Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Kernel unwinding support |
| 3 | * |
| 4 | * (c) 2002-2004 Randolph Chung <tausq@debian.org> |
| 5 | * |
| 6 | * Derived partially from the IA64 implementation. The PA-RISC |
| 7 | * Runtime Architecture Document is also a useful reference to |
| 8 | * understand what is happening here |
| 9 | */ |
| 10 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | #include <linux/kernel.h> |
| 12 | #include <linux/init.h> |
Matthew Wilcox | e6fc044 | 2006-12-15 09:34:36 -0700 | [diff] [blame] | 13 | #include <linux/sched.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/slab.h> |
| 15 | #include <linux/kallsyms.h> |
| 16 | |
| 17 | #include <asm/uaccess.h> |
| 18 | #include <asm/assembly.h> |
Randolph Chung | 05dc16d | 2007-06-12 14:27:32 +0800 | [diff] [blame] | 19 | #include <asm/asm-offsets.h> |
| 20 | #include <asm/ptrace.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | |
| 22 | #include <asm/unwind.h> |
| 23 | |
| 24 | /* #define DEBUG 1 */ |
| 25 | #ifdef DEBUG |
| 26 | #define dbg(x...) printk(x) |
| 27 | #else |
| 28 | #define dbg(x...) |
| 29 | #endif |
| 30 | |
Randolph Chung | e036306 | 2007-06-12 14:27:32 +0800 | [diff] [blame] | 31 | #define KERNEL_START (KERNEL_BINARY_TEXT_START - 0x1000) |
| 32 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | extern struct unwind_table_entry __start___unwind[]; |
| 34 | extern struct unwind_table_entry __stop___unwind[]; |
| 35 | |
| 36 | static spinlock_t unwind_lock; |
| 37 | /* |
| 38 | * the kernel unwind block is not dynamically allocated so that |
| 39 | * we can call unwind_init as early in the bootup process as |
| 40 | * possible (before the slab allocator is initialized) |
| 41 | */ |
Helge Deller | 8039de1 | 2006-01-10 20:35:03 -0500 | [diff] [blame] | 42 | static struct unwind_table kernel_unwind_table __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | static LIST_HEAD(unwind_tables); |
| 44 | |
| 45 | static inline const struct unwind_table_entry * |
| 46 | find_unwind_entry_in_table(const struct unwind_table *table, unsigned long addr) |
| 47 | { |
| 48 | const struct unwind_table_entry *e = NULL; |
| 49 | unsigned long lo, hi, mid; |
| 50 | |
| 51 | lo = 0; |
| 52 | hi = table->length - 1; |
| 53 | |
| 54 | while (lo <= hi) { |
| 55 | mid = (hi - lo) / 2 + lo; |
| 56 | e = &table->table[mid]; |
| 57 | if (addr < e->region_start) |
| 58 | hi = mid - 1; |
| 59 | else if (addr > e->region_end) |
| 60 | lo = mid + 1; |
| 61 | else |
| 62 | return e; |
| 63 | } |
| 64 | |
| 65 | return NULL; |
| 66 | } |
| 67 | |
| 68 | static const struct unwind_table_entry * |
| 69 | find_unwind_entry(unsigned long addr) |
| 70 | { |
| 71 | struct unwind_table *table; |
| 72 | const struct unwind_table_entry *e = NULL; |
| 73 | |
| 74 | if (addr >= kernel_unwind_table.start && |
| 75 | addr <= kernel_unwind_table.end) |
| 76 | e = find_unwind_entry_in_table(&kernel_unwind_table, addr); |
| 77 | else |
| 78 | list_for_each_entry(table, &unwind_tables, list) { |
| 79 | if (addr >= table->start && |
| 80 | addr <= table->end) |
