arch/powerpc/kernel/ftrace.c

   1 /*
   2  * Code for replacing ftrace calls with jumps.
   3  *
   4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   5  *
   6  * Thanks goes out to P.A. Semi, Inc for supplying me with a PPC64 box.
   7  *
   8  * Added function graph tracer code, taken from x86 that was written
   9  * by Frederic Weisbecker, and ported to PPC by Steven Rostedt.
  10  *
  11  */
  12
  13 #define pr_fmt(fmt) "ftrace-powerpc: " fmt
  14
  15 #include <linux/spinlock.h>
  16 #include <linux/hardirq.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/module.h>
  19 #include <linux/ftrace.h>
  20 #include <linux/percpu.h>
  21 #include <linux/init.h>
  22 #include <linux/list.h>
  23
  24 #include <asm/cacheflush.h>
  25 #include <asm/code-patching.h>
  26 #include <asm/ftrace.h>
  27 #include <asm/syscall.h>
  28
  29
  30 #ifdef CONFIG_DYNAMIC_FTRACE
  31 static unsigned int
  32 ftrace_call_replace(unsigned long ip, unsigned long addr, int link)
  33 {
  34         unsigned int op;
  35
  36         addr = ppc_function_entry((void *)addr);
  37
  38         /* if (link) set op to 'bl' else 'b' */
  39         op = create_branch((unsigned int *)ip, addr, link ? 1 : 0);
  40
  41         return op;
  42 }
  43
  44 static int
  45 ftrace_modify_code(unsigned long ip, unsigned int old, unsigned int new)
  46 {
  47         unsigned int replaced;
  48
  49         /*
  50          * Note: Due to modules and __init, code can
  51          *  disappear and change, we need to protect against faulting
  52          *  as well as code changing. We do this by using the
  53          *  probe_kernel_* functions.
  54          *
  55          * No real locking needed, this code is run through
  56          * kstop_machine, or before SMP starts.
  57          */
  58
  59         /* read the text we want to modify */
  60         if (probe_kernel_read(&replaced, (void *)ip, MCOUNT_INSN_SIZE))
  61                 return -EFAULT;
  62
  63         /* Make sure it is what we expect it to be */
  64         if (replaced != old)
  65                 return -EINVAL;
  66
  67         /* replace the text with the new text */
  68         if (patch_instruction((unsigned int *)ip, new))
  69                 return -EPERM;
  70
  71         return 0;
  72 }
  73
  74 /*
  75  * Helper functions that are the same for both PPC64 and PPC32.
  76  */
  77 static int test_24bit_addr(unsigned long ip, unsigned long addr)
  78 {
  79         addr = ppc_function_entry((void *)addr);
  80
  81         /* use the create_branch to verify that this offset can be branched */
  82         return create_branch((unsigned int *)ip, addr, 0);
  83 }
  84
  85 #ifdef CONFIG_MODULES
  86
  87 static int is_bl_op(unsigned int op)
  88 {
  89         return (op & 0xfc000003) == 0x48000001;
  90 }
  91
  92 static unsigned long find_bl_target(unsigned long ip, unsigned int op)
  93 {
  94         static int offset;
  95
  96         offset = (op & 0x03fffffc);
  97         /* make it signed */
  98         if (offset & 0x02000000)
  99                 offset |= 0xfe000000;
 100
 101         return ip + (long)offset;
 102 }
 103
 104 #ifdef CONFIG_PPC64
 105 static int
 106 __ftrace_make_nop(struct module *mod,
 107                   struct dyn_ftrace *rec, unsigned long addr)
 108 {
 109         unsigned int op;
 110         unsigned long entry, ptr;
 111         unsigned long ip = rec->ip;
 112         void *tramp;
 113
 114         /* read where this goes */
 115         if (probe_kernel_read(&op, (void *)ip, sizeof(int)))
 116                 return -EFAULT;
 117
 118         /* Make sure that that this is still a 24bit jump */
 119         if (!is_bl_op(op)) {
 120                 pr_err("Not expected bl: opcode is %x\n", op);
 121                 return -EINVAL;
 122         }
 123
 124         /* lets find where the pointer goes */
 125         tramp = (void *)find_bl_target(ip, op);
 126
 127         pr_devel("ip:%lx jumps to %p", ip, tramp);
 128
 129         if (!is_module_trampoline(tramp)) {
 130                 pr_err("Not a trampoline\n");
 131                 return -EINVAL;
 132         }
 133
 134         if (module_trampoline_target(mod, tramp, &ptr)) {
 135                 pr_err("Failed to get trampoline target\n");
 136                 return -EFAULT;
 137         }
 138
 139         pr_devel("trampoline target %lx", ptr);
 140
 141         entry = ppc_global_function_entry((void *)addr);
 142         /* This should match what was called */
 143         if (ptr != entry) {
 144                 pr_err("addr %lx does not match expected %lx\n", ptr, entry);
 145                 return -EINVAL;
 146         }
 147
 148         /*
 149          * Our original call site looks like:
 150          *
 151          * bl <tramp>
 152          * ld r2,XX(r1)
 153          *
 154          * Milton Miller pointed out that we can not simply nop the branch.
