From a84fbf205609313594b86065c67e823f09ebe29b Mon Sep 17 00:00:00 2001 From: Ian Rogers Date: Tue, 5 Sep 2023 17:39:12 -0700 Subject: [PATCH] perf stat: Fix aggr mode initialization Generating metrics llc_code_read_mpi_demand_plus_prefetch, llc_data_read_mpi_demand_plus_prefetch, llc_miss_local_memory_bandwidth_read, llc_miss_local_memory_bandwidth_write, nllc_miss_remote_memory_bandwidth_read, memory_bandwidth_read, memory_bandwidth_write, uncore_frequency, upi_data_transmit_bw, C2_Pkg_Residency, C3_Core_Residency, C3_Pkg_Residency, C6_Core_Residency, C6_Pkg_Residency, C7_Core_Residency, C7_Pkg_Residency, UNCORE_FREQ and tma_info_system_socket_clks would trigger an address sanitizer heap-buffer-overflows on a SkylakeX. ``` ==2567752==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x5020003ed098 at pc 0x5621a816654e bp 0x7fffb55d4da0 sp 0x7fffb55d4d98 READ of size 4 at 0x5020003eee78 thread T0 #0 0x558265d6654d in aggr_cpu_id__is_empty tools/perf/util/cpumap.c:694:12 #1 0x558265c914da in perf_stat__get_aggr tools/perf/builtin-stat.c:1490:6 #2 0x558265c914da in perf_stat__get_global_cached tools/perf/builtin-stat.c:1530:9 #3 0x558265e53290 in should_skip_zero_counter tools/perf/util/stat-display.c:947:31 #4 0x558265e53290 in print_counter_aggrdata tools/perf/util/stat-display.c:985:18 #5 0x558265e51931 in print_counter tools/perf/util/stat-display.c:1110:3 #6 0x558265e51931 in evlist__print_counters tools/perf/util/stat-display.c:1571:5 #7 0x558265c8ec87 in print_counters tools/perf/builtin-stat.c:981:2 #8 0x558265c8cc71 in cmd_stat tools/perf/builtin-stat.c:2837:3 #9 0x558265bb9bd4 in run_builtin tools/perf/perf.c:323:11 #10 0x558265bb98eb in handle_internal_command tools/perf/perf.c:377:8 #11 0x558265bb9389 in run_argv tools/perf/perf.c:421:2 #12 0x558265bb9389 in main tools/perf/perf.c:537:3 ``` The issue was the use of testing a cpumap with NULL rather than using empty, as a map containing the dummy value isn't NULL and the -1 results in an empty aggr map being allocated which legitimately overflows when any member is accessed. Fixes: 8a96f454f5668572 ("perf stat: Avoid SEGV if core.cpus isn't set") Signed-off-by: Ian Rogers Acked-by: Namhyung Kim Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Ian Rogers Cc: Ingo Molnar Cc: Jiri Olsa Cc: Mark Rutland Cc: Peter Zijlstra Link: https://lore.kernel.org/r/20230906003912.3317462-1-irogers@google.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/builtin-stat.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 07b48f6df48eb..a3af805a1d572 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -1622,7 +1622,7 @@ static int perf_stat_init_aggr_mode(void) * taking the highest cpu number to be the size of * the aggregation translate cpumap. */ - if (evsel_list->core.user_requested_cpus) + if (!perf_cpu_map__empty(evsel_list->core.user_requested_cpus)) nr = perf_cpu_map__max(evsel_list->core.user_requested_cpus).cpu; else nr = 0;