| 81 | e = find_unwind_entry_in_table(table, addr); |
| 82 | if (e) |
| 83 | break; |
| 84 | } |
| 85 | |
| 86 | return e; |
| 87 | } |
| 88 | |
| 89 | static void |
| 90 | unwind_table_init(struct unwind_table *table, const char *name, |
| 91 | unsigned long base_addr, unsigned long gp, |
| 92 | void *table_start, void *table_end) |
| 93 | { |
| 94 | struct unwind_table_entry *start = table_start; |
| 95 | struct unwind_table_entry *end = |
| 96 | (struct unwind_table_entry *)table_end - 1; |
| 97 | |
| 98 | table->name = name; |
| 99 | table->base_addr = base_addr; |
| 100 | table->gp = gp; |
| 101 | table->start = base_addr + start->region_start; |
| 102 | table->end = base_addr + end->region_end; |
| 103 | table->table = (struct unwind_table_entry *)table_start; |
| 104 | table->length = end - start + 1; |
| 105 | INIT_LIST_HEAD(&table->list); |
| 106 | |
| 107 | for (; start <= end; start++) { |
| 108 | if (start < end && |
| 109 | start->region_end > (start+1)->region_start) { |
| 110 | printk("WARNING: Out of order unwind entry! %p and %p\n", start, start+1); |
| 111 | } |
| 112 | |
| 113 | start->region_start += base_addr; |
| 114 | start->region_end += base_addr; |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | static void |
| 119 | unwind_table_sort(struct unwind_table_entry *start, |
| 120 | struct unwind_table_entry *finish) |
| 121 | { |
| 122 | struct unwind_table_entry el, *p, *q; |
| 123 | |
| 124 | for (p = start + 1; p < finish; ++p) { |
| 125 | if (p[0].region_start < p[-1].region_start) { |
| 126 | el = *p; |
| 127 | q = p; |
| 128 | do { |
| 129 | q[0] = q[-1]; |
| 130 | --q; |
| 131 | } while (q > start && |
| 132 | el.region_start < q[-1].region_start); |
| 133 | *q = el; |
| 134 | } |
| 135 | } |
| 136 | } |
| 137 | |
| 138 | struct unwind_table * |
| 139 | unwind_table_add(const char *name, unsigned long base_addr, |
| 140 | unsigned long gp, |
| 141 | void *start, void *end) |
| 142 | { |
| 143 | struct unwind_table *table; |
| 144 | unsigned long flags; |
| 145 | struct unwind_table_entry *s = (struct unwind_table_entry *)start; |
| 146 | struct unwind_table_entry *e = (struct unwind_table_entry *)end; |
| 147 | |
| 148 | unwind_table_sort(s, e); |
| 149 | |
| 150 | table = kmalloc(sizeof(struct unwind_table), GFP_USER); |
| 151 | if (table == NULL) |
| 152 | return NULL; |
| 153 | unwind_table_init(table, name, base_addr, gp, start, end); |
| 154 | spin_lock_irqsave(&unwind_lock, flags); |
| 155 | list_add_tail(&table->list, &unwind_tables); |
| 156 | spin_unlock_irqrestore(&unwind_lock, flags); |
| 157 | |
| 158 | return table; |
| 159 | } |
| 160 | |
| 161 | void unwind_table_remove(struct unwind_table *table) |
| 162 | { |
| 163 | unsigned long flags; |
| 164 | |
| 165 | spin_lock_irqsave(&unwind_lock, flags); |
| 166 | list_del(&table->list); |
| 167 | spin_unlock_irqrestore(&unwind_lock, flags); |
| 168 | |
| 169 | kfree(table); |
| 170 | } |
| 171 | |
| 172 | /* Called from setup_arch to import the kernel unwind info */ |
