Discussion:
Armeb removal before 12
Warner Losh
2018-06-13 16:55:00 UTC
Permalink
I'd like to remove armeb before 12.0.

It's poorly supported today.

Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.

Last time we asked, it took quite a while to find users.

It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.

The hardware was last made almost a decade ago.

It uses non-standard non-mainstream boot loaders (boot technology has moved
on from redboot).

The cost of doing API sweeps, make universe runs, etc exceeds the benefit
to the project.

So, given all these factors, it sounds like a good candidate for retirement.

Therefore, I'd like to remove it on July 15th.

Comments? (please keep them on topic to this specific thing: there's other
things that may also be past their freshness date, we'll discuss those in a
separate thread).

Warner
Rodney W. Grimes
2018-06-13 17:39:06 UTC
Permalink
Post by Warner Losh
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.
64MB is usable for 32 bit archs, I have many VM's running in
that configuration, so this item should not be in this list.
Post by Warner Losh
The hardware was last made almost a decade ago.
It uses non-standard non-mainstream boot loaders (boot technology has moved
on from redboot).
The cost of doing API sweeps, make universe runs, etc exceeds the benefit
to the project.
So, given all these factors, it sounds like a good candidate for retirement.
Therefore, I'd like to remove it on July 15th.
Comments? (please keep them on topic to this specific thing: there's other
things that may also be past their freshness date, we'll discuss those in a
separate thread).
Given all the other valid reasons, I have no objection to removal of armeb.
--
Rod Grimes ***@freebsd.org
Warner Losh
2018-06-13 18:09:08 UTC
Permalink
On Wed, Jun 13, 2018 at 11:39 AM, Rodney W. Grimes <
Post by Rodney W. Grimes
Post by Warner Losh
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.
64MB is usable for 32 bit archs, I have many VM's running in
that configuration, so this item should not be in this list.
Yes and no. A VM is a different beast to a real box. arm requires more
memory in general, and a little more on real machines.

You can run it in 128MB and do useful things, but not very many things. One
cannot, for example, run a wifi access point on arm in 128MB, at least on
this hardware. Adrian's ultra-stripped down stuff might be possible, but
nobody has ported it to this hardware despite it being ported to many weird
things. Likewise with the ZRouter project. You could run a simple sshd /
dns server on it, but there's lots of other alternatives for that which are
dirt cheap.

One can, with a lot of effort, do 64MB too, but it's more effort than for
i386. Even at 64MB on i386, though, the number of things you can do is
quite limited. You can't build anything on that machine. DNS + SSH is
possible here as well, as long as the zones are too big.

So I'll concede the point it's possible, and I'm not looking to make it not
be possible. However, we do need to draw the line somewhere, and this is
but one factor of many. Were it the only factor here, we wouldn't be
contemplating removal. Most people want to do more ambitious things that
can be done in 256MB is all I was trying to say.
Post by Rodney W. Grimes
The hardware was last made almost a decade ago.
Post by Warner Losh
It uses non-standard non-mainstream boot loaders (boot technology has
moved
Post by Warner Losh
on from redboot).
The cost of doing API sweeps, make universe runs, etc exceeds the benefit
to the project.
So, given all these factors, it sounds like a good candidate for
retirement.
Post by Warner Losh
Therefore, I'd like to remove it on July 15th.
Comments? (please keep them on topic to this specific thing: there's
other
Post by Warner Losh
things that may also be past their freshness date, we'll discuss those
in a
Post by Warner Losh
separate thread).
Given all the other valid reasons, I have no objection to removal of armeb.
Excellent.

Warner
Poul-Henning Kamp
2018-06-13 18:23:20 UTC
Permalink
--------
Post by Warner Losh
One can, with a lot of effort, do 64MB too, but it's more effort than for
i386. Even at 64MB on i386, though, the number of things you can do is
quite limited. You can't build anything on that machine.
Well, we *do* have code for something called "swap-devices", and
belive it or not, I actually ran make world in 64MB recently - just
to see if it would work. Took most of a week.

