arch/arc/include/asm/perf_event.h

   1 /* SPDX-License-Identifier: GPL-2.0-only */
   2 /*
   3  * Linux performance counter support for ARC
   4  *
   5  * Copyright (C) 2014-2015 Synopsys, Inc. (www.synopsys.com)
   6  * Copyright (C) 2011-2013 Synopsys, Inc. (www.synopsys.com)
   7  */
   8
   9 #ifndef __ASM_PERF_EVENT_H
  10 #define __ASM_PERF_EVENT_H
  11
  12 /* Max number of counters that PCT block may ever have */
  13 #define ARC_PERF_MAX_COUNTERS   32
  14
  15 #define ARC_REG_CC_BUILD        0xF6
  16 #define ARC_REG_CC_INDEX        0x240
  17 #define ARC_REG_CC_NAME0        0x241
  18 #define ARC_REG_CC_NAME1        0x242
  19
  20 #define ARC_REG_PCT_BUILD       0xF5
  21 #define ARC_REG_PCT_COUNTL      0x250
  22 #define ARC_REG_PCT_COUNTH      0x251
  23 #define ARC_REG_PCT_SNAPL       0x252
  24 #define ARC_REG_PCT_SNAPH       0x253
  25 #define ARC_REG_PCT_CONFIG      0x254
  26 #define ARC_REG_PCT_CONTROL     0x255
  27 #define ARC_REG_PCT_INDEX       0x256
  28 #define ARC_REG_PCT_INT_CNTL    0x25C
  29 #define ARC_REG_PCT_INT_CNTH    0x25D
  30 #define ARC_REG_PCT_INT_CTRL    0x25E
  31 #define ARC_REG_PCT_INT_ACT     0x25F
  32
  33 #define ARC_REG_PCT_CONFIG_USER (1 << 18)       /* count in user mode */
  34 #define ARC_REG_PCT_CONFIG_KERN (1 << 19)       /* count in kernel mode */
  35
  36 #define ARC_REG_PCT_CONTROL_CC  (1 << 16)       /* clear counts */
  37 #define ARC_REG_PCT_CONTROL_SN  (1 << 17)       /* snapshot */
  38
  39 struct arc_reg_pct_build {
  40 #ifdef CONFIG_CPU_BIG_ENDIAN
  41         unsigned int m:8, c:8, r:5, i:1, s:2, v:8;
  42 #else
  43         unsigned int v:8, s:2, i:1, r:5, c:8, m:8;
  44 #endif
  45 };
  46
  47 struct arc_reg_cc_build {
  48 #ifdef CONFIG_CPU_BIG_ENDIAN
  49         unsigned int c:16, r:8, v:8;
  50 #else
  51         unsigned int v:8, r:8, c:16;
  52 #endif
  53 };
  54
  55 #define PERF_COUNT_ARC_DCLM     (PERF_COUNT_HW_MAX + 0)
  56 #define PERF_COUNT_ARC_DCSM     (PERF_COUNT_HW_MAX + 1)
  57 #define PERF_COUNT_ARC_ICM      (PERF_COUNT_HW_MAX + 2)
  58 #define PERF_COUNT_ARC_BPOK     (PERF_COUNT_HW_MAX + 3)
  59 #define PERF_COUNT_ARC_EDTLB    (PERF_COUNT_HW_MAX + 4)
  60 #define PERF_COUNT_ARC_EITLB    (PERF_COUNT_HW_MAX + 5)
  61 #define PERF_COUNT_ARC_LDC      (PERF_COUNT_HW_MAX + 6)
  62 #define PERF_COUNT_ARC_STC      (PERF_COUNT_HW_MAX + 7)
  63
  64 #define PERF_COUNT_ARC_HW_MAX   (PERF_COUNT_HW_MAX + 8)
  65
  66 /*
  67  * Some ARC pct quirks:
  68  *
  69  * PERF_COUNT_HW_STALLED_CYCLES_BACKEND
  70  * PERF_COUNT_HW_STALLED_CYCLES_FRONTEND
  71  *      The ARC 700 can either measure stalls per pipeline stage, or all stalls
  72  *      combined; for now we assign all stalls to STALLED_CYCLES_BACKEND
  73  *      and all pipeline flushes (e.g. caused by mispredicts, etc.) to
  74  *      STALLED_CYCLES_FRONTEND.
  75  *
  76  *      We could start multiple performance counters and combine everything
  77  *      afterwards, but that makes it complicated.
  78  *
  79  *      Note that I$ cache misses aren't counted by either of the two!
  80  */
  81
  82 /*
  83  * ARC PCT has hardware conditions with fixed "names" but variable "indexes"
  84  * (based on a specific RTL build)
  85  * Below is the static map between perf generic/arc specific event_id and
  86  * h/w condition names.
