Madhavan Srinivasan | 8c002db | 2016-06-26 23:07:08 +0530 | [diff] [blame^] | 1 | /* |
| 2 | * Performance counter support for POWER9 processors. |
| 3 | * |
| 4 | * Copyright 2009 Paul Mackerras, IBM Corporation. |
| 5 | * Copyright 2013 Michael Ellerman, IBM Corporation. |
| 6 | * Copyright 2016 Madhavan Srinivasan, IBM Corporation. |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU General Public License |
| 10 | * as published by the Free Software Foundation; either version |
| 11 | * 2 of the License, or later version. |
| 12 | */ |
| 13 | |
| 14 | #define pr_fmt(fmt) "power9-pmu: " fmt |
| 15 | |
| 16 | #include "isa207-common.h" |
| 17 | |
| 18 | /* |
| 19 | * Some power9 event codes. |
| 20 | */ |
| 21 | #define EVENT(_name, _code) _name = _code, |
| 22 | |
| 23 | enum { |
| 24 | #include "power9-events-list.h" |
| 25 | }; |
| 26 | |
| 27 | #undef EVENT |
| 28 | |
| 29 | /* MMCRA IFM bits - POWER9 */ |
| 30 | #define POWER9_MMCRA_IFM1 0x0000000040000000UL |
| 31 | #define POWER9_MMCRA_IFM2 0x0000000080000000UL |
| 32 | #define POWER9_MMCRA_IFM3 0x00000000C0000000UL |
| 33 | |
| 34 | |
| 35 | PMU_FORMAT_ATTR(event, "config:0-49"); |
| 36 | PMU_FORMAT_ATTR(pmcxsel, "config:0-7"); |
| 37 | PMU_FORMAT_ATTR(mark, "config:8"); |
| 38 | PMU_FORMAT_ATTR(combine, "config:11"); |
| 39 | PMU_FORMAT_ATTR(unit, "config:12-15"); |
| 40 | PMU_FORMAT_ATTR(pmc, "config:16-19"); |
| 41 | PMU_FORMAT_ATTR(cache_sel, "config:20-23"); |
| 42 | PMU_FORMAT_ATTR(sample_mode, "config:24-28"); |
| 43 | PMU_FORMAT_ATTR(thresh_sel, "config:29-31"); |
| 44 | PMU_FORMAT_ATTR(thresh_stop, "config:32-35"); |
| 45 | PMU_FORMAT_ATTR(thresh_start, "config:36-39"); |
| 46 | PMU_FORMAT_ATTR(thresh_cmp, "config:40-49"); |
| 47 | |
| 48 | static struct attribute *power9_pmu_format_attr[] = { |
| 49 | &format_attr_event.attr, |
| 50 | &format_attr_pmcxsel.attr, |
| 51 | &format_attr_mark.attr, |
| 52 | &format_attr_combine.attr, |
| 53 | &format_attr_unit.attr, |
| 54 | &format_attr_pmc.attr, |
| 55 | &format_attr_cache_sel.attr, |
| 56 | &format_attr_sample_mode.attr, |
| 57 | &format_attr_thresh_sel.attr, |
| 58 | &format_attr_thresh_stop.attr, |
| 59 | &format_attr_thresh_start.attr, |
| 60 | &format_attr_thresh_cmp.attr, |
| 61 | NULL, |
| 62 | }; |
| 63 | |
| 64 | struct attribute_group power9_pmu_format_group = { |
| 65 | .name = "format", |
| 66 | .attrs = power9_pmu_format_attr, |
| 67 | }; |
| 68 | |
| 69 | static const struct attribute_group *power9_pmu_attr_groups[] = { |
| 70 | &power9_pmu_format_group, |
| 71 | NULL, |
| 72 | }; |
| 73 | |
| 74 | static int power9_generic_events[] = { |
| 75 | [PERF_COUNT_HW_CPU_CYCLES] = PM_CYC, |
| 76 | [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = PM_ICT_NOSLOT_CYC, |
| 77 | [PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = PM_CMPLU_STALL, |
| 78 | [PERF_COUNT_HW_INSTRUCTIONS] = PM_INST_CMPL, |
| 79 | [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = PM_BRU_CMPL, |
| 80 | [PERF_COUNT_HW_BRANCH_MISSES] = PM_BR_MPRED_CMPL, |
| 81 | [PERF_COUNT_HW_CACHE_REFERENCES] = PM_LD_REF_L1, |
