9cf201111d6c31b17d0596c26e8b543127b6bb01
[linux-2.6-block.git] / arch / powerpc / kernel / module_32.c
1 /*  Kernel module help for PPC.
2     Copyright (C) 2001 Rusty Russell.
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
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/module.h>
22 #include <linux/moduleloader.h>
23 #include <linux/elf.h>
24 #include <linux/vmalloc.h>
25 #include <linux/fs.h>
26 #include <linux/string.h>
27 #include <linux/kernel.h>
28 #include <linux/ftrace.h>
29 #include <linux/cache.h>
30 #include <linux/bug.h>
31 #include <linux/sort.h>
32 #include <asm/setup.h>
33
34 /* Count how many different relocations (different symbol, different
35    addend) */
36 static unsigned int count_relocs(const Elf32_Rela *rela, unsigned int num)
37 {
38         unsigned int i, r_info, r_addend, _count_relocs;
39
40         _count_relocs = 0;
41         r_info = 0;
42         r_addend = 0;
43         for (i = 0; i < num; i++)
44                 /* Only count 24-bit relocs, others don't need stubs */
45                 if (ELF32_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
46                     (r_info != ELF32_R_SYM(rela[i].r_info) ||
47                      r_addend != rela[i].r_addend)) {
48                         _count_relocs++;
49                         r_info = ELF32_R_SYM(rela[i].r_info);
50                         r_addend = rela[i].r_addend;
51                 }
52
53 #ifdef CONFIG_DYNAMIC_FTRACE
54         _count_relocs++;        /* add one for ftrace_caller */
55 #endif
56         return _count_relocs;
57 }
58
59 static int relacmp(const void *_x, const void *_y)
60 {
61         const Elf32_Rela *x, *y;
62
63         y = (Elf32_Rela *)_x;
64         x = (Elf32_Rela *)_y;
65
66         /* Compare the entire r_info (as opposed to ELF32_R_SYM(r_info) only) to
67          * make the comparison cheaper/faster. It won't affect the sorting or
68          * the counting algorithms' performance
69          */
70         if (x->r_info < y->r_info)
71                 return -1;
72         else if (x->r_info > y->r_info)
73                 return 1;
74         else if (x->r_addend < y->r_addend)
75                 return -1;
76         else if (x->r_addend > y->r_addend)
77                 return 1;
78         else
79                 return 0;
80 }
81
82 static void relaswap(void *_x, void *_y, int size)
83 {
84         uint32_t *x, *y, tmp;
85         int i;
86
87         y = (uint32_t *)_x;
88         x = (uint32_t *)_y;
89
90         for (i = 0; i < sizeof(Elf32_Rela) / sizeof(uint32_t); i++) {
91                 tmp = x[i];
92                 x[i] = y[i];
93                 y[i] = tmp;
94         }
95 }
96
97 /* Get the potential trampolines size required of the init and
98    non-init sections */
99 static unsigned long get_plt_size(const Elf32_Ehdr *hdr,
100                                   const Elf32_Shdr *sechdrs,
101                                   const char *secstrings,
102                                   int is_init)
103 {
104         unsigned long ret = 0;
105         unsigned i;
106
107         /* Everything marked ALLOC (this includes the exported
108            symbols) */
109         for (i = 1; i < hdr->e_shnum; i++) {
110                 /* If it's called *.init*, and we're not init, we're
111                    not interested */
112                 if ((strstr(secstrings + sechdrs[i].sh_name, ".init") != NULL)
113                     != is_init)
114                         continue;
115
116                 /* We don't want to look at debug sections. */
117                 if (strstr(secstrings + sechdrs[i].sh_name, ".debug"))
118                         continue;
119
120                 if (sechdrs[i].sh_type == SHT_RELA) {
121                         pr_debug("Found relocations in section %u\n", i);
122                         pr_debug("Ptr: %p.  Number: %u\n",
123                                (void *)hdr + sechdrs[i].sh_offset,
124                                sechdrs[i].sh_size / sizeof(Elf32_Rela));
125
126                         /* Sort the relocation information based on a symbol and
127                          * addend key. This is a stable O(n*log n) complexity
128                          * alogrithm but it will reduce the complexity of
