xref: /openbmc/linux/arch/powerpc/kernel/module_32.c (revision 55eb9a6c)
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 #include <asm/code-patching.h>
22 
23 /* Count how many different relocations (different symbol, different
24    addend) */
25 static unsigned int count_relocs(const Elf32_Rela *rela, unsigned int num)
26 {
27 	unsigned int i, r_info, r_addend, _count_relocs;
28 
29 	_count_relocs = 0;
30 	r_info = 0;
31 	r_addend = 0;
32 	for (i = 0; i < num; i++)
33 		/* Only count 24-bit relocs, others don't need stubs */
34 		if (ELF32_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
35 		    (r_info != ELF32_R_SYM(rela[i].r_info) ||
36 		     r_addend != rela[i].r_addend)) {
37 			_count_relocs++;
38 			r_info = ELF32_R_SYM(rela[i].r_info);
39 			r_addend = rela[i].r_addend;
40 		}
41 
42 #ifdef CONFIG_DYNAMIC_FTRACE
43 	_count_relocs++;	/* add one for ftrace_caller */
44 #endif
45 	return _count_relocs;
46 }
47 
48 static int relacmp(const void *_x, const void *_y)
49 {
50 	const Elf32_Rela *x, *y;
51 
52 	y = (Elf32_Rela *)_x;
53 	x = (Elf32_Rela *)_y;
54 
55 	/* Compare the entire r_info (as opposed to ELF32_R_SYM(r_info) only) to
56 	 * make the comparison cheaper/faster. It won't affect the sorting or
57 	 * the counting algorithms' performance
58 	 */
59 	if (x->r_info < y->r_info)
60 		return -1;
61 	else if (x->r_info > y->r_info)
62 		return 1;
63 	else if (x->r_addend < y->r_addend)
64 		return -1;
65 	else if (x->r_addend > y->r_addend)
66 		return 1;
67 	else
68 		return 0;
69 }
70 
71 /* Get the potential trampolines size required of the init and
72    non-init sections */
73 static unsigned long get_plt_size(const Elf32_Ehdr *hdr,
74 				  const Elf32_Shdr *sechdrs,
75 				  const char *secstrings,
76 				  int is_init)
77 {
78 	unsigned long ret = 0;
79 	unsigned i;
80 
81 	/* Everything marked ALLOC (this includes the exported
82            symbols) */
83 	for (i = 1; i < hdr->e_shnum; i++) {
84 		/* If it's called *.init*, and we're not init, we're
85                    not interested */
86 		if ((strstr(secstrings + sechdrs[i].sh_name, ".init") != NULL)
87 		    != is_init)
88 			continue;
89 
90 		/* We don't want to look at debug sections. */
91 		if (strstr(secstrings + sechdrs[i].sh_name, ".debug"))
92 			continue;
93 
94 		if (sechdrs[i].sh_type == SHT_RELA) {
95 			pr_debug("Found relocations in section %u\n", i);
96 			pr_debug("Ptr: %p.  Number: %u\n",
97 			       (void *)hdr + sechdrs[i].sh_offset,
98 			       sechdrs[i].sh_size / sizeof(Elf32_Rela));
99 
100 			/* Sort the relocation information based on a symbol and
101 			 * addend key. This is a stable O(n*log n) complexity
102 			 * algorithm but it will reduce the complexity of
103 			 * count_relocs() to linear complexity O(n)
104 			 */
105 			sort((void *)hdr + sechdrs[i].sh_offset,
106 			     sechdrs[i].sh_size / sizeof(Elf32_Rela),
107 			     sizeof(Elf32_Rela), relacmp, NULL);
108 
109 			ret += count_relocs((void *)hdr
110 					     + sechdrs[i].sh_offset,
111 					     sechdrs[i].sh_size
112 					     / sizeof(Elf32_Rela))
113 				* sizeof(struct ppc_plt_entry);
114 		}
115 	}
116 
117 	return ret;
118 }
119 
120 int module_frob_arch_sections(Elf32_Ehdr *hdr,
121 			      Elf32_Shdr *sechdrs,
122 			      char *secstrings,
123 			      struct module *me)
124 {
125 	unsigned int i;
126 
127 	/* Find .plt and .init.plt sections */
128 	for (i = 0; i < hdr->e_shnum; i++) {
129 		if (strcmp(secstrings + sechdrs[i].sh_name, ".init.plt") == 0)