James Bottomley | f0514ae | 2008-09-11 10:17:23 -0400 | [diff] [blame] | 173 | int unwind_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | { |
| 175 | long start, stop; |
| 176 | register unsigned long gp __asm__ ("r27"); |
| 177 | |
| 178 | start = (long)&__start___unwind[0]; |
| 179 | stop = (long)&__stop___unwind[0]; |
| 180 | |
| 181 | spin_lock_init(&unwind_lock); |
| 182 | |
| 183 | printk("unwind_init: start = 0x%lx, end = 0x%lx, entries = %lu\n", |
| 184 | start, stop, |
| 185 | (stop - start) / sizeof(struct unwind_table_entry)); |
| 186 | |
| 187 | unwind_table_init(&kernel_unwind_table, "kernel", KERNEL_START, |
| 188 | gp, |
| 189 | &__start___unwind[0], &__stop___unwind[0]); |
| 190 | #if 0 |
| 191 | { |
| 192 | int i; |
| 193 | for (i = 0; i < 10; i++) |
| 194 | { |
| 195 | printk("region 0x%x-0x%x\n", |
| 196 | __start___unwind[i].region_start, |
| 197 | __start___unwind[i].region_end); |
| 198 | } |
| 199 | } |
| 200 | #endif |
| 201 | return 0; |
| 202 | } |
| 203 | |
Randolph Chung | 05dc16d | 2007-06-12 14:27:32 +0800 | [diff] [blame] | 204 | #ifdef CONFIG_64BIT |
| 205 | #define get_func_addr(fptr) fptr[2] |
| 206 | #else |
| 207 | #define get_func_addr(fptr) fptr[0] |
| 208 | #endif |
| 209 | |
| 210 | static int unwind_special(struct unwind_frame_info *info, unsigned long pc, int frame_size) |
| 211 | { |
Kyle McMartin | 7819994 | 2007-10-18 00:03:56 -0700 | [diff] [blame] | 212 | extern void handle_interruption(int, struct pt_regs *); |
| 213 | static unsigned long *hi = (unsigned long *)&handle_interruption; |
Randolph Chung | 05dc16d | 2007-06-12 14:27:32 +0800 | [diff] [blame] | 214 | |
| 215 | if (pc == get_func_addr(hi)) { |
| 216 | struct pt_regs *regs = (struct pt_regs *)(info->sp - frame_size - PT_SZ_ALGN); |
| 217 | dbg("Unwinding through handle_interruption()\n"); |
| 218 | info->prev_sp = regs->gr[30]; |
| 219 | info->prev_ip = regs->iaoq[0]; |
| 220 | |
| 221 | return 1; |
| 222 | } |
| 223 | |
| 224 | return 0; |
| 225 | } |
| 226 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | static void unwind_frame_regs(struct unwind_frame_info *info) |
| 228 | { |
| 229 | const struct unwind_table_entry *e; |
| 230 | unsigned long npc; |
| 231 | unsigned int insn; |
| 232 | long frame_size = 0; |
| 233 | int looking_for_rp, rpoffset = 0; |
| 234 | |
| 235 | e = find_unwind_entry(info->ip); |
| 236 | if (e == NULL) { |
| 237 | unsigned long sp; |
| 238 | extern char _stext[], _etext[]; |
| 239 | |
| 240 | dbg("Cannot find unwind entry for 0x%lx; forced unwinding\n", info->ip); |
| 241 | |
| 242 | #ifdef CONFIG_KALLSYMS |
| 243 | /* Handle some frequent special cases.... */ |
| 244 | { |
Tejun Heo | 9281ace | 2007-07-17 04:03:51 -0700 | [diff] [blame] | 245 | char symname[KSYM_NAME_LEN]; |
Kyle McMartin | cb95779 | 2007-05-30 02:24:28 -0400 | [diff] [blame] | 246 | char *modname; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | |
Kyle McMartin | cb95779 | 2007-05-30 02:24:28 -0400 | [diff] [blame] | 248 | kallsyms_lookup(info->ip, NULL, NULL, &modname, |
| 249 | symname); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 250 | |