 155          * If a task was preempted when calling a trace function, the nops
 156          * will remove the way to restore the TOC in r2 and the r2 TOC will
 157          * get corrupted.
 158          *
 159          * Use a b +8 to jump over the load.
 160          */
 161         op = 0x48000008;        /* b +8 */
 162
 163         if (patch_instruction((unsigned int *)ip, op))
 164                 return -EPERM;
 165
 166         return 0;
 167 }
 168
 169 #else /* !PPC64 */
 170 static int
 171 __ftrace_make_nop(struct module *mod,
 172                   struct dyn_ftrace *rec, unsigned long addr)
 173 {
 174         unsigned int op;
 175         unsigned int jmp[4];
 176         unsigned long ip = rec->ip;
 177         unsigned long tramp;
 178
 179         if (probe_kernel_read(&op, (void *)ip, MCOUNT_INSN_SIZE))
 180                 return -EFAULT;
 181
 182         /* Make sure that that this is still a 24bit jump */
 183         if (!is_bl_op(op)) {
 184                 pr_err("Not expected bl: opcode is %x\n", op);
 185                 return -EINVAL;
 186         }
 187
 188         /* lets find where the pointer goes */
 189         tramp = find_bl_target(ip, op);
 190
 191         /*
 192          * On PPC32 the trampoline looks like:
 193          *  0x3d, 0x80, 0x00, 0x00  lis r12,sym@ha
 194          *  0x39, 0x8c, 0x00, 0x00  addi r12,r12,sym@l
 195          *  0x7d, 0x89, 0x03, 0xa6  mtctr r12
 196          *  0x4e, 0x80, 0x04, 0x20  bctr
 197          */
 198
 199         pr_devel("ip:%lx jumps to %lx", ip, tramp);
 200
 201         /* Find where the trampoline jumps to */
 202         if (probe_kernel_read(jmp, (void *)tramp, sizeof(jmp))) {
 203                 pr_err("Failed to read %lx\n", tramp);
 204                 return -EFAULT;
 205         }
 206
 207         pr_devel(" %08x %08x ", jmp[0], jmp[1]);
 208
 209         /* verify that this is what we expect it to be */
 210         if (((jmp[0] & 0xffff0000) != 0x3d800000) ||
 211             ((jmp[1] & 0xffff0000) != 0x398c0000) ||
 212             (jmp[2] != 0x7d8903a6) ||
 213             (jmp[3] != 0x4e800420)) {
 214                 pr_err("Not a trampoline\n");
 215                 return -EINVAL;
 216         }
 217
 218         tramp = (jmp[1] & 0xffff) |
 219                 ((jmp[0] & 0xffff) << 16);
 220         if (tramp & 0x8000)
 221                 tramp -= 0x10000;
 222
 223         pr_devel(" %lx ", tramp);
 224
 225         if (tramp != addr) {
 226                 pr_err("Trampoline location %08lx does not match addr\n",
 227                        tramp);
 228                 return -EINVAL;
 229         }
 230
 231         op = PPC_INST_NOP;
 232
 233         if (patch_instruction((unsigned int *)ip, op))
 234                 return -EPERM;
 235
 236         return 0;
 237 }
 238 #endif /* PPC64 */
 239 #endif /* CONFIG_MODULES */
 240
 241 int ftrace_make_nop(struct module *mod,
 242                     struct dyn_ftrace *rec, unsigned long addr)
 243 {
 244         unsigned long ip = rec->ip;
 245         unsigned int old, new;
 246
 247         /*
 248          * If the calling address is more that 24 bits away,
 249          * then we had to use a trampoline to make the call.