| 82 | [PERF_COUNT_HW_CACHE_MISSES] = PM_LD_MISS_L1_FIN, |
| 83 | }; |
| 84 | |
| 85 | static u64 power9_bhrb_filter_map(u64 branch_sample_type) |
| 86 | { |
| 87 | u64 pmu_bhrb_filter = 0; |
| 88 | |
| 89 | /* BHRB and regular PMU events share the same privilege state |
| 90 | * filter configuration. BHRB is always recorded along with a |
| 91 | * regular PMU event. As the privilege state filter is handled |
| 92 | * in the basic PMC configuration of the accompanying regular |
| 93 | * PMU event, we ignore any separate BHRB specific request. |
| 94 | */ |
| 95 | |
| 96 | /* No branch filter requested */ |
| 97 | if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY) |
| 98 | return pmu_bhrb_filter; |
| 99 | |
| 100 | /* Invalid branch filter options - HW does not support */ |
| 101 | if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_RETURN) |
| 102 | return -1; |
| 103 | |
| 104 | if (branch_sample_type & PERF_SAMPLE_BRANCH_IND_CALL) |
| 105 | return -1; |
| 106 | |
| 107 | if (branch_sample_type & PERF_SAMPLE_BRANCH_CALL) |
| 108 | return -1; |
| 109 | |
| 110 | if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_CALL) { |
| 111 | pmu_bhrb_filter |= POWER9_MMCRA_IFM1; |
| 112 | return pmu_bhrb_filter; |
| 113 | } |
| 114 | |
| 115 | /* Every thing else is unsupported */ |
| 116 | return -1; |
| 117 | } |
| 118 | |
| 119 | static void power9_config_bhrb(u64 pmu_bhrb_filter) |
| 120 | { |
| 121 | /* Enable BHRB filter in PMU */ |
| 122 | mtspr(SPRN_MMCRA, (mfspr(SPRN_MMCRA) | pmu_bhrb_filter)); |
| 123 | } |
| 124 | |
| 125 | #define C(x) PERF_COUNT_HW_CACHE_##x |
| 126 | |
| 127 | /* |
| 128 | * Table of generalized cache-related events. |
| 129 | * 0 means not supported, -1 means nonsensical, other values |
| 130 | * are event codes. |
| 131 | */ |
| 132 | static int power9_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = { |
| 133 | [ C(L1D) ] = { |
| 134 | [ C(OP_READ) ] = { |
| 135 | [ C(RESULT_ACCESS) ] = PM_LD_REF_L1, |
| 136 | [ C(RESULT_MISS) ] = PM_LD_MISS_L1_FIN, |
| 137 | }, |
| 138 | [ C(OP_WRITE) ] = { |
| 139 | [ C(RESULT_ACCESS) ] = 0, |
| 140 | [ C(RESULT_MISS) ] = PM_ST_MISS_L1, |
| 141 | }, |
| 142 | [ C(OP_PREFETCH) ] = { |
| 143 | [ C(RESULT_ACCESS) ] = PM_L1_PREF, |
| 144 | [ C(RESULT_MISS) ] = 0, |
| 145 | }, |
| 146 | }, |
| 147 | [ C(L1I) ] = { |
| 148 | [ C(OP_READ) ] = { |
| 149 | [ C(RESULT_ACCESS) ] = PM_INST_FROM_L1, |
| 150 | [ C(RESULT_MISS) ] = PM_L1_ICACHE_MISS, |
| 151 | }, |
| 152 | [ C(OP_WRITE) ] = { |
| 153 | [ C(RESULT_ACCESS) ] = PM_L1_DEMAND_WRITE, |
| 154 | [ C(RESULT_MISS) ] = -1, |
| 155 | }, |
| 156 | [ C(OP_PREFETCH) ] = { |
| 157 | [ C(RESULT_ACCESS) ] = PM_IC_PREF_WRITE, |
| 158 | [ C(RESULT_MISS) ] = 0, |
| 159 | }, |
| 160 | }, |
| 161 | [ C(LL) ] = { |
| 162 | [ C(OP_READ) ] = { |
| 163 | [ C(RESULT_ACCESS) ] = PM_DATA_FROM_L3, |
| 164 | [ C(RESULT_MISS) ] = PM_DATA_FROM_L3MISS, |
| 165 | }, |
| 166 | [ C(OP_WRITE) ] = { |
| 167 | [ C(RESULT_ACCESS) ] = PM_L2_ST, |
| 168 | [ C(RESULT_MISS) ] = PM_L2_ST_MISS, |
| 169 | }, |
| 170 | [ C(OP_PREFETCH) ] = { |
