tools/perf/arch/powerpc/util/skip-callchain-idx.c

   1 /*
   2  * Use DWARF Debug information to skip unnecessary callchain entries.
   3  *
   4  * Copyright (C) 2014 Sukadev Bhattiprolu, IBM Corporation.
   5  * Copyright (C) 2014 Ulrich Weigand, IBM Corporation.
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; either version
  10  * 2 of the License, or (at your option) any later version.
  11  */
  12 #include <inttypes.h>
  13 #include <dwarf.h>
  14 #include <elfutils/libdwfl.h>
  15
  16 #include "util/thread.h"
  17 #include "util/callchain.h"
  18 #include "util/debug.h"
  19 #include "util/dso.h"
  20 #include "util/map.h"
  21 #include "util/symbol.h"
  22
  23 /*
  24  * When saving the callchain on Power, the kernel conservatively saves
  25  * excess entries in the callchain. A few of these entries are needed
  26  * in some cases but not others. If the unnecessary entries are not
  27  * ignored, we end up with duplicate arcs in the call-graphs. Use
  28  * DWARF debug information to skip over any unnecessary callchain
  29  * entries.
  30  *
  31  * See function header for arch_adjust_callchain() below for more details.
  32  *
  33  * The libdwfl code in this file is based on code from elfutils
  34  * (libdwfl/argp-std.c, libdwfl/tests/addrcfi.c, etc).
  35  */
  36 static char *debuginfo_path;
  37
  38 static const Dwfl_Callbacks offline_callbacks = {
  39         .debuginfo_path = &debuginfo_path,
  40         .find_debuginfo = dwfl_standard_find_debuginfo,
  41         .section_address = dwfl_offline_section_address,
  42 };
  43
  44
  45 /*
  46  * Use the DWARF expression for the Call-frame-address and determine
  47  * if return address is in LR and if a new frame was allocated.
  48  */
  49 static int check_return_reg(int ra_regno, Dwarf_Frame *frame)
  50 {
  51         Dwarf_Op ops_mem[2];
  52         Dwarf_Op dummy;
  53         Dwarf_Op *ops = &dummy;
  54         size_t nops;
  55         int result;
  56
  57         result = dwarf_frame_register(frame, ra_regno, ops_mem, &ops, &nops);
  58         if (result < 0) {
  59                 pr_debug("dwarf_frame_register() %s\n", dwarf_errmsg(-1));
  60                 return -1;
  61         }
  62
  63         /*
  64          * Check if return address is on the stack. If return address
  65          * is in a register (typically R0), it is yet to be saved on
  66          * the stack.
  67          */
  68         if ((nops != 0 || ops != NULL) &&
  69                 !(nops == 1 && ops[0].atom == DW_OP_regx &&
  70                         ops[0].number2 == 0 && ops[0].offset == 0))
  71                 return 0;
  72
  73         /*
  74          * Return address is in LR. Check if a frame was allocated
  75          * but not-yet used.
  76          */
  77         result = dwarf_frame_cfa(frame, &ops, &nops);
  78         if (result < 0) {
  79                 pr_debug("dwarf_frame_cfa() returns %d, %s\n", result,
  80                                         dwarf_errmsg(-1));
  81                 return -1;
  82         }
  83
  84         /*
  85          * If call frame address is in r1, no new frame was allocated.
  86          */
  87         if (nops == 1 && ops[0].atom == DW_OP_bregx && ops[0].number == 1 &&
  88                                 ops[0].number2 == 0)
  89                 return 1;
  90
  91         /*
  92          * A new frame was allocated but has not yet been used.
  93          */
  94         return 2;
  95 }
  96
  97 /*
  98  * Get the DWARF frame from the .eh_frame section.
  99  */
 100 static Dwarf_Frame *get_eh_frame(Dwfl_Module *mod, Dwarf_Addr pc)
 101 {
 102         int             result;
 103         Dwarf_Addr      bias;
 104         Dwarf_CFI       *cfi;
 105         Dwarf_Frame     *frame;
 106
 107         cfi = dwfl_module_eh_cfi(mod, &bias);
 108         if (!cfi) {
 109                 pr_debug("%s(): no CFI - %s\n", __func__, dwfl_errmsg(-1));
 110                 return NULL;
 111         }
 112
 113         result = dwarf_cfi_addrframe(cfi, pc-bias, &frame);
 114         if (result) {
 115                 pr_debug("%s(): %s\n", __func__, dwfl_errmsg(-1));
 116                 return NULL;
 117         }
 118
 119         return frame;
 120 }
 121
 122 /*
 123  * Get the DWARF frame from the .debug_frame section.