 250          * Otherwise just update the call site.
 251          */
 252         if (test_24bit_addr(ip, addr)) {
 253                 /* within range */
 254                 old = ftrace_call_replace(ip, addr, 1);
 255                 new = PPC_INST_NOP;
 256                 return ftrace_modify_code(ip, old, new);
 257         }
 258
 259 #ifdef CONFIG_MODULES
 260         /*
 261          * Out of range jumps are called from modules.
 262          * We should either already have a pointer to the module
 263          * or it has been passed in.
 264          */
 265         if (!rec->arch.mod) {
 266                 if (!mod) {
 267                         pr_err("No module loaded addr=%lx\n", addr);
 268                         return -EFAULT;
 269                 }
 270                 rec->arch.mod = mod;
 271         } else if (mod) {
 272                 if (mod != rec->arch.mod) {
 273                         pr_err("Record mod %p not equal to passed in mod %p\n",
 274                                rec->arch.mod, mod);
 275                         return -EINVAL;
 276                 }
 277                 /* nothing to do if mod == rec->arch.mod */
 278         } else
 279                 mod = rec->arch.mod;
 280
 281         return __ftrace_make_nop(mod, rec, addr);
 282 #else
 283         /* We should not get here without modules */
 284         return -EINVAL;
 285 #endif /* CONFIG_MODULES */
 286 }
 287
 288 #ifdef CONFIG_MODULES
 289 #ifdef CONFIG_PPC64
 290 static int
 291 __ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
 292 {
 293         unsigned int op[2];
 294         void *ip = (void *)rec->ip;
 295
 296         /* read where this goes */
 297         if (probe_kernel_read(op, ip, sizeof(op)))
 298                 return -EFAULT;
 299
 300         /*
 301          * We expect to see:
 302          *
 303          * b +8
 304          * ld r2,XX(r1)
 305          *
 306          * The load offset is different depending on the ABI. For simplicity
 307          * just mask it out when doing the compare.
 308          */
 309         if ((op[0] != 0x48000008) || ((op[1] & 0xffff0000) != 0xe8410000)) {
 310                 pr_err("Unexpected call sequence: %x %x\n", op[0], op[1]);
 311                 return -EINVAL;
 312         }
 313
 314         /* If we never set up a trampoline to ftrace_caller, then bail */
 315         if (!rec->arch.mod->arch.tramp) {
 316                 pr_err("No ftrace trampoline\n");
 317                 return -EINVAL;
 318         }
 319
 320         /* Ensure branch is within 24 bits */
 321         if (!create_branch(ip, rec->arch.mod->arch.tramp, BRANCH_SET_LINK)) {
 322                 pr_err("Branch out of range\n");
 323                 return -EINVAL;
 324         }
 325
 326         if (patch_branch(ip, rec->arch.mod->arch.tramp, BRANCH_SET_LINK)) {
 327                 pr_err("REL24 out of range!\n");
 328                 return -EINVAL;
 329         }
 330
 331         return 0;
 332 }
 333 #else
 334 static int
 335 __ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
 336 {
 337         unsigned int op;
 338         unsigned long ip = rec->ip;
 339
 340         /* read where this goes */
 341         if (probe_kernel_read(&op, (void *)ip, MCOUNT_INSN_SIZE))
 342                 return -EFAULT;
 343
 344         /* It should be pointing to a nop */
 345         if (op != PPC_INST_NOP) {
 346                 pr_err("Expected NOP but have %x\n", op);
 347                 return -EINVAL;
 348         }
 349
 350         /* If we never set up a trampoline to ftrace_caller, then bail */
 351         if (!rec->arch.mod->arch.tramp) {
 352                 pr_err("No ftrace trampoline\n");
 353                 return -EINVAL;
 354         }
 355
 356         /* create the branch to the trampoline */
 357         op = create_branch((unsigned int *)ip,
 358                            rec->arch.mod->arch.tramp, BRANCH_SET_LINK);
 359         if (!op) {
 360                 pr_err("REL24 out of range!\n");
 361                 return -EINVAL;
 362         }
 363
 364         pr_devel("write to %lx\n", rec->ip);
 365
 366         if (patch_instruction((unsigned int *)ip, op))
 367                 return -EPERM;
 368
 369         return 0;
 370 }
 371 #endif /* CONFIG_PPC64 */
 372 #endif /* CONFIG_MODULES */
 373
 374 int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
 375 {
 376         unsigned long ip = rec->ip;
 377         unsigned int old, new;
 378
 379         /*
 380          * If the calling address is more that 24 bits away,
 381          * then we had to use a trampoline to make the call.
