Merge branch 'linus' into x86/apic
[linux-2.6] / 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 #include <linux/module.h>
19 #include <linux/moduleloader.h>
20 #include <linux/elf.h>
21 #include <linux/vmalloc.h>
22 #include <linux/fs.h>
23 #include <linux/string.h>
24 #include <linux/kernel.h>
25 #include <linux/ftrace.h>
26 #include <linux/cache.h>
27 #include <linux/bug.h>
28 #include <linux/sort.h>
29
30 #include "setup.h"
31
32 #if 0
33 #define DEBUGP printk
34 #else
35 #define DEBUGP(fmt , ...)
36 #endif
37
38 /* Count how many different relocations (different symbol, different
39    addend) */
40 static unsigned int count_relocs(const Elf32_Rela *rela, unsigned int num)
41 {
42         unsigned int i, r_info, r_addend, _count_relocs;
43
44         _count_relocs = 0;
45         r_info = 0;
46         r_addend = 0;
47         for (i = 0; i < num; i++)
48                 /* Only count 24-bit relocs, others don't need stubs */
49                 if (ELF32_R_TYPE(rela[i].r_info) == R_PPC_REL24 &&
50                     (r_info != ELF32_R_SYM(rela[i].r_info) ||
51                      r_addend != rela[i].r_addend)) {
52                         _count_relocs++;
53                         r_info = ELF32_R_SYM(rela[i].r_info);
54                         r_addend = rela[i].r_addend;
55                 }
56
57 #ifdef CONFIG_DYNAMIC_FTRACE
58         _count_relocs++;        /* add one for ftrace_caller */
59 #endif
60         return _count_relocs;
61 }
62
63 static int relacmp(const void *_x, const void *_y)
64 {
65         const Elf32_Rela *x, *y;
66
67         y = (Elf32_Rela *)_x;
68         x = (Elf32_Rela *)_y;
69
70         /* Compare the entire r_info (as opposed to ELF32_R_SYM(r_info) only) to
71          * make the comparison cheaper/faster. It won't affect the sorting or
72          * the counting algorithms' performance
73          */
74         if (x->r_info < y->r_info)
75                 return -1;
76         else if (x->r_info > y->r_info)
77                 return 1;
78         else if (x->r_addend < y->r_addend)
79                 return -1;
80         else if (x->r_addend > y->r_addend)
81                 return 1;
82         else
83                 return 0;
84 }
85
86 static void relaswap(void *_x, void *_y, int size)
87 {
88         uint32_t *x, *y, tmp;
89         int i;
90
91         y = (uint32_t *)_x;
92         x = (uint32_t *)_y;
93
94         for (i = 0; i < sizeof(Elf32_Rela) / sizeof(uint32_t); i++) {
95                 tmp = x[i];
96                 x[i] = y[i];
97                 y[i] = tmp;
98         }
99 }
100
101 /* Get the potential trampolines size required of the init and
102    non-init sections */
103 static unsigned long get_plt_size(const Elf32_Ehdr *hdr,
104                                   const Elf32_Shdr *sechdrs,
105                                   const char *secstrings,
106                                   int is_init)
107 {
108         unsigned long ret = 0;
109         unsigned i;
110
111         /* Everything marked ALLOC (this includes the exported
112            symbols) */
113         for (i = 1; i < hdr->e_shnum; i++) {
114                 /* If it's called *.init*, and we're not init, we're
115                    not interested */
116                 if ((strstr(secstrings + sechdrs[i].sh_name, ".init") != 0)
117                     != is_init)
118                         continue;
119
120                 /* We don't want to look at debug sections. */
121                 if (strstr(secstrings + sechdrs[i].sh_name, ".debug") != 0)
122                         continue;
123
124                 if (sechdrs[i].sh_type == SHT_RELA) {
125                         DEBUGP("Found relocations in section %u\n", i);
126                         DEBUGP("Ptr: %p.  Number: %u\n",
127                                (void *)hdr + sechdrs[i].sh_offset,
128                                sechdrs[i].sh_size / sizeof(Elf32_Rela));
129
130                         /* Sort the relocation information based on a symbol and
131                          * addend key. This is a stable O(n*log n) complexity
132                          * alogrithm but it will reduce the complexity of