PS: No opinion on the proposed removal.
--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
***@FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
Steve O'Hara-Smith
2018-06-13 18:44:54 UTC
Permalink
On Wed, 13 Jun 2018 18:23:20 +0000
Post by Poul-Henning Kamp
Well, we *do* have code for something called "swap-devices", and
belive it or not, I actually ran make world in 64MB recently - just
to see if it would work. Took most of a week.
If memory serves correctly buildworld took a day or two back when
64MB was a fairly generous allocation of memory and there was a lot less
world.
--
Steve O'Hara-Smith <***@sohara.org>
Warner Losh
2018-06-13 19:13:06 UTC
Permalink
Post by Steve O'Hara-Smith
On Wed, 13 Jun 2018 18:23:20 +0000
Post by Poul-Henning Kamp
Well, we *do* have code for something called "swap-devices", and
belive it or not, I actually ran make world in 64MB recently - just
to see if it would work. Took most of a week.
If memory serves correctly buildworld took a day or two back when
64MB was a fairly generous allocation of memory and there was a lot less
world.
In the FreeBSD 1.x time frame, my 486DX2-66 with 32MB of RAM built world
over night (well, late afternoon start the build, it was done by breakfast
the next day so < 15 hours).

Warner
Poul-Henning Kamp
2018-06-13 19:39:08 UTC
Permalink
--------
Post by Steve O'Hara-Smith
On Wed, 13 Jun 2018 18:23:20 +0000
Post by Poul-Henning Kamp
Well, we *do* have code for something called "swap-devices", and
belive it or not, I actually ran make world in 64MB recently - just
to see if it would work. Took most of a week.
If memory serves correctly buildworld took a day or two back when
64MB was a fairly generous allocation of memory and there was a lot less
world.
Most of the cpu-time in buildworld is taken up compiling LLVM.
--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
***@FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
Warner Losh
2018-06-13 18:23:31 UTC
Permalink
Post by Warner Losh
On Wed, Jun 13, 2018 at 11:39 AM, Rodney W. Grimes <
Post by Rodney W. Grimes
Post by Warner Losh
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.
64MB is usable for 32 bit archs, I have many VM's running in
that configuration, so this item should not be in this list.
Yes and no. A VM is a different beast to a real box. arm requires more
memory in general, and a little more on real machines.
You can run it in 128MB and do useful things, but not very many things.
One cannot, for example, run a wifi access point on arm in 128MB, at least
on this hardware. Adrian's ultra-stripped down stuff might be possible, but
nobody has ported it to this hardware despite it being ported to many weird
things. Likewise with the ZRouter project. You could run a simple sshd /
dns server on it, but there's lots of other alternatives for that which are
dirt cheap.
One can, with a lot of effort, do 64MB too, but it's more effort than for
i386. Even at 64MB on i386, though, the number of things you can do is
quite limited. You can't build anything on that machine. DNS + SSH is
possible here as well, as long as the zones are too big.
So I'll concede the point it's possible, and I'm not looking to make it
not be possible. However, we do need to draw the line somewhere, and this
is but one factor of many. Were it the only factor here, we wouldn't be
contemplating removal. Most people want to do more ambitious things that
can be done in 256MB is all I was trying to say.
Just to followup...

On my 11.0 internal DHCP / DNS server (dnsmasq), I have a Allwinner A20
board with 1GB ram that runs between 100-120MB avm 120-1320MB wired with
~750MB free. That's what I've been basing my 256MB to run comfortably
statements on. It's possible to do it with less, but it takes a lot of
tuning, custom kernels, and legwork to make it happen. This box has only 14
processes, apart from kernel threads and is pretty minimal.

Warner
Mori Hiroki
2018-06-13 22:25:43 UTC
Permalink
Hi.


----- Original Message -----
Date: 2018/6/14, Thu 01:55
Subject: Armeb removal before 12
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.
sys/arm/ralink target is only 16M of RAM. I use bridge on this target
work fine.

I have FWIXP422BB target(Planex BRC-14VG) with 32M of RAM. I never build
FreeBSD for this target.

