linux-kernel - Building kernels under 5.10 takes twice as long as under 4.19

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <fdbc3b5e-9aeb-126a-1183-05326fd615f2@mailbox.org>
Date:   Thu, 18 Mar 2021 19:42:14 +0100
From:   Rainer Fiebig <jrf@...lbox.org>
To:     linux-kernel@...r.kernel.org
Subject: Building kernels under 5.10 takes twice as long as under 4.19

Hi!

I tried kernel 5.10.x and encountered a strange problem: building the
same kernel (5.10.22/23) took roughly twice as long as under 4.19.179.

This is reproducible under kernels 5.9, 5.8 and 5.6. The latter is where
it apparently starts, kernel 5.5 is not affected.

Here's an example:

5.6 built under 5.5, -j4
************************
[...]
  LD [M]  sound/pci/hda/snd-hda-intel.ko

real    11m44,125s
user    38m35,022s
sys     3m17,940s

5.6 under 5.6, -j4
******************
[...]
  LD [M]  sound/pci/hda/snd-hda-intel.ko

real    23m25,834s
user    26m7,738s
sys     2m42,292s

Always after "make mrproper", with -j4, identical .config and everything
else unchanged. It's also there with a default .config.

When I build the kernels under 4.19 or 5.5 with -j4, CPU-usage is 100 %
almost all the time. Under 5.6 and later, CPU-usage goes down to ~30 %
for longer periods during which I see long sections of sorted items like
"fs/xfs" which I had never noticed before.

CPUfreq-governor was "ondemand" in all cases (for 5.10 set via
commandline). The CPU is an Intel Core i3 (Clarkdale/Ironlake; still
fast enough for a snappy LFS-Linux with Plasma 5).

Before writing a bug-report: any ideas what might cause this or how to
narrow the search? A full bisect between 5.5 and 5.6 would be a real drag.

And please don't say this is a feature now. ;)

Thanks.

Rainer

-- 
The truth always turns out to be simpler than you thought.
Richard Feynman

Download attachment "signature.asc" of type "application/pgp-signature" (834 bytes)