Add decoding for the new "lvlx" and "snoopx" meminfo fields added earlier to the kernel so that "perf mem report" and other tools can print it properly. v2: Merge with persistent memory patch. Switch to new bit encoding for each combination. v3: Switch to generic lvlnum field.
Unified · Split
| @@ -954,14 +954,20 @@ union perf_mem_data_src { · 953 unmodified lines | |||
| 954 | mem_snoop:5, /* snoop mode */ | 954 | mem_snoop:5, /* snoop mode */ |
| 955 | mem_lock:2, /* lock instr */ | 955 | mem_lock:2, /* lock instr */ |
| 956 | mem_dtlb:7, /* tlb access */ | 956 | mem_dtlb:7, /* tlb access */ |
| 957 | mem_rsvd:31; | 957 | mem_lvl_num:4, /* memory hierarchy level number */ |
| 958 | mem_remote:1, /* remote */ | ||
| 959 | mem_snoopx:2, /* snoop mode, ext */ | ||
| 960 | mem_rsvd:24; | ||
| 958 | }; | 961 | }; |
| 959 | }; | 962 | }; |
| 960 | #elif defined(__BIG_ENDIAN_BITFIELD) | 963 | #elif defined(__BIG_ENDIAN_BITFIELD) |
| 961 | union perf_mem_data_src { | 964 | union perf_mem_data_src { |
| 962 | __u64 val; | 965 | __u64 val; |
| 963 | struct { | 966 | struct { |
| 964 | __u64 mem_rsvd:31, | 967 | __u64 mem_rsvd:24, |
| 968 | mem_snoopx:2, /* snoop mode, ext */ | ||
| 969 | mem_remote:1, /* remote */ | ||
| 970 | mem_lvl_num:4, /* memory hierarchy level number */ | ||
| 965 | mem_dtlb:7, /* tlb access */ | 971 | mem_dtlb:7, /* tlb access */ |
| 966 | mem_lock:2, /* lock instr */ | 972 | mem_lock:2, /* lock instr */ |
| 967 | mem_snoop:5, /* snoop mode */ | 973 | mem_snoop:5, /* snoop mode */ |
| @@ -998,6 +1004,22 @@ union perf_mem_data_src { · 30 unmodified lines | |||
| 998 | #define PERF_MEM_LVL_UNC 0x2000 /* Uncached memory */ | 1004 | #define PERF_MEM_LVL_UNC 0x2000 /* Uncached memory */ |
| 999 | #define PERF_MEM_LVL_SHIFT 5 | 1005 | #define PERF_MEM_LVL_SHIFT 5 |
| 1000 | 1006 | ||
| 1007 | #define PERF_MEM_REMOTE_REMOTE 0x01 /* Remote */ | ||
| 1008 | #define PERF_MEM_REMOTE_SHIFT 37 | ||
| 1009 | |||
| 1010 | #define PERF_MEM_LVLNUM_L1 0x01 /* L1 */ | ||
| 1011 | #define PERF_MEM_LVLNUM_L2 0x02 /* L2 */ | ||
| 1012 | #define PERF_MEM_LVLNUM_L3 0x03 /* L3 */ | ||
| 1013 | #define PERF_MEM_LVLNUM_L4 0x04 /* L4 */ | ||
| 1014 | /* 5-0xa available */ | ||
| 1015 | #define PERF_MEM_LVLNUM_ANY_CACHE 0x0b /* Any cache */ | ||
| 1016 | #define PERF_MEM_LVLNUM_LFB 0x0c /* LFB */ | ||
| 1017 | #define PERF_MEM_LVLNUM_RAM 0x0d /* RAM */ | ||
| 1018 | #define PERF_MEM_LVLNUM_PMEM 0x0e /* PMEM */ | ||
| 1019 | #define PERF_MEM_LVLNUM_NA 0x0f /* N/A */ | ||
| 1020 | |||
| 1021 | #define PERF_MEM_LVLNUM_SHIFT 33 | ||
| 1022 | |||
| 1001 | /* snoop mode */ | 1023 | /* snoop mode */ |
| 1002 | #define PERF_MEM_SNOOP_NA 0x01 /* not available */ | 1024 | #define PERF_MEM_SNOOP_NA 0x01 /* not available */ |
| 1003 | #define PERF_MEM_SNOOP_NONE 0x02 /* no snoop */ | 1025 | #define PERF_MEM_SNOOP_NONE 0x02 /* no snoop */ |
| @@ -1006,6 +1028,10 @@ union perf_mem_data_src { · 2 unmodified lines | |||
| 1006 | #define PERF_MEM_SNOOP_HITM 0x10 /* snoop hit modified */ | 1028 | #define PERF_MEM_SNOOP_HITM 0x10 /* snoop hit modified */ |
| 1007 | #define PERF_MEM_SNOOP_SHIFT 19 | 1029 | #define PERF_MEM_SNOOP_SHIFT 19 |
| 1008 | 1030 | ||
| 1031 | #define PERF_MEM_SNOOPX_FWD 0x01 /* forward */ | ||
| 1032 | /* 1 free */ | ||
| 1033 | #define PERF_MEM_SNOOPX_SHIFT 37 | ||
| 1034 | |||
| 1009 | /* locked instruction */ | 1035 | /* locked instruction */ |
| 1010 | #define PERF_MEM_LOCK_NA 0x01 /* not available */ | 1036 | #define PERF_MEM_LOCK_NA 0x01 /* not available */ |
