Re: [PATCH dwarves] btf_encoder: use ELF_C_READ_MMAP

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 







On Monday, October 14th, 2024 at 3:31 AM, Alan Maguire <alan.maguire@xxxxxxxxxx> wrote:

> 
> 
> This reduces peak memory utilization during BTF vmlinux encoding,
> replacing ELF section-sized malloc()s with mmap()s.
> 
> For -j1 peak memory resident set size drops from
> 
> Maximum resident set size (kbytes): 1069220
> 
> ...to:
> 
> Maximum resident set size (kbytes): 701888
> 
> ...with this patch.
> 
> For -j8, peak memory resident set size drops from
> 
> Maximum resident set size (kbytes): 1121172
> 
> ...to:
> 
> Maximum resident set size (kbytes): 963516
> 
> Signed-off-by: Alan Maguire alan.maguire@xxxxxxxxxx
> 
> ---
> btf_encoder.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/btf_encoder.c b/btf_encoder.c
> index 5954238..9599faf 100644
> --- a/btf_encoder.c
> +++ b/btf_encoder.c
> @@ -1879,7 +1879,7 @@ static int btf_encoder__tag_kfuncs(struct btf_encoder *encoder)
> goto out;
> }
> 
> - elf = elf_begin(fd, ELF_C_READ, NULL);
> + elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
> if (elf == NULL) {
> elf_error("Cannot update ELF file");
> goto out;
> --
> 2.43.5

Alan, I can confirm this reduces max resident set size when BTF
encoding vmlinux. Great improvement from a single flag change! See
below.

    for j in 1 2 4 8 16 32 64; do
        echo -n '-j$j: '
        /usr/bin/time -v \
        ./pahole -J -j$j \
             --btf_features=encode_force,var,float,enum64,decl_tag,type_tag,optimized_func,consistent_func,decl_tag_kfuncs \
             --btf_encode_detached=/dev/null \
             --lang_exclude=rust \
             ~/repo/bpf-dev-docker/linux/.tmp_vmlinux1 \
        2>&1 | grep 'Maximum resident set size'
    done

Current next (91bcd1d):

    -j1: 	Maximum resident set size (kbytes): 1600036
    -j2: 	Maximum resident set size (kbytes): 1623380
    -j4: 	Maximum resident set size (kbytes): 1635824
    -j8: 	Maximum resident set size (kbytes): 1631412
    -j16: 	Maximum resident set size (kbytes): 1711976
    -j32: 	Maximum resident set size (kbytes): 1799856
    -j64: 	Maximum resident set size (kbytes): 1976184

This patch on top of next (91bcd1d):

    -j1: 	Maximum resident set size (kbytes): 640344
    -j2: 	Maximum resident set size (kbytes): 695272
    -j4: 	Maximum resident set size (kbytes): 769772
    -j8: 	Maximum resident set size (kbytes): 895728
    -j16: 	Maximum resident set size (kbytes): 984908
    -j32: 	Maximum resident set size (kbytes): 1148420
    -j64: 	Maximum resident set size (kbytes): 1573944

This patch on top of elf_functions patch (with segfault fixes) [1]:

    -j1: 	Maximum resident set size (kbytes): 642992
    -j2: 	Maximum resident set size (kbytes): 695432
    -j4: 	Maximum resident set size (kbytes): 746888
    -j8: 	Maximum resident set size (kbytes): 851208
    -j16: 	Maximum resident set size (kbytes): 884232
    -j32: 	Maximum resident set size (kbytes): 930632
    -j64: 	Maximum resident set size (kbytes): 1127976

Tested-by: Ihor Solodrai <ihor.solodrai@xxxxx>

[1]: https://lore.kernel.org/dwarves/20241009233520.989453-1-ihor.solodrai@xxxxx/





[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux