[<prev] [next>] [day] [month] [year] [list]
Message-ID: <20210914012612.vwlowt5wsojmyfzr@shells.gnugeneration.com>
Date: Mon, 13 Sep 2021 18:26:12 -0700
From: Vito Caputo <vcaputo@...garu.com>
To: linux-kernel <linux-kernel@...r.kernel.org>
Subject: Broken WCHAN in modern x86 kernel defconfig? Intentional?
Hello lkml,
Recently I've been using an Arch machine with the distro-provided
kernel, and noticed the wchan is never set for any processes. This
seems rather odd for a generic distro-provided kernel, since it breaks
basic functionality like showing WCHAN in `top` or `ps -o wchan`.
Looking at /proc/config.gz I see only CONFIG_SCHED_OMIT_FRAME_POINTER=y
set WRT frame pointers, but I thought this meant I'd still see
non-zero wchans in /proc/*/wchan. It's not like it contains
CONFIG_FRAME_POINTER=n, it simply doesn't mention it at all.
For many years I've been running custom kernels on Debian built from
the same .config that's just kept evolving over time. So I wouldn't
have noticed if at some point CONFIG_FRAME_POINTER vanished from a
fresh x86_64 defconfig, and I doubt I originally explicitly turned
that on when seeding my .config back in the day.
To see if this indeed is a defconfig regression of sorts, I just tried
a `make defconfig; grep FRAME_POINTER .config` for both v4.0 and
master:
v4.0:
CONFIG_SCHED_OMIT_FRAME_POINTER=y
CONFIG_ARCH_WANT_FRAME_POINTERS=y
CONFIG_FRAME_POINTER=y
master:
CONFIG_SCHED_OMIT_FRAME_POINTER=y
# CONFIG_UNWINDER_FRAME_POINTER is not set
Is this intentional? At a glance, commit 81d3871 looks suspect in
removing ARCH_WANT_FRAME_POINTERS from x86/Kconfig.
Thanks,
Vito Caputo
Powered by blists - more mailing lists