 124  */
 125 static Dwarf_Frame *get_dwarf_frame(Dwfl_Module *mod, Dwarf_Addr pc)
 126 {
 127         Dwarf_CFI       *cfi;
 128         Dwarf_Addr      bias;
 129         Dwarf_Frame     *frame;
 130         int             result;
 131
 132         cfi = dwfl_module_dwarf_cfi(mod, &bias);
 133         if (!cfi) {
 134                 pr_debug("%s(): no CFI - %s\n", __func__, dwfl_errmsg(-1));
 135                 return NULL;
 136         }
 137
 138         result = dwarf_cfi_addrframe(cfi, pc-bias, &frame);
 139         if (result) {
 140                 pr_debug("%s(): %s\n", __func__, dwfl_errmsg(-1));
 141                 return NULL;
 142         }
 143
 144         return frame;
 145 }
 146
 147 /*
 148  * Return:
 149  *      0 if return address for the program counter @pc is on stack
 150  *      1 if return address is in LR and no new stack frame was allocated
 151  *      2 if return address is in LR and a new frame was allocated (but not
 152  *              yet used)
 153  *      -1 in case of errors
 154  */
 155 static int check_return_addr(struct dso *dso, u64 map_start, Dwarf_Addr pc)
 156 {
 157         int             rc = -1;
 158         Dwfl            *dwfl;
 159         Dwfl_Module     *mod;
 160         Dwarf_Frame     *frame;
 161         int             ra_regno;
 162         Dwarf_Addr      start = pc;
 163         Dwarf_Addr      end = pc;
 164         bool            signalp;
 165         const char      *exec_file = dso->long_name;
 166
 167         dwfl = dso->dwfl;
 168
 169         if (!dwfl) {
 170                 dwfl = dwfl_begin(&offline_callbacks);
 171                 if (!dwfl) {
 172                         pr_debug("dwfl_begin() failed: %s\n", dwarf_errmsg(-1));
 173                         return -1;
 174                 }
 175
 176                 mod = dwfl_report_elf(dwfl, exec_file, exec_file, -1,
 177                                                 map_start, false);
 178                 if (!mod) {
 179                         pr_debug("dwfl_report_elf() failed %s\n",
 180                                                 dwarf_errmsg(-1));
 181                         /*
 182                          * We normally cache the DWARF debug info and never
 183                          * call dwfl_end(). But to prevent fd leak, free in
 184                          * case of error.
 185                          */
 186                         dwfl_end(dwfl);
 187                         goto out;
 188                 }
 189                 dso->dwfl = dwfl;
 190         }
 191
 192         mod = dwfl_addrmodule(dwfl, pc);
 193         if (!mod) {
 194                 pr_debug("dwfl_addrmodule() failed, %s\n", dwarf_errmsg(-1));
 195                 goto out;
 196         }
 197
 198         /*
 199          * To work with split debug info files (eg: glibc), check both
 200          * .eh_frame and .debug_frame sections of the ELF header.
 201          */
 202         frame = get_eh_frame(mod, pc);
 203         if (!frame) {
 204                 frame = get_dwarf_frame(mod, pc);
 205                 if (!frame)
 206                         goto out;
 207         }
 208
 209         ra_regno = dwarf_frame_info(frame, &start, &end, &signalp);
 210         if (ra_regno < 0) {
 211                 pr_debug("Return address register unavailable: %s\n",
 212                                 dwarf_errmsg(-1));
 213                 goto out;
 214         }
 215
 216         rc = check_return_reg(ra_regno, frame);
 217
 218 out:
 219         return rc;
 220 }
 221
 222 /*
 223  * The callchain saved by the kernel always includes the link register (LR).
 224  *
 225  *      0:      PERF_CONTEXT_USER
 226  *      1:      Program counter (Next instruction pointer)
 227  *      2:      LR value
 228  *      3:      Caller's caller
 229  *      4:      ...
 230  *
 231  * The value in LR is only needed when it holds a return address. If the
 232  * return address is on the stack, we should ignore the LR value.
 233  *
 234  * Further, when the return address is in the LR, if a new frame was just
 235  * allocated but the LR was not saved into it, then the LR contains the
 236  * caller, slot 4: contains the caller's caller and the contents of slot 3:
 237  * (chain->ips[3]) is undefined and must be ignored.
 238  *
 239  * Use DWARF debug information to determine if any entries need to be skipped.
 240  *
 241  * Return:
 242  *      index:  of callchain entry that needs to be ignored (if any)
 243  *      -1      if no entry needs to be ignored or in case of errors
 244  */
 245 int arch_skip_callchain_idx(struct thread *thread, struct ip_callchain *chain)
 246 {
 247         struct addr_location al;
 248         struct dso *dso = NULL;
 249         int rc;
 250         u64 ip;
 251         u64 skip_slot = -1;
 252
 253         if (!chain || chain->nr < 3)
 254                 return skip_slot;
 255
 256         ip = chain->ips[1];
 257
 258         thread__find_symbol(thread, PERF_RECORD_MISC_USER, ip, &al);
 259
 260         if (al.map)
 261                 dso = al.map->dso;
 262
 263         if (!dso) {
 264                 pr_debug("%" PRIx64 " dso is NULL\n", ip);
 265                 return skip_slot;
 266         }
 267
 268         rc = check_return_addr(dso, al.map->start, ip);
 269
 270         pr_debug("[DSO %s, sym %s, ip 0x%" PRIx64 "] rc %d\n",
 271                                 dso->long_name, al.sym->name, ip, rc);
 272
 273         if (rc == 0) {
 274                 /*
 275                  * Return address on stack. Ignore LR value in callchain
 276                  */
 277                 skip_slot = 2;
 278         } else if (rc == 2) {
 279                 /*
 280                  * New frame allocated but return address still in LR.
 281                  * Ignore the caller's caller entry in callchain.
 282                  */
 283                 skip_slot = 3;
 284         }
 285         return skip_slot;
 286 }