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                 pr_err("%p: replaced (%#x) != old (%#x)",
  66                 (void *)ip, replaced, old);
  67                 return -EINVAL;
  68         }
  69
  70         /* replace the text with the new text */
  71         if (patch_instruction((unsigned int *)ip, new))
  72                 return -EPERM;
  73
  74         return 0;
  75 }
  76
  77 /*
  78  * Helper functions that are the same for both PPC64 and PPC32.
  79  */
  80 static int test_24bit_addr(unsigned long ip, unsigned long addr)
  81 {
  82         addr = ppc_function_entry((void *)addr);
  83
  84         /* use the create_branch to verify that this offset can be branched */
  85         return create_branch((unsigned int *)ip, addr, 0);
  86 }
  87
  88 #ifdef CONFIG_MODULES
  89
  90 static int is_bl_op(unsigned int op)
  91 {
  92         return (op & 0xfc000003) == 0x48000001;
  93 }
  94
  95 static unsigned long find_bl_target(unsigned long ip, unsigned int op)
  96 {
  97         static int offset;
  98
  99         offset = (op & 0x03fffffc);
 100         /* make it signed */
 101         if (offset & 0x02000000)
 102                 offset |= 0xfe000000;
 103
 104         return ip + (long)offset;
 105 }
 106
 107 #ifdef CONFIG_PPC64
 108 static int
 109 __ftrace_make_nop(struct module *mod,
 110                   struct dyn_ftrace *rec, unsigned long addr)
 111 {
 112         unsigned long entry, ptr, tramp;
 113         unsigned long ip = rec->ip;
 114         unsigned int op, pop;
 115
 116         /* read where this goes */
 117         if (probe_kernel_read(&op, (void *)ip, sizeof(int))) {
 118                 pr_err("Fetching opcode failed.\n");
 119                 return -EFAULT;
 120         }
 121
 122         /* Make sure that that this is still a 24bit jump */
 123         if (!is_bl_op(op)) {
 124                 pr_err("Not expected bl: opcode is %x\n", op);
 125                 return -EINVAL;
 126         }
 127
 128         /* lets find where the pointer goes */
 129         tramp = find_bl_target(ip, op);
 130
 131         pr_devel("ip:%lx jumps to %lx", ip, tramp);
 132
 133         if (module_trampoline_target(mod, tramp, &ptr)) {
 134                 pr_err("Failed to get trampoline target\n");
 135                 return -EFAULT;
 136         }
 137
 138         pr_devel("trampoline target %lx", ptr);
 139
 140         entry = ppc_global_function_entry((void *)addr);
 141         /* This should match what was called */
 142         if (ptr != entry) {
 143                 pr_err("addr %lx does not match expected %lx\n", ptr, entry);
 144                 return -EINVAL;
 145         }
 146
 147         /*
 148          * Our original call site looks like:
 149          *
 150          * bl <tramp>
 151          * ld r2,XX(r1)
 152          *
 153          * Milton Miller pointed out that we can not simply nop the branch.
 154          * If a task was preempted when calling a trace function, the nops
 155          * will remove the way to restore the TOC in r2 and the r2 TOC will
 156          * get corrupted.
 157          *
 158          * Use a b +8 to jump over the load.
 159          */
 160
 161         pop = PPC_INST_BRANCH | 8;      /* b +8 */
 162
 163         /*
 164          * Check what is in the next instruction. We can see ld r2,40(r1), but
 165          * on first pass after boot we will see mflr r0.