130 			me->arch.init_plt_section = i;
131 		else if (strcmp(secstrings + sechdrs[i].sh_name, ".plt") == 0)
132 			me->arch.core_plt_section = i;
133 	}
134 	if (!me->arch.core_plt_section || !me->arch.init_plt_section) {
135 		pr_err("Module doesn't contain .plt or .init.plt sections.\n");
136 		return -ENOEXEC;
137 	}
138 
139 	/* Override their sizes */
140 	sechdrs[me->arch.core_plt_section].sh_size
141 		= get_plt_size(hdr, sechdrs, secstrings, 0);
142 	sechdrs[me->arch.init_plt_section].sh_size
143 		= get_plt_size(hdr, sechdrs, secstrings, 1);
144 	return 0;
145 }
146 
147 static inline int entry_matches(struct ppc_plt_entry *entry, Elf32_Addr val)
148 {
149 	if (entry->jump[0] != PPC_RAW_LIS(_R12, PPC_HA(val)))
150 		return 0;
151 	if (entry->jump[1] != PPC_RAW_ADDI(_R12, _R12, PPC_LO(val)))
152 		return 0;
153 	return 1;
154 }
155 
156 /* Set up a trampoline in the PLT to bounce us to the distant function */
157 static uint32_t do_plt_call(void *location,
158 			    Elf32_Addr val,
159 			    const Elf32_Shdr *sechdrs,
160 			    struct module *mod)
161 {
162 	struct ppc_plt_entry *entry;
163 
164 	pr_debug("Doing plt for call to 0x%x at 0x%x\n", val, (unsigned int)location);
165 	/* Init, or core PLT? */
166 	if (location >= mod->core_layout.base
167 	    && location < mod->core_layout.base + mod->core_layout.size)
168 		entry = (void *)sechdrs[mod->arch.core_plt_section].sh_addr;
169 	else
170 		entry = (void *)sechdrs[mod->arch.init_plt_section].sh_addr;
171 
172 	/* Find this entry, or if that fails, the next avail. entry */
173 	while (entry->jump[0]) {
174 		if (entry_matches(entry, val)) return (uint32_t)entry;
175 		entry++;
176 	}
177 
178 	if (patch_instruction(&entry->jump[0], ppc_inst(PPC_RAW_LIS(_R12, PPC_HA(val)))))
179 		return 0;
180 	if (patch_instruction(&entry->jump[1], ppc_inst(PPC_RAW_ADDI(_R12, _R12, PPC_LO(val)))))
181 		return 0;
182 	if (patch_instruction(&entry->jump[2], ppc_inst(PPC_RAW_MTCTR(_R12))))
183 		return 0;
184 	if (patch_instruction(&entry->jump[3], ppc_inst(PPC_RAW_BCTR())))
185 		return 0;
186 
187 	pr_debug("Initialized plt for 0x%x at %p\n", val, entry);
188 	return (uint32_t)entry;
189 }
190 
191 static int patch_location_16(uint32_t *loc, u16 value)
192 {
193 	loc = PTR_ALIGN_DOWN(loc, sizeof(u32));
194 	return patch_instruction(loc, ppc_inst((*loc & 0xffff0000) | value));
195 }
196 
197 int apply_relocate_add(Elf32_Shdr *sechdrs,
198 		       const char *strtab,
199 		       unsigned int symindex,
200 		       unsigned int relsec,
201 		       struct module *module)
202 {
203 	unsigned int i;
204 	Elf32_Rela *rela = (void *)sechdrs[relsec].sh_addr;
205 	Elf32_Sym *sym;
206 	uint32_t *location;
207 	uint32_t value;
208 
209 	pr_debug("Applying ADD relocate section %u to %u\n", relsec,
210 	       sechdrs[relsec].sh_info);
211 	for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
212 		/* This is where to make the change */
213 		location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
214 			+ rela[i].r_offset;
215 		/* This is the symbol it is referring to.  Note that all
216 		   undefined symbols have been resolved.  */
217 		sym = (Elf32_Sym *)sechdrs[symindex].sh_addr
218 			+ ELF32_R_SYM(rela[i].r_info);
219 		/* `Everything is relative'. */
220 		value = sym->st_value + rela[i].r_addend;
221 
222 		switch (ELF32_R_TYPE(rela[i].r_info)) {
223 		case R_PPC_ADDR32:
224 			/* Simply set it */
225 			*(uint32_t *)location = value;
226 			break;
227 
228 		case R_PPC_ADDR16_LO:
229 			/* Low half of the symbol */
