechdrs, ".pv_table");
505 	if (s)
506 		fixup_pv_table((void *)s->sh_addr, s->sh_size);
507 #endif
508 	s = find_mod_section(hdr, sechdrs, ".alt.smp.init");
509 	if (s && !is_smp())
510 #ifdef CONFIG_SMP_ON_UP
511 		fixup_smp((void *)s->sh_addr, s->sh_size);
512 #else
513 		return -EINVAL;
514 #endif
515 	return 0;
516 }
517 
518 void
519 module_arch_cleanup(struct module *mod)
520 {
521 #ifdef CONFIG_ARM_UNWIND
522 	int i;
523 
524 	for (i = 0; i < ARM_SEC_MAX; i++) {
525 		unwind_table_del(mod->arch.unwind[i]);
526 		mod->arch.unwind[i] = NULL;
527 	}
528 #endif
529 }
530 
531 void __weak module_arch_freeing_init(struct module *mod)
532 {
533 #ifdef CONFIG_ARM_UNWIND
534 	unwind_table_del(mod->arch.unwind[ARM_SEC_INIT]);
535 	mod->arch.unwind[ARM_SEC_INIT] = NULL;
536 #endif
537 }
538