xref: /openbmc/linux/arch/powerpc/kernel/module_32.c (revision 8e8e69d6)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*  Kernel module help for PPC.
3     Copyright (C) 2001 Rusty Russell.
4 
5 */
6 
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 
9 #include <linux/module.h>
10 #include <linux/moduleloader.h>
11 #include <linux/elf.h>
12 #include <linux/vmalloc.h>
13 #include <linux/fs.h>
14 #include <linux/string.h>
15 #include <linux/kernel.h>
16 #include <linux/ftrace.h>
17 #include <linux/cache.h>
18 #include <linux/bug.h>
19 #include <linux/sort.h>
20 #include <asm/setup.h>
21 
22 /* Count how many different relocations (different symbol, different
23    addend) */
24 static unsigned int count_relocs(const Elf32_Rela *rela, unsigned int num)
25 {
26 	unsigned int i, r_info, r_addend, _count_relocs;
27 
28 	_count_relocs = 0;
29 	r_info = 0;
30 	r_addend = 0;
31 	for (i = 0; i < num; i++)
32 		/* Only count 24-bit relocs, others don't need stubs */
33 		if (ELF32_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
34 		    (r_info != ELF32_R_SYM(rela[i].r_info) ||
35 		     r_addend != rela[i].r_addend)) {
36 			_count_relocs++;
37 			r_info = ELF32_R_SYM(rela[i].r_info);
38 			r_addend = rela[i].r_addend;
39 		}
40 
41 #ifdef CONFIG_DYNAMIC_FTRACE
42 	_count_relocs++;	/* add one for ftrace_caller */
43 #endif
44 	return _count_relocs;
45 }
46 
47 static int relacmp(const void *_x, const void *_y)
48 {
49 	const Elf32_Rela *x, *y;
50 
51 	y = (Elf32_Rela *)_x;
52 	x = (Elf32_Rela *)_y;
53 
54 	/* Compare the entire r_info (as opposed to ELF32_R_SYM(r_info) only) to
55 	 * make the comparison cheaper/faster. It won't affect the sorting or
56 	 * the counting algorithms' performance
57 	 */
58 	if (x->r_info < y->r_info)
59 		return -1;
60 	else if (x->r_info > y->r_info)
61 		return 1;
62 	else if (x->r_addend < y->r_addend)
63 		return -1;
64 	else if (x->r_addend > y->r_addend)
65 		return 1;
66 	else
67 		return 0;
68 }
69 
70 static void relaswap(void *_x, void *_y, int size)
71 {
72 	uint32_t *x, *y, tmp;
73 	int i;
74 
75 	y = (uint32_t *)_x;
76 	x = (uint32_t *)_y;
77 
78 	for (i = 0; i < sizeof(Elf32_Rela) / sizeof(uint32_t); i++) {
79 		tmp = x[i];
80 		x[i] = y[i];
81 		y[i] = tmp;
82 	}
83 }
84 
85 /* Get the potential trampolines size required of the init and
86    non-init sections */
87 static unsigned long get_plt_size(const Elf32_Ehdr *hdr,
88 				  const Elf32_Shdr *sechdrs,
89 				  const char *secstrings,
90 				  int is_init)
91 {
92 	unsigned long ret = 0;
93 	unsigned i;
94 
95 	/* Everything marked ALLOC (this includes the exported
96            symbols) */
97 	for (i = 1; i < hdr->e_shnum; i++) {
98 		/* If it's called *.init*, and we're not init, we're
99                    not interested */
100 		if ((strstr(secstrings + sechdrs[i].sh_name, ".init") != NULL)
101 		    != is_init)
102 			continue;
103 
104 		/* We don't want to look at debug sections. */
105 		if (strstr(secstrings + sechdrs[i].sh_name, ".debug"))
106 			continue;
107 
108 		if (sechdrs[i].sh_type == SHT_RELA) {
109 			pr_debug("Found relocations in section %u\n", i);
110 			pr_debug("Ptr: %p.  Number: %u\n",
