1 /* 2 * Linux performance counter support for ARC 3 * 4 * Copyright (C) 2014-2015 Synopsys, Inc. (www.synopsys.com) 5 * Copyright (C) 2011-2013 Synopsys, Inc. (www.synopsys.com) 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 * 11 */ 12 13 #ifndef __ASM_PERF_EVENT_H 14 #define __ASM_PERF_EVENT_H 15 16 /* Max number of counters that PCT block may ever have */ 17 #define ARC_PERF_MAX_COUNTERS 32 18 19 #define ARC_REG_CC_BUILD 0xF6 20 #define ARC_REG_CC_INDEX 0x240 21 #define ARC_REG_CC_NAME0 0x241 22 #define ARC_REG_CC_NAME1 0x242 23 24 #define ARC_REG_PCT_BUILD 0xF5 25 #define ARC_REG_PCT_COUNTL 0x250 26 #define ARC_REG_PCT_COUNTH 0x251 27 #define ARC_REG_PCT_SNAPL 0x252 28 #define ARC_REG_PCT_SNAPH 0x253 29 #define ARC_REG_PCT_CONFIG 0x254 30 #define ARC_REG_PCT_CONTROL 0x255 31 #define ARC_REG_PCT_INDEX 0x256 32 #define ARC_REG_PCT_INT_CNTL 0x25C 33 #define ARC_REG_PCT_INT_CNTH 0x25D 34 #define ARC_REG_PCT_INT_CTRL 0x25E 35 #define ARC_REG_PCT_INT_ACT 0x25F 36 37 #define ARC_REG_PCT_CONFIG_USER (1 << 18) /* count in user mode */ 38 #define ARC_REG_PCT_CONFIG_KERN (1 << 19) /* count in kernel mode */ 39 40 #define ARC_REG_PCT_CONTROL_CC (1 << 16) /* clear counts */ 41 #define ARC_REG_PCT_CONTROL_SN (1 << 17) /* snapshot */ 42 43 struct arc_reg_pct_build { 44 #ifdef CONFIG_CPU_BIG_ENDIAN 45 unsigned int m:8, c:8, r:5, i:1, s:2, v:8; 46 #else 47 unsigned int v:8, s:2, i:1, r:5, c:8, m:8; 48 #endif 49 }; 50 51 struct arc_reg_cc_build { 52 #ifdef CONFIG_CPU_BIG_ENDIAN 53 unsigned int c:16, r:8, v:8; 54 #else 55 unsigned int v:8, r:8, c:16; 56 #endif 57 }; 58 59 #define PERF_COUNT_ARC_DCLM (PERF_COUNT_HW_MAX + 0) 60 #define PERF_COUNT_ARC_DCSM (PERF_COUNT_HW_MAX + 1) 61 #define PERF_COUNT_ARC_ICM (PERF_COUNT_HW_MAX + 2) 62 #define PERF_COUNT_ARC_BPOK (PERF_COUNT_HW_MAX + 3) 63 #define PERF_COUNT_ARC_EDTLB (PERF_COUNT_HW_MAX + 4) 64 #define PERF_COUNT_ARC_EITLB (PERF_COUNT_HW_MAX + 5) 65 #define PERF_COUNT_ARC_LDC (PERF_COUNT_HW_MAX + 6) 66 #define PERF_COUNT_ARC_STC (PERF_COUNT_HW_MAX + 7) 67 68 #define PERF_COUNT_ARC_HW_MAX (PERF_COUNT_HW_MAX + 8) 69 70 /* 71 * Some ARC pct quirks: 72 * 73 * PERF_COUNT_HW_STALLED_CYCLES_BACKEND 74 * PERF_COUNT_HW_STALLED_CYCLES_FRONTEND 75 * The ARC 700 can either measure stalls per pipeline stage, or all stalls 76 * combined; for now we assign all stalls to STALLED_CYCLES_BACKEND 77 * and all pipeline flushes (e.g. caused by mispredicts, etc.) to 78 * STALLED_CYCLES_FRONTEND. 79 * 80 * We could start multiple performance counters and combine everything 81 * afterwards, but that makes it complicated. 82 * 83 * Note that I$ cache misses aren't counted by either of the two! 84 */ 85 86 /* 87 * ARC PCT has hardware conditions with fixed "names" but variable "indexes" 88 * (based on a specific RTL build) 89 * Below is the static map between perf generic/arc specific event_id and 90 * h/w condition names. 91 * At the time of probe, we loop thru each index and find it's name to 92 * complete the mapping of perf event_id to h/w index as latter is needed 93 * to program the counter really 94 */ 95 static const char * const arc_pmu_ev_hw_map[] = { 96 /* count cycles */ 97 [PERF_COUNT_HW_CPU_CYCLES] = "crun", 98 [PERF_COUNT_HW_REF_CPU_CYCLES] = "crun", 99 [PERF_COUNT_HW_BUS_CYCLES] = "crun", 100 101 [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = "bflush", 102 [PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = "bstall", 103 104 /* counts condition */ 105 [PERF_COUNT_HW_INSTRUCTIONS] = "iall", 106 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = "ijmp", /* Excludes ZOL jumps */ 107 [PERF_COUNT_ARC_BPOK] = "bpok", /* NP-NT, PT-T, PNT-NT */ 108 #ifdef CONFIG_ISA_ARCV2 109 [PERF_COUNT_HW_BRANCH_MISSES] = "bpmp", 110 #else 111 [PERF_COUNT_HW_BRANCH_MISSES] = "bpfail", /* NP-T, PT-NT, PNT-T */ 112 #endif 113 [PERF_COUNT_ARC_LDC] = "imemrdc", /* Instr: mem read cached */ 114 [PERF_COUNT_ARC_STC] = "imemwrc", /* Instr: mem write cached */ 115 116 [PERF_COUNT_ARC_DCLM] = "dclm", /* D-cache Load Miss */ 117 [PERF_COUNT_ARC_DCSM] = "dcsm", /* D-cache Store Miss */ 118 [PERF_COUNT_ARC_ICM] = "icm", /* I-cache Miss */ 119 [PERF_COUNT_ARC_EDTLB] = "edtlb", /* D-TLB Miss */ 120 [PERF_COUNT_ARC_EITLB] = "eitlb", /* I-TLB Miss */ 121 122 [PERF_COUNT_HW_CACHE_REFERENCES] = "imemrdc", /* Instr: mem read cached */ 123 [PERF_COUNT_HW_CACHE_MISSES] = "dclm", /* D-cache Load Miss */ 124 }; 125 126 #define C(_x) PERF_COUNT_HW_CACHE_##_x 127 #define CACHE_OP_UNSUPPORTED 0xffff 128 129 static const unsigned arc_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = { 130 [C(L1D)] = { 131 [C(OP_READ)] = { 132 [C(RESULT_ACCESS)] = PERF_COUNT_ARC_LDC, 133 [C(RESULT_MISS)] = PERF_COUNT_ARC_DCLM, 134 }, 135 [C(OP_WRITE)] = { 136 [C(RESULT_ACCESS)] = PERF_COUNT_ARC_STC, 137 [C(RESULT_MISS)] = PERF_COUNT_ARC_DCSM, 138 }, 139 [C(OP_PREFETCH)] = { 140 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 141 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 142 }, 143 }, 144 [C(L1I)] = { 145 [C(OP_READ)] = { 146 [C(RESULT_ACCESS)] = PERF_COUNT_HW_INSTRUCTIONS, 147 [C(RESULT_MISS)] = PERF_COUNT_ARC_ICM, 148 }, 149 [C(OP_WRITE)] = { 150 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 151 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 152 }, 153 [C(OP_PREFETCH)] = { 154 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 155 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 156 }, 157 }, 158 [C(LL)] = { 159 [C(OP_READ)] = { 160 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 161 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 162 }, 163 [C(OP_WRITE)] = { 164 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 165 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 166 }, 167 [C(OP_PREFETCH)] = { 168 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 169 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 170 }, 171 }, 172 [C(DTLB)] = { 173 [C(OP_READ)] = { 174 [C(RESULT_ACCESS)] = PERF_COUNT_ARC_LDC, 175 [C(RESULT_MISS)] = PERF_COUNT_ARC_EDTLB, 176 }, 177 /* DTLB LD/ST Miss not segregated by h/w*/ 178 [C(OP_WRITE)] = { 179 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 180 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 181 }, 182 [C(OP_PREFETCH)] = { 183 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 184 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 185 }, 186 }, 187 [C(ITLB)] = { 188 [C(OP_READ)] = { 189 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 190 [C(RESULT_MISS)] = PERF_COUNT_ARC_EITLB, 191 }, 192 [C(OP_WRITE)] = { 193 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 194 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 195 }, 196 [C(OP_PREFETCH)] = { 197 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 198 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 199 }, 200 }, 201 [C(BPU)] = { 202 [C(OP_READ)] = { 203 [C(RESULT_ACCESS)] = PERF_COUNT_HW_BRANCH_INSTRUCTIONS, 204 [C(RESULT_MISS)] = PERF_COUNT_HW_BRANCH_MISSES, 205 }, 206 [C(OP_WRITE)] = { 207 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 208 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 209 }, 210 [C(OP_PREFETCH)] = { 211 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 212 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 213 }, 214 }, 215 [C(NODE)] = { 216 [C(OP_READ)] = { 217 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 218 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 219 }, 220 [C(OP_WRITE)] = { 221 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 222 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 223 }, 224 [C(OP_PREFETCH)] = { 225 [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED, 226 [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED, 227 }, 228 }, 229 }; 230 231 #endif /* __ASM_PERF_EVENT_H */ 232