 166          */
 167         if (probe_kernel_read(&op, (void *)(ip+4), MCOUNT_INSN_SIZE)) {
 168                 pr_err("Fetching op failed.\n");
 169                 return -EFAULT;
 170         }
 171
 172         if (op != PPC_INST_LD_TOC) {
 173                 unsigned int inst;
 174
 175                 if (probe_kernel_read(&inst, (void *)(ip - 4), 4)) {
 176                         pr_err("Fetching instruction at %lx failed.\n", ip - 4);
 177                         return -EFAULT;
 178                 }
 179
 180                 /* We expect either a mlfr r0, or a std r0, LRSAVE(r1) */
 181                 if (inst != PPC_INST_MFLR && inst != PPC_INST_STD_LR) {
 182                         pr_err("Unexpected instructions around bl _mcount\n"
 183                                "when enabling dynamic ftrace!\t"
 184                                "(%08x,bl,%08x)\n", inst, op);
 185                         return -EINVAL;
 186                 }
 187
 188                 /* When using -mkernel_profile there is no load to jump over */
 189                 pop = PPC_INST_NOP;
 190         }
 191
 192         if (patch_instruction((unsigned int *)ip, pop)) {
 193                 pr_err("Patching NOP failed.\n");
 194                 return -EPERM;
 195         }
 196
 197         return 0;
 198 }
 199
 200 #else /* !PPC64 */
 201 static int
 202 __ftrace_make_nop(struct module *mod,
 203                   struct dyn_ftrace *rec, unsigned long addr)
 204 {
 205         unsigned int op;
 206         unsigned int jmp[4];
 207         unsigned long ip = rec->ip;
 208         unsigned long tramp;
 209
 210         if (probe_kernel_read(&op, (void *)ip, MCOUNT_INSN_SIZE))
 211                 return -EFAULT;
 212
 213         /* Make sure that that this is still a 24bit jump */
 214         if (!is_bl_op(op)) {
 215                 pr_err("Not expected bl: opcode is %x\n", op);
 216                 return -EINVAL;
 217         }
 218
 219         /* lets find where the pointer goes */
 220         tramp = find_bl_target(ip, op);
 221
 222         /*
 223          * On PPC32 the trampoline looks like:
 224          *  0x3d, 0x80, 0x00, 0x00  lis r12,sym@ha
 225          *  0x39, 0x8c, 0x00, 0x00  addi r12,r12,sym@l
 226          *  0x7d, 0x89, 0x03, 0xa6  mtctr r12
 227          *  0x4e, 0x80, 0x04, 0x20  bctr
 228          */
 229
 230         pr_devel("ip:%lx jumps to %lx", ip, tramp);
 231
 232         /* Find where the trampoline jumps to */
 233         if (probe_kernel_read(jmp, (void *)tramp, sizeof(jmp))) {
 234                 pr_err("Failed to read %lx\n", tramp);
 235                 return -EFAULT;
 236         }
 237
 238         pr_devel(" %08x %08x ", jmp[0], jmp[1]);
 239
 240         /* verify that this is what we expect it to be */
 241         if (((jmp[0] & 0xffff0000) != 0x3d800000) ||
 242             ((jmp[1] & 0xffff0000) != 0x398c0000) ||
 243             (jmp[2] != 0x7d8903a6) ||
 244             (jmp[3] != 0x4e800420)) {
 245                 pr_err("Not a trampoline\n");
 246                 return -EINVAL;
 247         }
 248
 249         tramp = (jmp[1] & 0xffff) |
 250                 ((jmp[0] & 0xffff) << 16);
 251         if (tramp & 0x8000)
 252                 tramp -= 0x10000;
 253
 254         pr_devel(" %lx ", tramp);
 255
 256         if (tramp != addr) {
 257                 pr_err("Trampoline location %08lx does not match addr\n",
 258                        tramp);
 259                 return -EINVAL;
 260         }
 261
 262         op = PPC_INST_NOP;
 263
 264         if (patch_instruction((unsigned int *)ip, op))
 265                 return -EPERM;
 266
 267         return 0;
 268 }
 269 #endif /* PPC64 */
 270 #endif /* CONFIG_MODULES */
 271
 272 int ftrace_make_nop(struct module *mod,
 273                     struct dyn_ftrace *rec, unsigned long addr)
 274 {
 275         unsigned long ip = rec->ip;
 276         unsigned int old, new;
 277
 278         /*
 279          * If the calling address is more that 24 bits away,
 280          * then we had to use a trampoline to make the call.