| 171 | [ C(RESULT_ACCESS) ] = PM_L3_PREF_ALL, |
| 172 | [ C(RESULT_MISS) ] = 0, |
| 173 | }, |
| 174 | }, |
| 175 | [ C(DTLB) ] = { |
| 176 | [ C(OP_READ) ] = { |
| 177 | [ C(RESULT_ACCESS) ] = 0, |
| 178 | [ C(RESULT_MISS) ] = PM_DTLB_MISS, |
| 179 | }, |
| 180 | [ C(OP_WRITE) ] = { |
| 181 | [ C(RESULT_ACCESS) ] = -1, |
| 182 | [ C(RESULT_MISS) ] = -1, |
| 183 | }, |
| 184 | [ C(OP_PREFETCH) ] = { |
| 185 | [ C(RESULT_ACCESS) ] = -1, |
| 186 | [ C(RESULT_MISS) ] = -1, |
| 187 | }, |
| 188 | }, |
| 189 | [ C(ITLB) ] = { |
| 190 | [ C(OP_READ) ] = { |
| 191 | [ C(RESULT_ACCESS) ] = 0, |
| 192 | [ C(RESULT_MISS) ] = PM_ITLB_MISS, |
| 193 | }, |
| 194 | [ C(OP_WRITE) ] = { |
| 195 | [ C(RESULT_ACCESS) ] = -1, |
| 196 | [ C(RESULT_MISS) ] = -1, |
| 197 | }, |
| 198 | [ C(OP_PREFETCH) ] = { |
| 199 | [ C(RESULT_ACCESS) ] = -1, |
| 200 | [ C(RESULT_MISS) ] = -1, |
| 201 | }, |
| 202 | }, |
| 203 | [ C(BPU) ] = { |
| 204 | [ C(OP_READ) ] = { |
| 205 | [ C(RESULT_ACCESS) ] = PM_BRU_CMPL, |
| 206 | [ C(RESULT_MISS) ] = PM_BR_MPRED_CMPL, |
| 207 | }, |
| 208 | [ C(OP_WRITE) ] = { |
| 209 | [ C(RESULT_ACCESS) ] = -1, |
| 210 | [ C(RESULT_MISS) ] = -1, |
| 211 | }, |
| 212 | [ C(OP_PREFETCH) ] = { |
| 213 | [ C(RESULT_ACCESS) ] = -1, |
| 214 | [ C(RESULT_MISS) ] = -1, |
| 215 | }, |
| 216 | }, |
| 217 | [ C(NODE) ] = { |
| 218 | [ C(OP_READ) ] = { |
| 219 | [ C(RESULT_ACCESS) ] = -1, |
| 220 | [ C(RESULT_MISS) ] = -1, |
| 221 | }, |
| 222 | [ C(OP_WRITE) ] = { |
| 223 | [ C(RESULT_ACCESS) ] = -1, |
| 224 | [ C(RESULT_MISS) ] = -1, |
| 225 | }, |
| 226 | [ C(OP_PREFETCH) ] = { |
| 227 | [ C(RESULT_ACCESS) ] = -1, |
| 228 | [ C(RESULT_MISS) ] = -1, |
| 229 | }, |
| 230 | }, |
| 231 | }; |
| 232 | |
| 233 | #undef C |
| 234 | |
| 235 | static struct power_pmu power9_pmu = { |
| 236 | .name = "POWER9", |
| 237 | .n_counter = MAX_PMU_COUNTERS, |
| 238 | .add_fields = ISA207_ADD_FIELDS, |
| 239 | .test_adder = ISA207_TEST_ADDER, |
| 240 | .compute_mmcr = isa207_compute_mmcr, |
| 241 | .config_bhrb = power9_config_bhrb, |
| 242 | .bhrb_filter_map = power9_bhrb_filter_map, |
| 243 | .get_constraint = isa207_get_constraint, |
| 244 | .disable_pmc = isa207_disable_pmc, |
| 245 | .flags = PPMU_HAS_SIER | PPMU_ARCH_207S, |
| 246 | .n_generic = ARRAY_SIZE(power9_generic_events), |
| 247 | .generic_events = power9_generic_events, |
| 248 | .cache_events = &power9_cache_events, |
| 249 | .attr_groups = power9_pmu_attr_groups, |
| 250 | .bhrb_nr = 32, |
| 251 | }; |
| 252 | |
| 253 | static int __init init_power9_pmu(void) |
| 254 | { |
| 255 | int rc; |
| 256 | |
| 257 | /* Comes from cpu_specs[] */ |
| 258 | if (!cur_cpu_spec->oprofile_cpu_type || |
| 259 | strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power9")) |
| 260 | return -ENODEV; |
| 261 | |
| 262 | rc = register_power_pmu(&power9_pmu); |
| 263 | if (rc) |
| 264 | return rc; |
| 265 | |
| 266 | /* Tell userspace that EBB is supported */ |
| 267 | cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_EBB; |
| 268 | |
| 269 | return 0; |
| 270 | } |
| 271 | early_initcall(init_power9_pmu); |