| 251 | dbg("info->ip = 0x%lx, name = %s\n", info->ip, symname); |
| 252 | |
| 253 | if (strcmp(symname, "_switch_to_ret") == 0) { |
| 254 | info->prev_sp = info->sp - CALLEE_SAVE_FRAME_SIZE; |
| 255 | info->prev_ip = *(unsigned long *)(info->prev_sp - RP_OFFSET); |
| 256 | dbg("_switch_to_ret @ %lx - setting " |
| 257 | "prev_sp=%lx prev_ip=%lx\n", |
| 258 | info->ip, info->prev_sp, |
| 259 | info->prev_ip); |
| 260 | return; |
| 261 | } else if (strcmp(symname, "ret_from_kernel_thread") == 0 || |
| 262 | strcmp(symname, "syscall_exit") == 0) { |
| 263 | info->prev_ip = info->prev_sp = 0; |
| 264 | return; |
| 265 | } |
| 266 | } |
| 267 | #endif |
| 268 | |
| 269 | /* Since we are doing the unwinding blind, we don't know if |
| 270 | we are adjusting the stack correctly or extracting the rp |
| 271 | correctly. The rp is checked to see if it belongs to the |
| 272 | kernel text section, if not we assume we don't have a |
| 273 | correct stack frame and we continue to unwind the stack. |
| 274 | This is not quite correct, and will fail for loadable |
| 275 | modules. */ |
| 276 | sp = info->sp & ~63; |
| 277 | do { |
| 278 | unsigned long tmp; |
| 279 | |
| 280 | info->prev_sp = sp - 64; |
| 281 | info->prev_ip = 0; |
| 282 | if (get_user(tmp, (unsigned long *)(info->prev_sp - RP_OFFSET))) |
| 283 | break; |
| 284 | info->prev_ip = tmp; |
| 285 | sp = info->prev_sp; |
| 286 | } while (info->prev_ip < (unsigned long)_stext || |
| 287 | info->prev_ip > (unsigned long)_etext); |
| 288 | |
| 289 | info->rp = 0; |
| 290 | |
| 291 | dbg("analyzing func @ %lx with no unwind info, setting " |
| 292 | "prev_sp=%lx prev_ip=%lx\n", info->ip, |
| 293 | info->prev_sp, info->prev_ip); |
| 294 | } else { |
| 295 | dbg("e->start = 0x%x, e->end = 0x%x, Save_SP = %d, " |
| 296 | "Save_RP = %d, Millicode = %d size = %u\n", |
| 297 | e->region_start, e->region_end, e->Save_SP, e->Save_RP, |
| 298 | e->Millicode, e->Total_frame_size); |
| 299 | |
| 300 | looking_for_rp = e->Save_RP; |
| 301 | |
| 302 | for (npc = e->region_start; |
| 303 | (frame_size < (e->Total_frame_size << 3) || |
| 304 | looking_for_rp) && |
| 305 | npc < info->ip; |
| 306 | npc += 4) { |
| 307 | |
| 308 | insn = *(unsigned int *)npc; |
| 309 | |
| 310 | if ((insn & 0xffffc000) == 0x37de0000 || |
| 311 | (insn & 0xffe00000) == 0x6fc00000) { |
| 312 | /* ldo X(sp), sp, or stwm X,D(sp) */ |
| 313 | frame_size += (insn & 0x1 ? -1 << 13 : 0) | |
| 314 | ((insn & 0x3fff) >> 1); |
| 315 | dbg("analyzing func @ %lx, insn=%08x @ " |
| 316 | "%lx, frame_size = %ld\n", info->ip, |
| 317 | insn, npc, frame_size); |
| 318 | } else if ((insn & 0xffe00008) == 0x73c00008) { |
| 319 | /* std,ma X,D(sp) */ |
| 320 | frame_size += (insn & 0x1 ? -1 << 13 : 0) | |
| 321 | (((insn >> 4) & 0x3ff) << 3); |
| 322 | dbg("analyzing func @ %lx, insn=%08x @ " |
| 323 | "%lx, frame_size = %ld\n", info->ip, |
| 324 | insn, npc, frame_size); |
| 325 | } else if (insn == 0x6bc23fd9) { |
| 326 | /* stw rp,-20(sp) */ |
| 327 | rpoffset = 20; |
| 328 | looking_for_rp = 0; |
| 329 | dbg("analyzing func @ %lx, insn=stw rp," |