 281          * Otherwise just update the call site.
 282          */
 283         if (test_24bit_addr(ip, addr)) {
 284                 /* within range */
 285                 old = ftrace_call_replace(ip, addr, 1);
 286                 new = PPC_INST_NOP;
 287                 return ftrace_modify_code(ip, old, new);
 288         }
 289
 290 #ifdef CONFIG_MODULES
 291         /*
 292          * Out of range jumps are called from modules.
 293          * We should either already have a pointer to the module
 294          * or it has been passed in.
 295          */
 296         if (!rec->arch.mod) {
 297                 if (!mod) {
 298                         pr_err("No module loaded addr=%lx\n", addr);
 299                         return -EFAULT;
 300                 }
 301                 rec->arch.mod = mod;
 302         } else if (mod) {
 303                 if (mod != rec->arch.mod) {
 304                         pr_err("Record mod %p not equal to passed in mod %p\n",
 305                                rec->arch.mod, mod);
 306                         return -EINVAL;
 307                 }
 308                 /* nothing to do if mod == rec->arch.mod */
 309         } else
 310                 mod = rec->arch.mod;
 311
 312         return __ftrace_make_nop(mod, rec, addr);
 313 #else
 314         /* We should not get here without modules */
 315         return -EINVAL;
 316 #endif /* CONFIG_MODULES */
 317 }
 318
 319 #ifdef CONFIG_MODULES
 320 #ifdef CONFIG_PPC64
 321 /*
 322  * Examine the existing instructions for __ftrace_make_call.
 323  * They should effectively be a NOP, and follow formal constraints,
 324  * depending on the ABI. Return false if they don't.
 325  */
 326 #ifndef CC_USING_MPROFILE_KERNEL
 327 static int
 328 expected_nop_sequence(void *ip, unsigned int op0, unsigned int op1)
 329 {
 330         /*
 331          * We expect to see:
 332          *
 333          * b +8
 334          * ld r2,XX(r1)
 335          *
 336          * The load offset is different depending on the ABI. For simplicity
 337          * just mask it out when doing the compare.
 338          */
 339         if ((op0 != 0x48000008) || ((op1 & 0xffff0000) != 0xe8410000))
 340                 return 0;
 341         return 1;
 342 }
 343 #else
 344 static int
 345 expected_nop_sequence(void *ip, unsigned int op0, unsigned int op1)
 346 {
 347         /* look for patched "NOP" on ppc64 with -mprofile-kernel */
 348         if (op0 != PPC_INST_NOP)
 349                 return 0;
 350         return 1;
 351 }
 352 #endif
 353
 354 static int
 355 __ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
 356 {
 357         unsigned int op[2];
 358         void *ip = (void *)rec->ip;
 359
 360         /* read where this goes */
 361         if (probe_kernel_read(op, ip, sizeof(op)))
 362                 return -EFAULT;
 363
 364         if (!expected_nop_sequence(ip, op[0], op[1])) {
 365                 pr_err("Unexpected call sequence at %p: %x %x\n",
 366                 ip, op[0], op[1]);
 367                 return -EINVAL;
 368         }
 369
 370         /* If we never set up a trampoline to ftrace_caller, then bail */
 371         if (!rec->arch.mod->arch.tramp) {
 372                 pr_err("No ftrace trampoline\n");
 373                 return -EINVAL;
 374         }
 375
 376         /* Ensure branch is within 24 bits */
 377         if (!create_branch(ip, rec->arch.mod->arch.tramp, BRANCH_SET_LINK)) {
 378                 pr_err("Branch out of range\n");
 379                 return -EINVAL;
 380         }
 381
 382         if (patch_branch(ip, rec->arch.mod->arch.tramp, BRANCH_SET_LINK)) {
 383                 pr_err("REL24 out of range!\n");
 384                 return -EINVAL;
 385         }
 386
 387         return 0;
 388 }
 389
 390 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
 391 int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
 392                         unsigned long addr)
 393 {
 394         return ftrace_make_call(rec, addr);
 395 }
 396 #endif
 397
 398 #else  /* !CONFIG_PPC64: */
 399 static int
 400 __ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