230 			if (patch_location_16(location, PPC_LO(value)))
231 				return -EFAULT;
232 			break;
233 
234 		case R_PPC_ADDR16_HI:
235 			/* Higher half of the symbol */
236 			if (patch_location_16(location, PPC_HI(value)))
237 				return -EFAULT;
238 			break;
239 
240 		case R_PPC_ADDR16_HA:
241 			if (patch_location_16(location, PPC_HA(value)))
242 				return -EFAULT;
243 			break;
244 
245 		case R_PPC_REL24:
246 			if ((int)(value - (uint32_t)location) < -0x02000000
247 			    || (int)(value - (uint32_t)location) >= 0x02000000) {
248 				value = do_plt_call(location, value,
249 						    sechdrs, module);
250 				if (!value)
251 					return -EFAULT;
252 			}
253 
254 			/* Only replace bits 2 through 26 */
255 			pr_debug("REL24 value = %08X. location = %08X\n",
256 			       value, (uint32_t)location);
257 			pr_debug("Location before: %08X.\n",
258 			       *(uint32_t *)location);
259 			value = (*(uint32_t *)location & ~PPC_LI_MASK) |
260 				PPC_LI(value - (uint32_t)location);
261 
262 			if (patch_instruction(location, ppc_inst(value)))
263 				return -EFAULT;
264 
265 			pr_debug("Location after: %08X.\n",
266 			       *(uint32_t *)location);
267 			pr_debug("ie. jump to %08X+%08X = %08X\n",
268 				 *(uint32_t *)PPC_LI((uint32_t)location), (uint32_t)location,
269 				 (*(uint32_t *)PPC_LI((uint32_t)location)) + (uint32_t)location);
270 			break;
271 
272 		case R_PPC_REL32:
273 			/* 32-bit relative jump. */
274 			*(uint32_t *)location = value - (uint32_t)location;
275 			break;
276 
277 		default:
278 			pr_err("%s: unknown ADD relocation: %u\n",
279 			       module->name,
280 			       ELF32_R_TYPE(rela[i].r_info));
281 			return -ENOEXEC;
282 		}
283 	}
284 
285 	return 0;
286 }
287 
288 #ifdef CONFIG_DYNAMIC_FTRACE
289 notrace int module_trampoline_target(struct module *mod, unsigned long addr,
290 				     unsigned long *target)
291 {
292 	ppc_inst_t jmp[4];
293 
294 	/* Find where the trampoline jumps to */
295 	if (copy_inst_from_kernel_nofault(jmp, (void *)addr))
296 		return -EFAULT;
297 	if (__copy_inst_from_kernel_nofault(jmp + 1, (void *)addr + 4))
298 		return -EFAULT;
299 	if (__copy_inst_from_kernel_nofault(jmp + 2, (void *)addr + 8))
300 		return -EFAULT;
301 	if (__copy_inst_from_kernel_nofault(jmp + 3, (void *)addr + 12))
302 		return -EFAULT;
303 
304 	/* verify that this is what we expect it to be */
305 	if ((ppc_inst_val(jmp[0]) & 0xffff0000) != PPC_RAW_LIS(_R12, 0))
306 		return -EINVAL;
307 	if ((ppc_inst_val(jmp[1]) & 0xffff0000) != PPC_RAW_ADDI(_R12, _R12, 0))
308 		return -EINVAL;
309 	if (ppc_inst_val(jmp[2]) != PPC_RAW_MTCTR(_R12))
310 		return -EINVAL;
311 	if (ppc_inst_val(jmp[3]) != PPC_RAW_BCTR())
312 		return -EINVAL;
313 
314 	addr = (ppc_inst_val(jmp[1]) & 0xffff) | ((ppc_inst_val(jmp[0]) & 0xffff) << 16);
315 	if (addr & 0x8000)
316 		addr -= 0x10000;
317 
318 	*target = addr;
319 
320 	return 0;
321 }
322 
323 int module_finalize_ftrace(struct module *module, const Elf_Shdr *sechdrs)
324 {
325 	module->arch.tramp = do_plt_call(module->core_layout.base,
326 					 (unsigned long)ftrace_caller,
327 					 sechdrs, module);
328 	if (!module->arch.tramp)
329 		return -ENOENT;
330 
331 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
332 	module->arch.tramp_regs = do_plt_call(module->core_layout.base,
333 					      (unsigned long)ftrace_regs_caller,
334 					      sechdrs, module);
335 	if (!module->arch.tramp_regs)
336 		return -ENOENT;
337 #endif
338 
339 	return 0;
340 }
341 #endif
342