| 330 | "-20(sp) @ %lx\n", info->ip, npc); |
| 331 | } else if (insn == 0x0fc212c1) { |
| 332 | /* std rp,-16(sr0,sp) */ |
| 333 | rpoffset = 16; |
| 334 | looking_for_rp = 0; |
| 335 | dbg("analyzing func @ %lx, insn=std rp," |
| 336 | "-16(sp) @ %lx\n", info->ip, npc); |
| 337 | } |
| 338 | } |
| 339 | |
Randolph Chung | 05dc16d | 2007-06-12 14:27:32 +0800 | [diff] [blame] | 340 | if (!unwind_special(info, e->region_start, frame_size)) { |
| 341 | info->prev_sp = info->sp - frame_size; |
| 342 | if (e->Millicode) |
| 343 | info->rp = info->r31; |
| 344 | else if (rpoffset) |
| 345 | info->rp = *(unsigned long *)(info->prev_sp - rpoffset); |
| 346 | info->prev_ip = info->rp; |
| 347 | info->rp = 0; |
| 348 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | |
| 350 | dbg("analyzing func @ %lx, setting prev_sp=%lx " |
| 351 | "prev_ip=%lx npc=%lx\n", info->ip, info->prev_sp, |
| 352 | info->prev_ip, npc); |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | void unwind_frame_init(struct unwind_frame_info *info, struct task_struct *t, |
| 357 | struct pt_regs *regs) |
| 358 | { |
| 359 | memset(info, 0, sizeof(struct unwind_frame_info)); |
| 360 | info->t = t; |
| 361 | info->sp = regs->gr[30]; |
| 362 | info->ip = regs->iaoq[0]; |
| 363 | info->rp = regs->gr[2]; |
| 364 | info->r31 = regs->gr[31]; |
| 365 | |
| 366 | dbg("(%d) Start unwind from sp=%08lx ip=%08lx\n", |
| 367 | t ? (int)t->pid : -1, info->sp, info->ip); |
| 368 | } |
| 369 | |
| 370 | void unwind_frame_init_from_blocked_task(struct unwind_frame_info *info, struct task_struct *t) |
| 371 | { |
| 372 | struct pt_regs *r = &t->thread.regs; |
| 373 | struct pt_regs *r2; |
| 374 | |
Helge Deller | e0e7ed4 | 2008-12-17 23:57:34 +0100 | [diff] [blame] | 375 | r2 = kmalloc(sizeof(struct pt_regs), GFP_ATOMIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | if (!r2) |
| 377 | return; |
| 378 | *r2 = *r; |
| 379 | r2->gr[30] = r->ksp; |
| 380 | r2->iaoq[0] = r->kpc; |
| 381 | unwind_frame_init(info, t, r2); |
| 382 | kfree(r2); |
| 383 | } |
| 384 | |
| 385 | void unwind_frame_init_running(struct unwind_frame_info *info, struct pt_regs *regs) |
| 386 | { |
| 387 | unwind_frame_init(info, current, regs); |
| 388 | } |
| 389 | |
| 390 | int unwind_once(struct unwind_frame_info *next_frame) |
| 391 | { |
| 392 | unwind_frame_regs(next_frame); |
| 393 | |
| 394 | if (next_frame->prev_sp == 0 || |
| 395 | next_frame->prev_ip == 0) |
| 396 | return -1; |
| 397 | |
| 398 | next_frame->sp = next_frame->prev_sp; |
| 399 | next_frame->ip = next_frame->prev_ip; |
| 400 | next_frame->prev_sp = 0; |
| 401 | next_frame->prev_ip = 0; |
| 402 | |
| 403 | dbg("(%d) Continue unwind to sp=%08lx ip=%08lx\n", |
| 404 | next_frame->t ? (int)next_frame->t->pid : -1, |
| 405 | next_frame->sp, next_frame->ip); |
| 406 | |
| 407 | return 0; |
| 408 | } |
| 409 | |
| 410 | int unwind_to_user(struct unwind_frame_info *info) |
| 411 | { |
| 412 | int ret; |
| 413 | |
| 414 | do { |
| 415 | ret = unwind_once(info); |
| 416 | } while (!ret && !(info->ip & 3)); |
| 417 | |
| 418 | return ret; |
| 419 | } |