 401 {
 402         unsigned int op;
 403         unsigned long ip = rec->ip;
 404
 405         /* read where this goes */
 406         if (probe_kernel_read(&op, (void *)ip, MCOUNT_INSN_SIZE))
 407                 return -EFAULT;
 408
 409         /* It should be pointing to a nop */
 410         if (op != PPC_INST_NOP) {
 411                 pr_err("Expected NOP but have %x\n", op);
 412                 return -EINVAL;
 413         }
 414
 415         /* If we never set up a trampoline to ftrace_caller, then bail */
 416         if (!rec->arch.mod->arch.tramp) {
 417                 pr_err("No ftrace trampoline\n");
 418                 return -EINVAL;
 419         }
 420
 421         /* create the branch to the trampoline */
 422         op = create_branch((unsigned int *)ip,
 423                            rec->arch.mod->arch.tramp, BRANCH_SET_LINK);
 424         if (!op) {
 425                 pr_err("REL24 out of range!\n");
 426                 return -EINVAL;
 427         }
 428
 429         pr_devel("write to %lx\n", rec->ip);
 430
 431         if (patch_instruction((unsigned int *)ip, op))
 432                 return -EPERM;
 433
 434         return 0;
 435 }
 436 #endif /* CONFIG_PPC64 */
 437 #endif /* CONFIG_MODULES */
 438
 439 int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
 440 {
 441         unsigned long ip = rec->ip;
 442         unsigned int old, new;
 443
 444         /*
 445          * If the calling address is more that 24 bits away,
 446          * then we had to use a trampoline to make the call.
 447          * Otherwise just update the call site.
 448          */
 449         if (test_24bit_addr(ip, addr)) {
 450                 /* within range */
 451                 old = PPC_INST_NOP;
 452                 new = ftrace_call_replace(ip, addr, 1);
 453                 return ftrace_modify_code(ip, old, new);
 454         }
 455
 456 #ifdef CONFIG_MODULES
 457         /*
 458          * Out of range jumps are called from modules.
 459          * Being that we are converting from nop, it had better
 460          * already have a module defined.
 461          */
 462         if (!rec->arch.mod) {
 463                 pr_err("No module loaded\n");
 464                 return -EINVAL;
 465         }
 466
 467         return __ftrace_make_call(rec, addr);
 468 #else
 469         /* We should not get here without modules */
 470         return -EINVAL;
 471 #endif /* CONFIG_MODULES */
 472 }
 473
 474 int ftrace_update_ftrace_func(ftrace_func_t func)
 475 {
 476         unsigned long ip = (unsigned long)(&ftrace_call);
 477         unsigned int old, new;
 478         int ret;
 479
 480         old = *(unsigned int *)&ftrace_call;
 481         new = ftrace_call_replace(ip, (unsigned long)func, 1);
 482         ret = ftrace_modify_code(ip, old, new);
 483
 484         return ret;
 485 }
 486
 487 static int __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
 488 {
 489         unsigned long ftrace_addr = (unsigned long)FTRACE_ADDR;
 490         int ret;
 491
 492         ret = ftrace_update_record(rec, enable);
 493
 494         switch (ret) {
 495         case FTRACE_UPDATE_IGNORE:
 496                 return 0;
 497         case FTRACE_UPDATE_MAKE_CALL:
 498                 return ftrace_make_call(rec, ftrace_addr);
 499         case FTRACE_UPDATE_MAKE_NOP:
 500                 return ftrace_make_nop(NULL, rec, ftrace_addr);
 501         }
 502
 503         return 0;
 504 }
 505
 506 void ftrace_replace_code(int enable)
 507 {
 508         struct ftrace_rec_iter *iter;
 509         struct dyn_ftrace *rec;
 510         int ret;
 511
 512         for (iter = ftrace_rec_iter_start(); iter;
 513              iter = ftrace_rec_iter_next(iter)) {
 514                 rec = ftrace_rec_iter_record(iter);
 515                 ret = __ftrace_replace_code(rec, enable);
 516                 if (ret) {
 517                         ftrace_bug(ret, rec);
 518                         return;
 519                 }
 520         }
 521 }
 522
 523 /*
 524  * Use the default ftrace_modify_all_code, but without
 525  * stop_machine().