| 1011 | #define PERF_MEM_LOCK_LOCKED 0x02 /* locked transaction */ | 1037 | #define PERF_MEM_LOCK_LOCKED 0x02 /* locked transaction */ |
| @@ -166,11 +166,20 @@ static const char * const mem_lvl[] = { · 165 unmodified lines | |||
| 166 | "Uncached", | 166 | "Uncached", |
| 167 | }; | 167 | }; |
| 168 | 168 | ||
| 169 | static const char * const mem_lvlnum[] = { | ||
| 170 | [PERF_MEM_LVLNUM_ANY_CACHE] = "Any cache", | ||
| 171 | [PERF_MEM_LVLNUM_LFB] = "LFB", | ||
| 172 | [PERF_MEM_LVLNUM_RAM] = "RAM", | ||
| 173 | [PERF_MEM_LVLNUM_PMEM] = "PMEM", | ||
| 174 | [PERF_MEM_LVLNUM_NA] = "N/A", | ||
| 175 | }; | ||
| 176 | |||
| 169 | int perf_mem__lvl_scnprintf(char *out, size_t sz, struct mem_info *mem_info) | 177 | int perf_mem__lvl_scnprintf(char *out, size_t sz, struct mem_info *mem_info) |
| 170 | { | 178 | { |
| 171 | size_t i, l = 0; | 179 | size_t i, l = 0; |
| 172 | u64 m = PERF_MEM_LVL_NA; | 180 | u64 m = PERF_MEM_LVL_NA; |
| 173 | u64 hit, miss; | 181 | u64 hit, miss; |
| 182 | int printed; | ||
| 174 | 183 | ||
| 175 | if (mem_info) | 184 | if (mem_info) |
| 176 | m = mem_info->data_src.mem_lvl; | 185 | m = mem_info->data_src.mem_lvl; |
| @@ -184,17 +193,37 @@ int perf_mem__lvl_scnprintf(char *out, size_t sz, struct mem_info *mem_info) · 7 unmodified lines | |||
| 184 | /* already taken care of */ | 193 | /* already taken care of */ |
| 185 | m &= ~(PERF_MEM_LVL_HIT|PERF_MEM_LVL_MISS); | 194 | m &= ~(PERF_MEM_LVL_HIT|PERF_MEM_LVL_MISS); |
| 186 | 195 | ||
| 196 | |||
| 197 | if (mem_info && mem_info->data_src.mem_remote) { | ||
| 198 | strcat(out, "Remote "); | ||
| 199 | l += 7; | ||
| 200 | } | ||
| 201 | |||
| 202 | printed = 0; | ||
| 187 | for (i = 0; m && i < ARRAY_SIZE(mem_lvl); i++, m >>= 1) { | 203 | for (i = 0; m && i < ARRAY_SIZE(mem_lvl); i++, m >>= 1) { |
| 188 | if (!(m & 0x1)) | 204 | if (!(m & 0x1)) |
| 189 | continue; | 205 | continue; |
| 190 | if (l) { | 206 | if (printed++) { |
| 191 | strcat(out, " or "); | 207 | strcat(out, " or "); |
| 192 | l += 4; | 208 | l += 4; |
| 193 | } | 209 | } |
| 194 | l += scnprintf(out + l, sz - l, mem_lvl[i]); | 210 | l += scnprintf(out + l, sz - l, mem_lvl[i]); |
| 195 | } | 211 | } |
| 196 | if (*out == '\0') | 212 | |
| 197 | l += scnprintf(out, sz - l, "N/A"); | 213 | if (mem_info && mem_info->data_src.mem_lvl_num) { |
| 214 | int lvl = mem_info->data_src.mem_lvl_num; | ||
| 215 | if (printed++) { | ||
| 216 | strcat(out, " or "); | ||
| 217 | l += 4; | ||
| 218 | } | ||
| 219 | if (mem_lvlnum[lvl]) | ||
| 220 | l += scnprintf(out + l, sz - l, mem_lvlnum[lvl]); | ||
| 221 | else | ||
| 222 | l += scnprintf(out + l, sz - l, "L%d", lvl); | ||
| 223 | } | ||
| 224 | |||
| 225 | if (l == 0) | ||
| 226 | l += scnprintf(out + l, sz - l, "N/A"); | ||
| 198 | if (hit) | 227 | if (hit) |
| 199 | l += scnprintf(out + l, sz - l, " hit"); | 228 | l += scnprintf(out + l, sz - l, " hit"); |
| 200 | if (miss) | 229 | if (miss) |
| @@ -231,6 +260,14 @@ int perf_mem__snp_scnprintf(char *out, size_t sz, struct mem_info *mem_info) · 30 unmodified lines | |||
| 231 | } | 260 | } |
| 232 | l += scnprintf(out + l, sz - l, snoop_access[i]); | 261 | l += scnprintf(out + l, sz - l, snoop_access[i]); |
| 233 | } | 262 | } |
| 263 | if (mem_info && | ||
| 264 | (mem_info->data_src.mem_snoopx & PERF_MEM_SNOOPX_FWD)) { | ||
| 265 | if (l) { | ||
| 266 | strcat(out, " or "); | ||
| 267 | l += 4; | ||
| 268 | } | ||
| 269 | l += scnprintf(out + l, sz - l, "Fwd"); | ||
| 270 | } | ||
| 234 | 271 | ||
| 235 | if (*out == '\0') | 272 | if (*out == '\0') |
| 236 | l += scnprintf(out, sz - l, "N/A"); | 273 | l += scnprintf(out, sz - l, "N/A"); |