  87  * At the time of probe, we loop thru each index and find it's name to
  88  * complete the mapping of perf event_id to h/w index as latter is needed
  89  * to program the counter really
  90  */
  91 static const char * const arc_pmu_ev_hw_map[] = {
  92         /* count cycles */
  93         [PERF_COUNT_HW_CPU_CYCLES] = "crun",
  94         [PERF_COUNT_HW_REF_CPU_CYCLES] = "crun",
  95         [PERF_COUNT_HW_BUS_CYCLES] = "crun",
  96
  97         [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = "bflush",
  98         [PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = "bstall",
  99
 100         /* counts condition */
 101         [PERF_COUNT_HW_INSTRUCTIONS] = "iall",
 102         /* All jump instructions that are taken */
 103         [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = "ijmptak",
 104 #ifdef CONFIG_ISA_ARCV2
 105         [PERF_COUNT_HW_BRANCH_MISSES] = "bpmp",
 106 #else
 107         [PERF_COUNT_ARC_BPOK]         = "bpok",   /* NP-NT, PT-T, PNT-NT */
 108         [PERF_COUNT_HW_BRANCH_MISSES] = "bpfail", /* NP-T, PT-NT, PNT-T */
 109 #endif
 110         [PERF_COUNT_ARC_LDC] = "imemrdc",       /* Instr: mem read cached */
 111         [PERF_COUNT_ARC_STC] = "imemwrc",       /* Instr: mem write cached */
 112
 113         [PERF_COUNT_ARC_DCLM] = "dclm",         /* D-cache Load Miss */
 114         [PERF_COUNT_ARC_DCSM] = "dcsm",         /* D-cache Store Miss */
 115         [PERF_COUNT_ARC_ICM] = "icm",           /* I-cache Miss */
 116         [PERF_COUNT_ARC_EDTLB] = "edtlb",       /* D-TLB Miss */
 117         [PERF_COUNT_ARC_EITLB] = "eitlb",       /* I-TLB Miss */
 118
 119         [PERF_COUNT_HW_CACHE_REFERENCES] = "imemrdc",   /* Instr: mem read cached */
 120         [PERF_COUNT_HW_CACHE_MISSES] = "dclm",          /* D-cache Load Miss */
 121 };
 122
 123 #define C(_x)                   PERF_COUNT_HW_CACHE_##_x
 124 #define CACHE_OP_UNSUPPORTED    0xffff
 125
 126 static const unsigned arc_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 127         [C(L1D)] = {
 128                 [C(OP_READ)] = {
 129                         [C(RESULT_ACCESS)]      = PERF_COUNT_ARC_LDC,
 130                         [C(RESULT_MISS)]        = PERF_COUNT_ARC_DCLM,
 131                 },
 132                 [C(OP_WRITE)] = {
 133                         [C(RESULT_ACCESS)]      = PERF_COUNT_ARC_STC,
 134                         [C(RESULT_MISS)]        = PERF_COUNT_ARC_DCSM,
 135                 },
 136                 [C(OP_PREFETCH)] = {
 137                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 138                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 139                 },
 140         },
 141         [C(L1I)] = {
 142                 [C(OP_READ)] = {
 143                         [C(RESULT_ACCESS)]      = PERF_COUNT_HW_INSTRUCTIONS,
 144                         [C(RESULT_MISS)]        = PERF_COUNT_ARC_ICM,
 145                 },
 146                 [C(OP_WRITE)] = {
 147                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 148                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 149                 },
 150                 [C(OP_PREFETCH)] = {
 151                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 152                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 153                 },
 154         },
 155         [C(LL)] = {
 156                 [C(OP_READ)] = {
 157                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 158                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 159                 },
 160                 [C(OP_WRITE)] = {
 161                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 162                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 163                 },
 164                 [C(OP_PREFETCH)] = {
 165                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 166                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 167                 },
 168         },
 169         [C(DTLB)] = {
 170                 [C(OP_READ)] = {
 171                         [C(RESULT_ACCESS)]      = PERF_COUNT_ARC_LDC,
 172                         [C(RESULT_MISS)]        = PERF_COUNT_ARC_EDTLB,
 173                 },
 174                         /* DTLB LD/ST Miss not segregated by h/w*/
 175                 [C(OP_WRITE)] = {
 176                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 177                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 178                 },
 179                 [C(OP_PREFETCH)] = {
 180                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 181                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 182                 },
 183         },
 184         [C(ITLB)] = {
 185                 [C(OP_READ)] = {
 186                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 187                         [C(RESULT_MISS)]        = PERF_COUNT_ARC_EITLB,
 188                 },
 189                 [C(OP_WRITE)] = {
 190                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 191                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 192                 },
 193                 [C(OP_PREFETCH)] = {
 194                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 195                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 196                 },
 197         },
 198         [C(BPU)] = {
 199                 [C(OP_READ)] = {
 200                         [C(RESULT_ACCESS)] = PERF_COUNT_HW_BRANCH_INSTRUCTIONS,
 201                         [C(RESULT_MISS)]        = PERF_COUNT_HW_BRANCH_MISSES,
 202                 },
 203                 [C(OP_WRITE)] = {
 204                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 205                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 206                 },
 207                 [C(OP_PREFETCH)] = {
 208                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 209                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 210                 },
 211         },
 212         [C(NODE)] = {
 213                 [C(OP_READ)] = {
 214                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 215                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 216                 },
 217                 [C(OP_WRITE)] = {
 218                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 219                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 220                 },
 221                 [C(OP_PREFETCH)] = {
 222                         [C(RESULT_ACCESS)]      = CACHE_OP_UNSUPPORTED,
 223                         [C(RESULT_MISS)]        = CACHE_OP_UNSUPPORTED,
 224                 },
 225         },
 226 };
 227
 228 #endif /* __ASM_PERF_EVENT_H */