133                          * count_relocs() to linear complexity O(n)
134                          */
135                         sort((void *)hdr + sechdrs[i].sh_offset,
136                              sechdrs[i].sh_size / sizeof(Elf32_Rela),
137                              sizeof(Elf32_Rela), relacmp, relaswap);
138
139                         ret += count_relocs((void *)hdr
140                                              + sechdrs[i].sh_offset,
141                                              sechdrs[i].sh_size
142                                              / sizeof(Elf32_Rela))
143                                 * sizeof(struct ppc_plt_entry);
144                 }
145         }
146
147         return ret;
148 }
149
150 int module_frob_arch_sections(Elf32_Ehdr *hdr,
151                               Elf32_Shdr *sechdrs,
152                               char *secstrings,
153                               struct module *me)
154 {
155         unsigned int i;
156
157         /* Find .plt and .init.plt sections */
158         for (i = 0; i < hdr->e_shnum; i++) {
159                 if (strcmp(secstrings + sechdrs[i].sh_name, ".init.plt") == 0)
160                         me->arch.init_plt_section = i;
161                 else if (strcmp(secstrings + sechdrs[i].sh_name, ".plt") == 0)
162                         me->arch.core_plt_section = i;
163         }
164         if (!me->arch.core_plt_section || !me->arch.init_plt_section) {
165                 printk("Module doesn't contain .plt or .init.plt sections.\n");
166                 return -ENOEXEC;
167         }
168
169         /* Override their sizes */
170         sechdrs[me->arch.core_plt_section].sh_size
171                 = get_plt_size(hdr, sechdrs, secstrings, 0);
172         sechdrs[me->arch.init_plt_section].sh_size
173                 = get_plt_size(hdr, sechdrs, secstrings, 1);
174         return 0;
175 }
176
177 int apply_relocate(Elf32_Shdr *sechdrs,
178                    const char *strtab,
179                    unsigned int symindex,
180                    unsigned int relsec,
181                    struct module *module)
182 {
183         printk(KERN_ERR "%s: Non-ADD RELOCATION unsupported\n",
184                module->name);
185         return -ENOEXEC;
186 }
187
188 static inline int entry_matches(struct ppc_plt_entry *entry, Elf32_Addr val)
189 {
190         if (entry->jump[0] == 0x3d600000 + ((val + 0x8000) >> 16)
191             && entry->jump[1] == 0x396b0000 + (val & 0xffff))
192                 return 1;
193         return 0;
194 }
195
196 /* Set up a trampoline in the PLT to bounce us to the distant function */
197 static uint32_t do_plt_call(void *location,
198                             Elf32_Addr val,
199                             Elf32_Shdr *sechdrs,
200                             struct module *mod)
201 {
202         struct ppc_plt_entry *entry;
203
204         DEBUGP("Doing plt for call to 0x%x at 0x%x\n", val, (unsigned int)location);
205         /* Init, or core PLT? */
206         if (location >= mod->module_core
207             && location < mod->module_core + mod->core_size)
208                 entry = (void *)sechdrs[mod->arch.core_plt_section].sh_addr;
209         else
210                 entry = (void *)sechdrs[mod->arch.init_plt_section].sh_addr;
211
212         /* Find this entry, or if that fails, the next avail. entry */
213         while (entry->jump[0]) {
214                 if (entry_matches(entry, val)) return (uint32_t)entry;
215                 entry++;
216         }
217
218         /* Stolen from Paul Mackerras as well... */
219         entry->jump[0] = 0x3d600000+((val+0x8000)>>16); /* lis r11,sym@ha */
220         entry->jump[1] = 0x396b0000 + (val&0xffff);     /* addi r11,r11,sym@l*/
221         entry->jump[2] = 0x7d6903a6;                    /* mtctr r11 */
222         entry->jump[3] = 0x4e800420;                    /* bctr */
223
224         DEBUGP("Initialized plt for 0x%x at %p\n", val, entry);
225         return (uint32_t)entry;
226 }
227
228 int apply_relocate_add(Elf32_Shdr *sechdrs,
229                        const char *strtab,
230                        unsigned int symindex,
231                        unsigned int relsec,
232                        struct module *module)