111 			       (void *)hdr + sechdrs[i].sh_offset,
112 			       sechdrs[i].sh_size / sizeof(Elf32_Rela));
113 
114 			/* Sort the relocation information based on a symbol and
115 			 * addend key. This is a stable O(n*log n) complexity
116 			 * alogrithm but it will reduce the complexity of
117 			 * count_relocs() to linear complexity O(n)
118 			 */
119 			sort((void *)hdr + sechdrs[i].sh_offset,
120 			     sechdrs[i].sh_size / sizeof(Elf32_Rela),
121 			     sizeof(Elf32_Rela), relacmp, relaswap);
122 
123 			ret += count_relocs((void *)hdr
124 					     + sechdrs[i].sh_offset,
125 					     sechdrs[i].sh_size
126 					     / sizeof(Elf32_Rela))
127 				* sizeof(struct ppc_plt_entry);
128 		}
129 	}
130 
131 	return ret;
132 }
133 
134 int module_frob_arch_sections(Elf32_Ehdr *hdr,
135 			      Elf32_Shdr *sechdrs,
136 			      char *secstrings,
137 			      struct module *me)
138 {
139 	unsigned int i;
140 
141 	/* Find .plt and .init.plt sections */
142 	for (i = 0; i < hdr->e_shnum; i++) {
143 		if (strcmp(secstrings + sechdrs[i].sh_name, ".init.plt") == 0)
144 			me->arch.init_plt_section = i;
145 		else if (strcmp(secstrings + sechdrs[i].sh_name, ".plt") == 0)
146 			me->arch.core_plt_section = i;
147 	}
148 	if (!me->arch.core_plt_section || !me->arch.init_plt_section) {
149 		pr_err("Module doesn't contain .plt or .init.plt sections.\n");
150 		return -ENOEXEC;
151 	}
152 
153 	/* Override their sizes */
154 	sechdrs[me->arch.core_plt_section].sh_size
155 		= get_plt_size(hdr, sechdrs, secstrings, 0);
156 	sechdrs[me->arch.init_plt_section].sh_size
157 		= get_plt_size(hdr, sechdrs, secstrings, 1);
158 	return 0;
159 }
160 
161 static inline int entry_matches(struct ppc_plt_entry *entry, Elf32_Addr val)
162 {
163 	if (entry->jump[0] == 0x3d800000 + ((val + 0x8000) >> 16)
164 	    && entry->jump[1] == 0x398c0000 + (val & 0xffff))
165 		return 1;
166 	return 0;
167 }
168 
169 /* Set up a trampoline in the PLT to bounce us to the distant function */
170 static uint32_t do_plt_call(void *location,
171 			    Elf32_Addr val,
172 			    const Elf32_Shdr *sechdrs,
173 			    struct module *mod)
174 {
175 	struct ppc_plt_entry *entry;
176 
177 	pr_debug("Doing plt for call to 0x%x at 0x%x\n", val, (unsigned int)location);
178 	/* Init, or core PLT? */
179 	if (location >= mod->core_layout.base
180 	    && location < mod->core_layout.base + mod->core_layout.size)
181 		entry = (void *)sechdrs[mod->arch.core_plt_section].sh_addr;
182 	else
183 		entry = (void *)sechdrs[mod->arch.init_plt_section].sh_addr;
184 
185 	/* Find this entry, or if that fails, the next avail. entry */
186 	while (entry->jump[0]) {
187 		if (entry_matches(entry, val)) return (uint32_t)entry;
188 		entry++;
189 	}
190 
191 	entry->jump[0] = 0x3d800000+((val+0x8000)>>16); /* lis r12,sym@ha */
192 	entry->jump[1] = 0x398c0000 + (val&0xffff);     /* addi r12,r12,sym@l*/
193 	entry->jump[2] = 0x7d8903a6;                    /* mtctr r12 */
194 	entry->jump[3] = 0x4e800420;			/* bctr */
195 
196 	pr_debug("Initialized plt for 0x%x at %p\n", val, entry);
197 	return (uint32_t)entry;
198 }
199 
200 int apply_relocate_add(Elf32_Shdr *sechdrs,