 382          * Otherwise just update the call site.
 383          */
 384         if (test_24bit_addr(ip, addr)) {
 385                 /* within range */
 386                 old = PPC_INST_NOP;
 387                 new = ftrace_call_replace(ip, addr, 1);
 388                 return ftrace_modify_code(ip, old, new);
 389         }
 390
 391 #ifdef CONFIG_MODULES
 392         /*
 393          * Out of range jumps are called from modules.
 394          * Being that we are converting from nop, it had better
 395          * already have a module defined.
 396          */
 397         if (!rec->arch.mod) {
 398                 pr_err("No module loaded\n");
 399                 return -EINVAL;
 400         }
 401
 402         return __ftrace_make_call(rec, addr);
 403 #else
 404         /* We should not get here without modules */
 405         return -EINVAL;
 406 #endif /* CONFIG_MODULES */
 407 }
 408
 409 int ftrace_update_ftrace_func(ftrace_func_t func)
 410 {
 411         unsigned long ip = (unsigned long)(&ftrace_call);
 412         unsigned int old, new;
 413         int ret;
 414
 415         old = *(unsigned int *)&ftrace_call;
 416         new = ftrace_call_replace(ip, (unsigned long)func, 1);
 417         ret = ftrace_modify_code(ip, old, new);
 418
 419         return ret;
 420 }
 421
 422 static int __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
 423 {
 424         unsigned long ftrace_addr = (unsigned long)FTRACE_ADDR;
 425         int ret;
 426
 427         ret = ftrace_update_record(rec, enable);
 428
 429         switch (ret) {
 430         case FTRACE_UPDATE_IGNORE:
 431                 return 0;
 432         case FTRACE_UPDATE_MAKE_CALL:
 433                 return ftrace_make_call(rec, ftrace_addr);
 434         case FTRACE_UPDATE_MAKE_NOP:
 435                 return ftrace_make_nop(NULL, rec, ftrace_addr);
 436         }
 437
 438         return 0;
 439 }
 440
 441 void ftrace_replace_code(int enable)
 442 {
 443         struct ftrace_rec_iter *iter;
 444         struct dyn_ftrace *rec;
 445         int ret;
 446
 447         for (iter = ftrace_rec_iter_start(); iter;
 448              iter = ftrace_rec_iter_next(iter)) {
 449                 rec = ftrace_rec_iter_record(iter);
 450                 ret = __ftrace_replace_code(rec, enable);
 451                 if (ret) {
 452                         ftrace_bug(ret, rec->ip);
 453                         return;
 454                 }
 455         }
 456 }
 457
 458 void arch_ftrace_update_code(int command)
 459 {
 460         if (command & FTRACE_UPDATE_CALLS)
 461                 ftrace_replace_code(1);
 462         else if (command & FTRACE_DISABLE_CALLS)
 463                 ftrace_replace_code(0);
 464
 465         if (command & FTRACE_UPDATE_TRACE_FUNC)
 466                 ftrace_update_ftrace_func(ftrace_trace_function);
 467
 468         if (command & FTRACE_START_FUNC_RET)
 469                 ftrace_enable_ftrace_graph_caller();
 470         else if (command & FTRACE_STOP_FUNC_RET)
 471                 ftrace_disable_ftrace_graph_caller();
 472 }
 473
 474 int __init ftrace_dyn_arch_init(void)
 475 {
 476         return 0;
 477 }
 478 #endif /* CONFIG_DYNAMIC_FTRACE */
 479
 480 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 481
 482 #ifdef CONFIG_DYNAMIC_FTRACE
 483 extern void ftrace_graph_call(void);
 484 extern void ftrace_graph_stub(void);
 485
 486 int ftrace_enable_ftrace_graph_caller(void)
 487 {
 488         unsigned long ip = (unsigned long)(&ftrace_graph_call);
 489         unsigned long addr = (unsigned long)(&ftrace_graph_caller);
 490         unsigned long stub = (unsigned long)(&ftrace_graph_stub);