233 {
234         unsigned int i;
235         Elf32_Rela *rela = (void *)sechdrs[relsec].sh_addr;
236         Elf32_Sym *sym;
237         uint32_t *location;
238         uint32_t value;
239
240         DEBUGP("Applying ADD relocate section %u to %u\n", relsec,
241                sechdrs[relsec].sh_info);
242         for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rela); i++) {
243                 /* This is where to make the change */
244                 location = (void *)sechdrs[sechdrs[relsec].sh_info].sh_addr
245                         + rela[i].r_offset;
246                 /* This is the symbol it is referring to.  Note that all
247                    undefined symbols have been resolved.  */
248                 sym = (Elf32_Sym *)sechdrs[symindex].sh_addr
249                         + ELF32_R_SYM(rela[i].r_info);
250                 /* `Everything is relative'. */
251                 value = sym->st_value + rela[i].r_addend;
252
253                 switch (ELF32_R_TYPE(rela[i].r_info)) {
254                 case R_PPC_ADDR32:
255                         /* Simply set it */
256                         *(uint32_t *)location = value;
257                         break;
258
259                 case R_PPC_ADDR16_LO:
260                         /* Low half of the symbol */
261                         *(uint16_t *)location = value;
262                         break;
263
264                 case R_PPC_ADDR16_HI:
265                         /* Higher half of the symbol */
266                         *(uint16_t *)location = (value >> 16);
267                         break;
268
269                 case R_PPC_ADDR16_HA:
270                         /* Sign-adjusted lower 16 bits: PPC ELF ABI says:
271                            (((x >> 16) + ((x & 0x8000) ? 1 : 0))) & 0xFFFF.
272                            This is the same, only sane.
273                          */
274                         *(uint16_t *)location = (value + 0x8000) >> 16;
275                         break;
276
277                 case R_PPC_REL24:
278                         if ((int)(value - (uint32_t)location) < -0x02000000
279                             || (int)(value - (uint32_t)location) >= 0x02000000)
280                                 value = do_plt_call(location, value,
281                                                     sechdrs, module);
282
283                         /* Only replace bits 2 through 26 */
284                         DEBUGP("REL24 value = %08X. location = %08X\n",
285                                value, (uint32_t)location);
286                         DEBUGP("Location before: %08X.\n",
287                                *(uint32_t *)location);
288                         *(uint32_t *)location
289                                 = (*(uint32_t *)location & ~0x03fffffc)
290                                 | ((value - (uint32_t)location)
291                                    & 0x03fffffc);
292                         DEBUGP("Location after: %08X.\n",
293                                *(uint32_t *)location);
294                         DEBUGP("ie. jump to %08X+%08X = %08X\n",
295                                *(uint32_t *)location & 0x03fffffc,
296                                (uint32_t)location,
297                                (*(uint32_t *)location & 0x03fffffc)
298                                + (uint32_t)location);
299                         break;
300
301                 case R_PPC_REL32:
302                         /* 32-bit relative jump. */
303                         *(uint32_t *)location = value - (uint32_t)location;
304                         break;
305
306                 default:
307                         printk("%s: unknown ADD relocation: %u\n",
308                                module->name,
309                                ELF32_R_TYPE(rela[i].r_info));
310                         return -ENOEXEC;
311                 }
312         }
313 #ifdef CONFIG_DYNAMIC_FTRACE
314         module->arch.tramp =
315                 do_plt_call(module->module_core,
316                             (unsigned long)ftrace_caller,
317                             sechdrs, module);
318 #endif
319         return 0;
320 }