201 		       const char *strtab,
202 		       unsigned int symindex,
203 		       unsigned int relsec,
204 		       struct module *module)
205 {
206 	unsigned int i;
207 	Elf32_Rela *rela = (void *)sechdrs[relsec].sh_addr;
208 	Elf32_Sym *sym;
209 	uint32_t *location;
210 	uint32_t value;
211 
212 	pr_debug("Applying ADD relocate section %u to %u\n", relsec,
213 	       sechdrs[relsec].sh_info);
214 	for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
215 		/* This is where to make the change */
216 		location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
217 			+ rela[i].r_offset;
218 		/* This is the symbol it is referring to.  Note that all
219 		   undefined symbols have been resolved.  */
220 		sym = (Elf32_Sym *)sechdrs[symindex].sh_addr
221 			+ ELF32_R_SYM(rela[i].r_info);
222 		/* `Everything is relative'. */
223 		value = sym->st_value + rela[i].r_addend;
224 
225 		switch (ELF32_R_TYPE(rela[i].r_info)) {
226 		case R_PPC_ADDR32:
227 			/* Simply set it */
228 			*(uint32_t *)location = value;
229 			break;
230 
231 		case R_PPC_ADDR16_LO:
232 			/* Low half of the symbol */
233 			*(uint16_t *)location = value;
234 			break;
235 
236 		case R_PPC_ADDR16_HI:
237 			/* Higher half of the symbol */
238 			*(uint16_t *)location = (value >> 16);
239 			break;
240 
241 		case R_PPC_ADDR16_HA:
242 			/* Sign-adjusted lower 16 bits: PPC ELF ABI says:
243 			   (((x >> 16) + ((x & 0x8000) ? 1 : 0))) & 0xFFFF.
244 			   This is the same, only sane.
245 			 */
246 			*(uint16_t *)location = (value + 0x8000) >> 16;
247 			break;
248 
249 		case R_PPC_REL24:
250 			if ((int)(value - (uint32_t)location) < -0x02000000
251 			    || (int)(value - (uint32_t)location) >= 0x02000000)
252 				value = do_plt_call(location, value,
253 						    sechdrs, module);
254 
255 			/* Only replace bits 2 through 26 */
256 			pr_debug("REL24 value = %08X. location = %08X\n",
257 			       value, (uint32_t)location);
258 			pr_debug("Location before: %08X.\n",
259 			       *(uint32_t *)location);
260 			*(uint32_t *)location
261 				= (*(uint32_t *)location & ~0x03fffffc)
262 				| ((value - (uint32_t)location)
263 				   & 0x03fffffc);
264 			pr_debug("Location after: %08X.\n",
265 			       *(uint32_t *)location);
266 			pr_debug("ie. jump to %08X+%08X = %08X\n",
267 			       *(uint32_t *)location & 0x03fffffc,
268 			       (uint32_t)location,
269 			       (*(uint32_t *)location & 0x03fffffc)
270 			       + (uint32_t)location);
271 			break;
272 
273 		case R_PPC_REL32:
274 			/* 32-bit relative jump. */
275 			*(uint32_t *)location = value - (uint32_t)location;
276 			break;
277 
278 		default:
279 			pr_err("%s: unknown ADD relocation: %u\n",
280 			       module->name,
281 			       ELF32_R_TYPE(rela[i].r_info));
282 			return -ENOEXEC;
283 		}
284 	}
285 
286 	return 0;
287 }
288 
289 #ifdef CONFIG_DYNAMIC_FTRACE
290 int module_finalize_ftrace(struct module *module, const Elf_Shdr *sechdrs)
291 {
292 	module->arch.tramp = do_plt_call(module->core_layout.base,
293 					 (unsigned long)ftrace_caller,
294 					 sechdrs, module);
295 	if (!module->arch.tramp)
296 		return -ENOENT;
297 
298 	return 0;
299 }
300 #endif
301