 526  */
 527 void arch_ftrace_update_code(int command)
 528 {
 529         ftrace_modify_all_code(command);
 530 }
 531
 532 int __init ftrace_dyn_arch_init(void)
 533 {
 534         return 0;
 535 }
 536 #endif /* CONFIG_DYNAMIC_FTRACE */
 537
 538 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 539
 540 #ifdef CONFIG_DYNAMIC_FTRACE
 541 extern void ftrace_graph_call(void);
 542 extern void ftrace_graph_stub(void);
 543
 544 int ftrace_enable_ftrace_graph_caller(void)
 545 {
 546         unsigned long ip = (unsigned long)(&ftrace_graph_call);
 547         unsigned long addr = (unsigned long)(&ftrace_graph_caller);
 548         unsigned long stub = (unsigned long)(&ftrace_graph_stub);
 549         unsigned int old, new;
 550
 551         old = ftrace_call_replace(ip, stub, 0);
 552         new = ftrace_call_replace(ip, addr, 0);
 553
 554         return ftrace_modify_code(ip, old, new);
 555 }
 556
 557 int ftrace_disable_ftrace_graph_caller(void)
 558 {
 559         unsigned long ip = (unsigned long)(&ftrace_graph_call);
 560         unsigned long addr = (unsigned long)(&ftrace_graph_caller);
 561         unsigned long stub = (unsigned long)(&ftrace_graph_stub);
 562         unsigned int old, new;
 563
 564         old = ftrace_call_replace(ip, addr, 0);
 565         new = ftrace_call_replace(ip, stub, 0);
 566
 567         return ftrace_modify_code(ip, old, new);
 568 }
 569 #endif /* CONFIG_DYNAMIC_FTRACE */
 570
 571 /*
 572  * Hook the return address and push it in the stack of return addrs
 573  * in current thread info. Return the address we want to divert to.
 574  */
 575 unsigned long prepare_ftrace_return(unsigned long parent, unsigned long ip)
 576 {
 577         struct ftrace_graph_ent trace;
 578         unsigned long return_hooker;
 579
 580         if (unlikely(ftrace_graph_is_dead()))
 581                 goto out;
 582
 583         if (unlikely(atomic_read(&current->tracing_graph_pause)))
 584                 goto out;
 585
 586         return_hooker = ppc_function_entry(return_to_handler);
 587
 588         trace.func = ip;
 589         trace.depth = current->curr_ret_stack + 1;
 590
 591         /* Only trace if the calling function expects to */
 592         if (!ftrace_graph_entry(&trace))
 593                 goto out;
 594
 595         if (ftrace_push_return_trace(parent, ip, &trace.depth, 0) == -EBUSY)
 596                 goto out;
 597
 598         parent = return_hooker;
 599 out:
 600         return parent;
 601 }
 602 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 603
 604 #if defined(CONFIG_FTRACE_SYSCALLS) && defined(CONFIG_PPC64)
 605 unsigned long __init arch_syscall_addr(int nr)
 606 {
 607         return sys_call_table[nr*2];
 608 }
 609 #endif /* CONFIG_FTRACE_SYSCALLS && CONFIG_PPC64 */