129                          * count_relocs() to linear complexity O(n)
130                          */
131                         sort((void *)hdr + sechdrs[i].sh_offset,
132                              sechdrs[i].sh_size / sizeof(Elf32_Rela),
133                              sizeof(Elf32_Rela), relacmp, relaswap);
134
135                         ret += count_relocs((void *)hdr
136                                              + sechdrs[i].sh_offset,
137                                              sechdrs[i].sh_size
138                                              / sizeof(Elf32_Rela))
139                                 * sizeof(struct ppc_plt_entry);
140                 }
141         }
142
143         return ret;
144 }
145
146 int module_frob_arch_sections(Elf32_Ehdr *hdr,
147                               Elf32_Shdr *sechdrs,
148                               char *secstrings,
149                               struct module *me)
150 {
151         unsigned int i;
152
153         /* Find .plt and .init.plt sections */
154         for (i = 0; i < hdr->e_shnum; i++) {
155                 if (strcmp(secstrings + sechdrs[i].sh_name, ".init.plt") == 0)
156                         me->arch.init_plt_section = i;
157                 else if (strcmp(secstrings + sechdrs[i].sh_name, ".plt") == 0)
158                         me->arch.core_plt_section = i;
159         }
160         if (!me->arch.core_plt_section || !me->arch.init_plt_section) {
161                 pr_err("Module doesn't contain .plt or .init.plt sections.\n");
162                 return -ENOEXEC;
163         }
164
165         /* Override their sizes */
166         sechdrs[me->arch.core_plt_section].sh_size
167                 = get_plt_size(hdr, sechdrs, secstrings, 0);
168         sechdrs[me->arch.init_plt_section].sh_size
169                 = get_plt_size(hdr, sechdrs, secstrings, 1);
170         return 0;
171 }
172
173 static inline int entry_matches(struct ppc_plt_entry *entry, Elf32_Addr val)
174 {
175         if (entry->jump[0] != (PPC_INST_ADDIS | __PPC_RT(R12) | PPC_HA(val)))
176                 return 0;
177         if (entry->jump[1] != (PPC_INST_ADDI | __PPC_RT(R12) | __PPC_RA(R12) |
178                                PPC_LO(val)))
179                 return 0;
180         return 1;
181 }
182
183 /* Set up a trampoline in the PLT to bounce us to the distant function */
184 static uint32_t do_plt_call(void *location,
185                             Elf32_Addr val,
186                             const Elf32_Shdr *sechdrs,
187                             struct module *mod)
188 {
189         struct ppc_plt_entry *entry;
190
191         pr_debug("Doing plt for call to 0x%x at 0x%x\n", val, (unsigned int)location);
192         /* Init, or core PLT? */
193         if (location >= mod->core_layout.base
194             && location < mod->core_layout.base + mod->core_layout.size)
195                 entry = (void *)sechdrs[mod->arch.core_plt_section].sh_addr;
196         else
197                 entry = (void *)sechdrs[mod->arch.init_plt_section].sh_addr;
198
199         /* Find this entry, or if that fails, the next avail. entry */
200         while (entry->jump[0]) {
201                 if (entry_matches(entry, val)) return (uint32_t)entry;
202                 entry++;
203         }
204
205         /*
206          * lis r12, sym@ha
207          * addi r12, r12, sym@l
208          * mtctr r12
209          * bctr
210          */
211         entry->jump[0] = PPC_INST_ADDIS | __PPC_RT(R12) | PPC_HA(val);
212         entry->jump[1] = PPC_INST_ADDI | __PPC_RT(R12) | __PPC_RA(R12) | PPC_LO(val);
213         entry->jump[2] = PPC_INST_MTCTR | __PPC_RS(R12);
214         entry->jump[3] = PPC_INST_BCTR;
215
216         pr_debug("Initialized plt for 0x%x at %p\n", val, entry);
217         return (uint32_t)entry;
218 }
219
220 int apply_relocate_add(Elf32_Shdr *sechdrs,
221                        const char *strtab,
222                        unsigned int symindex,
223                        unsigned int relsec,
224                        struct module *module)
225 {
226         unsigned int i;
227         Elf32_Rela *rela = (void *)sechdrs[relsec].sh_addr;
228         Elf32_Sym *sym;
229         uint32_t *location;