 491         unsigned int old, new;
 492
 493         old = ftrace_call_replace(ip, stub, 0);
 494         new = ftrace_call_replace(ip, addr, 0);
 495
 496         return ftrace_modify_code(ip, old, new);
 497 }
 498
 499 int ftrace_disable_ftrace_graph_caller(void)
 500 {
 501         unsigned long ip = (unsigned long)(&ftrace_graph_call);
 502         unsigned long addr = (unsigned long)(&ftrace_graph_caller);
 503         unsigned long stub = (unsigned long)(&ftrace_graph_stub);
 504         unsigned int old, new;
 505
 506         old = ftrace_call_replace(ip, addr, 0);
 507         new = ftrace_call_replace(ip, stub, 0);
 508
 509         return ftrace_modify_code(ip, old, new);
 510 }
 511 #endif /* CONFIG_DYNAMIC_FTRACE */
 512
 513 #ifdef CONFIG_PPC64
 514 extern void mod_return_to_handler(void);
 515 #endif
 516
 517 /*
 518  * Hook the return address and push it in the stack of return addrs
 519  * in current thread info.
 520  */
 521 void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr)
 522 {
 523         unsigned long old;
 524         int faulted;
 525         struct ftrace_graph_ent trace;
 526         unsigned long return_hooker = (unsigned long)&return_to_handler;
 527
 528         if (unlikely(ftrace_graph_is_dead()))
 529                 return;
 530
 531         if (unlikely(atomic_read(&current->tracing_graph_pause)))
 532                 return;
 533
 534 #ifdef CONFIG_PPC64
 535         /* non core kernel code needs to save and restore the TOC */
 536         if (REGION_ID(self_addr) != KERNEL_REGION_ID)
 537                 return_hooker = (unsigned long)&mod_return_to_handler;
 538 #endif
 539
 540         return_hooker = ppc_function_entry((void *)return_hooker);
 541
 542         /*
 543          * Protect against fault, even if it shouldn't
 544          * happen. This tool is too much intrusive to
 545          * ignore such a protection.
 546          */
 547         asm volatile(
 548                 "1: " PPC_LL "%[old], 0(%[parent])\n"
 549                 "2: " PPC_STL "%[return_hooker], 0(%[parent])\n"
 550                 "   li %[faulted], 0\n"
 551                 "3:\n"
 552
 553                 ".section .fixup, \"ax\"\n"
 554                 "4: li %[faulted], 1\n"
 555                 "   b 3b\n"
 556                 ".previous\n"
 557
 558                 ".section __ex_table,\"a\"\n"
 559                         PPC_LONG_ALIGN "\n"
 560                         PPC_LONG "1b,4b\n"
 561                         PPC_LONG "2b,4b\n"
 562                 ".previous"
 563
 564                 : [old] "=&r" (old), [faulted] "=r" (faulted)
 565                 : [parent] "r" (parent), [return_hooker] "r" (return_hooker)
 566                 : "memory"
 567         );
 568
 569         if (unlikely(faulted)) {
 570                 ftrace_graph_stop();
 571                 WARN_ON(1);
 572                 return;
 573         }
 574
 575         trace.func = self_addr;
 576         trace.depth = current->curr_ret_stack + 1;
 577
 578         /* Only trace if the calling function expects to */
 579         if (!ftrace_graph_entry(&trace)) {
 580                 *parent = old;
 581                 return;
 582         }
 583
 584         if (ftrace_push_return_trace(old, self_addr, &trace.depth, 0) == -EBUSY)
 585                 *parent = old;
 586 }
 587 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 588
 589 #if defined(CONFIG_FTRACE_SYSCALLS) && defined(CONFIG_PPC64)
 590 unsigned long __init arch_syscall_addr(int nr)
 591 {
 592         return sys_call_table[nr*2];
 593 }
 594 #endif /* CONFIG_FTRACE_SYSCALLS && CONFIG_PPC64 */