Date: Thu, 13 Feb 2014 09:23:19 -0800 From: Marcel Moolenaar <marcel@xcllnt.net> To: Glen Barber <gjb@FreeBSD.org> Cc: freebsd-toolchain@FreeBSD.org Subject: Re: bmake(1) "sometimes" honors '-j' ? Message-ID: <2510126A-ECD7-4DC7-A936-23824B233821@xcllnt.net> In-Reply-To: <20140209213618.GA63417@glenbarber.us> References: <20140209213618.GA63417@glenbarber.us>
next in thread | previous in thread | raw e-mail | index | archive | help
[-- Attachment #1 --]
On Feb 9, 2014, at 1:36 PM, Glen Barber <gjb@FreeBSD.org> wrote:
> Maybe there is an obvious answer to this, or maybe I just do not
> understand how computers work.
>
> For snapshot builds for RPI-B and BEAGLEBONE, I am hard-coding '-j10'
> for buildworld and '-j6' for buildkernel, because these values are
> somewhat "safe" considering various race conditions with high '-j'
> values. Crochet sets the '-j' value to '1', if WORLDJOBS and
> KERNJOBS are not set, so what I end up with is basically:
> 'make [...] -j10 -j1 ${.TARGET}'.
>
> What I see in ps(1) is a bit confusing to me, though.
>
> For example, the following suggests '-j10' is being honored for the
> 'libraries' target, but '-j1' is being forced for the 'buildworld'
> target, and I'm not sure I understand why. (Lines broken up for
> readability.)
>
> root@grind:~ # ps auxww | grep make | grep 'j'
>
> root 67766 0.0 0.0 9024 1048 4 S+ 7:20PM 0:03.02 make
> TARGET_ARCH=armv6 SRCCONF=/dev/null __MAKE_CONF=/dev/null -j10 -j 1 buildworld
>
> root 13112 0.0 0.0 9024 1488 4 S 8:24PM 0:00.94 make
> -j10 -f Makefile.inc1 DESTDIR=/usr/obj/arm.armv6/usr/src/tmp -DNO_FSCHG
> -DWITHOUT_HTML -DWITHOUT_INFO -DNO_LINT -DWITHOUT_MAN -DNO_PROFILE
> -DNO_TESTS libraries
>
> So, if I am reading this correctly, the 'libraries' target is honoring
> the '-j10', and 'buildworld' is honoring the later-specified '-j1'.
>
> Am I reading this incorrectly, or is this something I just do not
> understand about our toolchain?
I can't give an authoritative answer, but there are a few
things to consider:
1. MAKEFLAGS is set by make(1) for later use. There's also the
.MAKEFLAGS variant. I don't know the *exact* difference
between them, but our makefile, including Makefile.inc1, uses
the variable and passes it to sub-makes.
2. make/bmake itself sets MAKEFLAGS/.MAKEFLAGS and automatically
passes it to sub-makes. So when a makefile sets it, it's to
override the default.
3. The .MAKEFLAGS variable holds the exact set of options passed
to make. For example:
% make -j10 -j1 -V .MAKEFLAGS
-j 10 -j 1 -V .MAKEFLAGS
What this means is that -j10 gets passed around and as such
can be used, not withstanding the fact that -j1 follows and
make itself may use that instead. Point 1 above applies.
For best results, pass only 1 -j flag or alternatively we need
to fix our makefiles to only keep the last -j in MAKEFLAGS, etc.
HTH,
--
Marcel Moolenaar
marcel@xcllnt.net
[-- Attachment #2 --]
-----BEGIN PGP SIGNATURE-----
Comment: GPGTools - http://gpgtools.org
iEYEARECAAYFAlL8/4cACgkQpgWlLWHuifYkAgCfQfo0GOUJw+V/EwFgtOlRgJkd
ytYAnROn9MWmCsL+3dNWghI5e4zG8eWo
=aqza
-----END PGP SIGNATURE-----
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?2510126A-ECD7-4DC7-A936-23824B233821>