The following commit has been merged into the perf/core branch of tip: Commit-ID: cae1d759065ee989de246d4a72bc2bfe9ad9d262 Gitweb: https://git.kernel.org/tip/cae1d759065ee989de246d4a72bc2bfe9ad9d262 Author: Kajol Jain <kjain@xxxxxxxxxxxxx> AuthorDate: Wed, 06 Oct 2021 19:36:53 +05:30 Committer: Peter Zijlstra <peterz@xxxxxxxxxxxxx> CommitterDate: Tue, 19 Oct 2021 17:27:00 +02:00 tools/perf: Add mem_hops field in perf_mem_data_src structure Going forward, future generation systems can have more hierarchy within the node/package level but currently we don't have any data source encoding field in perf, which can be used to represent this level of data. Add a new field called 'mem_hops' in the perf_mem_data_src structure which can be used to represent intra-node/package or inter-node/off-package details. This field is of size 3 bits where PERF_MEM_HOPS_{NA, 0..6} value can be used to present different hop levels data. Also add corresponding macros to define mem_hop field values and shift value. Currently we define macro for HOPS_0 which corresponds to data coming from another core but same node. Add functionality to represent mem_hop field data in perf_mem__lvl_scnprintf function with the help of added string array called mem_hops. For ex: Encodings for mem_hops fields with L2 cache: L2 - local L2 L2 | REMOTE | HOPS_0 - remote core, same node L2 Since with the addition of HOPS field, now remote can be used to denote cache access from the same node but different core, a check is added in the c2c_decode_stats function to set mrem only when HOPS is zero along with set remote field. Signed-off-by: Kajol Jain <kjain@xxxxxxxxxxxxx> Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx> Link: https://lore.kernel.org/r/20211006140654.298352-4-kjain@xxxxxxxxxxxxx --- tools/include/uapi/linux/perf_event.h | 11 +++++++++-- tools/perf/util/mem-events.c | 19 ++++++++++++++++++- 2 files changed, 27 insertions(+), 3 deletions(-) diff --git a/tools/include/uapi/linux/perf_event.h b/tools/include/uapi/linux/perf_event.h index e1701e9..2fc0957 100644 --- a/tools/include/uapi/linux/perf_event.h +++ b/tools/include/uapi/linux/perf_event.h @@ -1210,14 +1210,16 @@ union perf_mem_data_src { mem_remote:1, /* remote */ mem_snoopx:2, /* snoop mode, ext */ mem_blk:3, /* access blocked */ - mem_rsvd:21; + mem_hops:3, /* hop level */ + mem_rsvd:18; }; }; #elif defined(__BIG_ENDIAN_BITFIELD) union perf_mem_data_src { __u64 val; struct { - __u64 mem_rsvd:21, + __u64 mem_rsvd:18, + mem_hops:3, /* hop level */ mem_blk:3, /* access blocked */ mem_snoopx:2, /* snoop mode, ext */ mem_remote:1, /* remote */ @@ -1313,6 +1315,11 @@ union perf_mem_data_src { #define PERF_MEM_BLK_ADDR 0x04 /* address conflict */ #define PERF_MEM_BLK_SHIFT 40 +/* hop level */ +#define PERF_MEM_HOPS_0 0x01 /* remote core, same node */ +/* 2-7 available */ +#define PERF_MEM_HOPS_SHIFT 43 + #define PERF_MEM_S(a, s) \ (((__u64)PERF_MEM_##a##_##s) << PERF_MEM_##a##_SHIFT) diff --git a/tools/perf/util/mem-events.c b/tools/perf/util/mem-events.c index ff7289e..3167b46 100644 --- a/tools/perf/util/mem-events.c +++ b/tools/perf/util/mem-events.c @@ -301,6 +301,16 @@ static const char * const mem_lvlnum[] = { [PERF_MEM_LVLNUM_NA] = "N/A", }; +static const char * const mem_hops[] = { + "N/A", + /* + * While printing, 'Remote' will be added to represent + * 'Remote core, same node' accesses as remote field need + * to be set with mem_hops field. + */ + "core, same node", +}; + int perf_mem__lvl_scnprintf(char *out, size_t sz, struct mem_info *mem_info) { size_t i, l = 0; @@ -325,6 +335,9 @@ int perf_mem__lvl_scnprintf(char *out, size_t sz, struct mem_info *mem_info) l += 7; } + if (mem_info && mem_info->data_src.mem_hops) + l += scnprintf(out + l, sz - l, "%s ", mem_hops[mem_info->data_src.mem_hops]); + printed = 0; for (i = 0; m && i < ARRAY_SIZE(mem_lvl); i++, m >>= 1) { if (!(m & 0x1)) @@ -471,8 +484,12 @@ int c2c_decode_stats(struct c2c_stats *stats, struct mem_info *mi) /* * Skylake might report unknown remote level via this * bit, consider it when evaluating remote HITMs. + * + * Incase of power, remote field can also be used to denote cache + * accesses from the another core of same node. Hence, setting + * mrem only when HOPS is zero along with set remote field. */ - bool mrem = data_src->mem_remote; + bool mrem = (data_src->mem_remote && !data_src->mem_hops); int err = 0; #define HITM_INC(__f) \
![]() |