230         uint32_t value;
231
232         pr_debug("Applying ADD relocate section %u to %u\n", relsec,
233                sechdrs[relsec].sh_info);
234         for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
235                 /* This is where to make the change */
236                 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
237                         + rela[i].r_offset;
238                 /* This is the symbol it is referring to.  Note that all
239                    undefined symbols have been resolved.  */
240                 sym = (Elf32_Sym *)sechdrs[symindex].sh_addr
241                         + ELF32_R_SYM(rela[i].r_info);
242                 /* `Everything is relative'. */
243                 value = sym->st_value + rela[i].r_addend;
244
245                 switch (ELF32_R_TYPE(rela[i].r_info)) {
246                 case R_PPC_ADDR32:
247                         /* Simply set it */
248                         *(uint32_t *)location = value;
249                         break;
250
251                 case R_PPC_ADDR16_LO:
252                         /* Low half of the symbol */
253                         *(uint16_t *)location = value;
254                         break;
255
256                 case R_PPC_ADDR16_HI:
257                         /* Higher half of the symbol */
258                         *(uint16_t *)location = (value >> 16);
259                         break;
260
261                 case R_PPC_ADDR16_HA:
262                         /* Sign-adjusted lower 16 bits: PPC ELF ABI says:
263                            (((x >> 16) + ((x & 0x8000) ? 1 : 0))) & 0xFFFF.
264                            This is the same, only sane.
265                          */
266                         *(uint16_t *)location = (value + 0x8000) >> 16;
267                         break;
268
269                 case R_PPC_REL24:
270                         if ((int)(value - (uint32_t)location) < -0x02000000
271                             || (int)(value - (uint32_t)location) >= 0x02000000)
272                                 value = do_plt_call(location, value,
273                                                     sechdrs, module);
274
275                         /* Only replace bits 2 through 26 */
276                         pr_debug("REL24 value = %08X. location = %08X\n",
277                                value, (uint32_t)location);
278                         pr_debug("Location before: %08X.\n",
279                                *(uint32_t *)location);
280                         *(uint32_t *)location
281                                 = (*(uint32_t *)location & ~0x03fffffc)
282                                 | ((value - (uint32_t)location)
283                                    & 0x03fffffc);
284                         pr_debug("Location after: %08X.\n",
285                                *(uint32_t *)location);
286                         pr_debug("ie. jump to %08X+%08X = %08X\n",
287                                *(uint32_t *)location & 0x03fffffc,
288                                (uint32_t)location,
289                                (*(uint32_t *)location & 0x03fffffc)
290                                + (uint32_t)location);
291                         break;
292
293                 case R_PPC_REL32:
294                         /* 32-bit relative jump. */
295                         *(uint32_t *)location = value - (uint32_t)location;
296                         break;
297
298                 default:
299                         pr_err("%s: unknown ADD relocation: %u\n",
300                                module->name,
301                                ELF32_R_TYPE(rela[i].r_info));
302                         return -ENOEXEC;
303                 }
304         }
305
306         return 0;
307 }
308
309 #ifdef CONFIG_DYNAMIC_FTRACE
310 int module_finalize_ftrace(struct module *module, const Elf_Shdr *sechdrs)
311 {
312         module->arch.tramp = do_plt_call(module->core_layout.base,
313                                          (unsigned long)ftrace_caller,
314                                          sechdrs, module);
315         if (!module->arch.tramp)
316                 return -ENOENT;
317
318         return 0;
319 }
320 #endif