[Bug libc/34394] New: Regression in cpu2006 403.gcc after removal of dynamic mmap/trim threshold

[anton at ozlabs dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=34394

            Bug ID: 34394
           Summary: Regression in cpu2006 403.gcc after removal of dynamic
                    mmap/trim threshold
           Product: glibc
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libc
          Assignee: unassigned at sourceware dot org
          Reporter: anton at ozlabs dot org
                CC: drepper.fsp at gmail dot com
  Target Milestone: ---

We noticed cpu2006 403.gcc performance dropped significantly on a RISC-V
Spacemit-K3 system after a glibc update. The regression varies, but the worst
case is s04.i at 35% slower. It was isolated to the following patch:

17a79a51208c5648fe70983085833bf15d83d0f1
Author: Wilco Dijkstra <Wilco.Dijkstra@arm.com>
Date:   Thu Apr 2 13:56:10 2026 +0000

    malloc: Remove dynamic mmap/trim threshold [BZ #30769]

    v2: Update documentation

    Whenever a large mmap is released the mmap and trim thresholds are updated.
    As a result these thresholds grow ever larger which means huge allocations
    are always served by arenas rather than mmap.  The thresholds can end up as
    large as an arena, which completely stops all trimming of the top block.
    Remove the code completely - the default thresholds seem way too low for
    modern 64-bit targets, but they can be increased seperately.

    Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>

Not surprisingly the regression disappears when the threshold is increased:

GLIBC_TUNABLES=glibc.malloc.mmap_threshold=4194304

We should bump the threshold as suggested in the commit message.

-- 
You are receiving this mail because:
You are on the CC list for the bug.