Date: Wed, 1 Nov 2017 18:11:00 -0700 From: bob prohaska <fbsd@www.zefox.net> To: Mark Millard <markmi@dsl-only.net> Cc: Carl Johnson <carlj@peak.org>, freebsd-arm@freebsd.org, bob prohaska <fbsd@www.zefox.net> Subject: Re: RPi2 snapshot for armv7 won't boot. Message-ID: <20171102011100.GB66048@www.zefox.net> In-Reply-To: <497A9769-0BEF-4B2A-B0C8-0CA399F4DA29@dsl-only.net> References: <86fu9xmx3u.fsf@elm.localnet> <20171101231700.GA66048@www.zefox.net> <497A9769-0BEF-4B2A-B0C8-0CA399F4DA29@dsl-only.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Nov 01, 2017 at 05:27:52PM -0700, Mark Millard wrote:
>
> [Sometime between -r324743 and -r325156 armv7
> has been broken such that /sbin/init fails,
> apparently no matter how the build is done.
> I'm ignoring that issue here.]
>
I've never come close to getting far enough for init to
be an issue. Something has always stopped buildworld, or
at least interfered enough to stop installworld.
>
> Crochet has one major configuration problem for
> reliability when swapping is to be involved on
> the target: Crochet creates a context using a
> swapfile instead of using a swap partition.
>
At least on 11-stable the swap file was enough for
buildworld/kernel, though I don't know how thin the
margin is. That particular host doesn't get fiddled
with much, so long as it works.
> See bugzilla 206048 about the swapfile issue.
>
> An RPI2 self-hosted likely(?) swaps extensively at
> times. ("Only" 1 GiByte of RAM but clang is built,
> for example.) I expect that having swap space is a
> requirement to have self-hosting work. I'm not
> familiar with what a near-minimal size might be.
>
The box I'm trying to get from v6 to v7 uses a dedicated
swap partition on a usb-flash drive. Highest swap usage
I've noticed is a little under 1GB out of 2GB total.
> I'm not claiming this explains what you might be
> seeing. But the use of swapfiles is a known
> problem, apparently known long before that bugzilla
> submittal was made (2016-Jan-08).
>
The first problem encounted was fixed per you, simply
hardcoding the result of uname -p (=armv7) in Makefile.inc1.
Off and on, there were pmap faults (pt2map abort) which
superficially resemble what you reported about a year ago.
At first, they went away spontaneously, then returned (always
late in buildworld). A single-user buildworld succeeded,
but installworld reported a strip error in clang and clang
was never updated. Seemingly most else was updated. At this
point installworld is failing very quickly in /usr/src/rescue,
so I'm trying again to do a clean kernel, world, install cycle.
> My experience has lead me to always use swap
> partitions --and never swap files.
>
Agreed entirely.
>
> Another issue is significant swapping activity is
> likely not a good match to most sdcards. For this
> class of devices, I normally configure causing the
> root file system and swap space to be on a USB
> drive, such as a USB SSD. So mostly it is the
> kernel and earlier stages that come from the
> sdcard and the sdcard is otherwise little used
> in normal operation.
>
So far, I've gotten away with leaving / on the microSD card
and (most) everything else on a USB flash drive. I'm rather
curious to see if I can wear either one out. So far, no.
I'd be very encouraged to know if anyone at all is trying to self
host -current on RPI2, in any configuration.
Thanks for reading!
bob prohaska
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20171102011100.GB66048>