Re: [PATCH] perf: Increase MAX_NR_CPUS to 4096
From: Ian Rogers
Date: Thu Dec 05 2024 - 12:36:57 EST
On Thu, Dec 5, 2024 at 9:01 AM Kyle Meyer <kyle.meyer@xxxxxxx> wrote:
>
> Systems have surpassed 2048 CPUs. Increase MAX_NR_CPUS to 4096.
>
> Bitmaps declared with MAX_NR_CPUS bits will increase from 256B to 512B,
> and cpus_runtime will increase from 81960B to 163880B.
>
> Signed-off-by: Kyle Meyer <kyle.meyer@xxxxxxx>
This is very interesting, thanks Kyle! Just noting, having the same
#define in many places is clearly error prone and there are other
redefinitions of this value:
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/bpf_skel/kwork_top.bpf.c?h=perf-tools-next#n21
I wonder I can refactor `tools/lib/perf/cpumap.c` to get rid of that
constant as it is only used when parsing from a file/string.
Could the kwork developers perhaps look at their many uses? The other
uses in the tool may be removable too.
Wrt this change, perhaps bump
`tools/perf/util/bpf_skel/kwork_top.bpf.c` too and then we merge that
while trying to remove other uses.
Thanks,
Ian
> ---
>
> Tested on a 32 socket Sapphire Rapids system with 3840 CPUs.
>
> tools/lib/perf/include/internal/cpumap.h | 2 +-
> tools/perf/perf.h | 2 +-
> 2 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/tools/lib/perf/include/internal/cpumap.h b/tools/lib/perf/include/internal/cpumap.h
> index 49649eb51ce4..3cf28522004e 100644
> --- a/tools/lib/perf/include/internal/cpumap.h
> +++ b/tools/lib/perf/include/internal/cpumap.h
> @@ -22,7 +22,7 @@ DECLARE_RC_STRUCT(perf_cpu_map) {
> };
>
> #ifndef MAX_NR_CPUS
> -#define MAX_NR_CPUS 2048
> +#define MAX_NR_CPUS 4096
> #endif
>
> struct perf_cpu_map *perf_cpu_map__alloc(int nr_cpus);
> diff --git a/tools/perf/perf.h b/tools/perf/perf.h
> index c004dd4e65a3..3cb40965549f 100644
> --- a/tools/perf/perf.h
> +++ b/tools/perf/perf.h
> @@ -3,7 +3,7 @@
> #define _PERF_PERF_H
>
> #ifndef MAX_NR_CPUS
> -#define MAX_NR_CPUS 2048
> +#define MAX_NR_CPUS 4096
> #endif
>
> enum perf_affinity {
> --
> 2.47.1
>