I want build for FWIXP422BB but clang cross build is very slow. It take
7 hour on my machine.
The hardware was last made almost a decade ago.
It uses non-standard non-mainstream boot loaders (boot technology has moved
on from redboot).
The cost of doing API sweeps, make universe runs, etc exceeds the benefit
to the project.
So, given all these factors, it sounds like a good candidate for retirement.
Therefore, I'd like to remove it on July 15th.
Comments? (please keep  them on topic to this specific thing: there's other
things that may also be past their freshness date, we'll discuss those in a
separate thread).
Warner
_______________________________________________
https://lists.freebsd.org/mailman/listinfo/freebsd-arm
Warner Losh
2018-06-13 22:57:02 UTC
Permalink
Post by Mori Hiroki
Hi.
----- Original Message -----
Date: 2018/6/14, Thu 01:55
Subject: Armeb removal before 12
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.
sys/arm/ralink target is only 16M of RAM. I use bridge on this target
work fine.
Even with current? It seems to have grown a lot. Even FreeBSD 11 on my
atmel boards, the 32MB ones can't really do much of anything (though a
custom /etc/rc.d might be usable). The 64MB ones I have troubles logging
into without disabling a ton of stuff. And FreeBSD-current is even larger.

Are you using a custom rc.d in this setup?
Post by Mori Hiroki
I have FWIXP422BB target(Planex BRC-14VG) with 32M of RAM. I never build
FreeBSD for this target.
I want build for FWIXP422BB but clang cross build is very slow. It take
7 hour on my machine.
I'm sorry to hear that. Even on my fast machines, it takes long time.

Warner
Post by Mori Hiroki
The hardware was last made almost a decade ago.
It uses non-standard non-mainstream boot loaders (boot technology has
moved
on from redboot).
The cost of doing API sweeps, make universe runs, etc exceeds the benefit
to the project.
So, given all these factors, it sounds like a good candidate for
retirement.
Therefore, I'd like to remove it on July 15th.
Comments? (please keep them on topic to this specific thing: there's
other
things that may also be past their freshness date, we'll discuss those
in a
separate thread).
Warner
_______________________________________________
https://lists.freebsd.org/mailman/listinfo/freebsd-arm
Mori Hiroki
2018-06-13 23:10:12 UTC
Permalink
Hi

----- Original Message -----
Date: 2018/6/14, Thu 07:57
Subject: Re: Armeb removal before 12
Hi.
Post by Mori Hiroki
----- Original Message -----
Date: 2018/6/14, Thu 01:55
Subject: Armeb removal before 12
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.
sys/arm/ralink target is only 16M of RAM. I use bridge on this target
work fine.
Even with current? It seems to have grown a lot. Even FreeBSD 11 on my atmel boards, the 32MB ones can't really do much of anything (though a custom /etc/rc.d might be usable). The 64MB ones I have troubles logging into without disabling a ton of stuff. And FreeBSD-current is even larger.
Are you using a custom rc.d in this setup?
Of course CURRENT.

https://gist.github.com/yamori813/88224f1c96c9c592fb611b12a15e4ab5


I use custom rc.d by ZRouter build system.

Thanks

Hiroki Mori
 
I have FWIXP422BB target(Planex BRC-14VG) with 32M of RAM. I never build
Post by Mori Hiroki
FreeBSD for this target.
I want build for FWIXP422BB but clang cross build is very slow. It take
7 hour on my machine.
I'm sorry to hear that. Even on my fast machines, it takes long time.
Warner
 
Post by Mori Hiroki
The hardware was last made almost a decade ago.
It uses non-standard non-mainstream boot loaders (boot technology has moved
on from redboot).
The cost of doing API sweeps, make universe runs, etc exceeds the benefit
to the project.
So, given all these factors, it sounds like a good candidate for retirement.
Therefore, I'd like to remove it on July 15th.
Comments? (please keep  them on topic to this specific thing: there's other
things that may also be past their freshness date, we'll discuss those in a
separate thread).
Warner
______________________________ _________________
https://lists.freebsd.org/ mailman/listinfo/freebsd-arm
Warner Losh
2018-06-13 23:24:13 UTC
Permalink
Post by Mori Hiroki
Hi
----- Original Message -----
Date: 2018/6/14, Thu 07:57
Subject: Re: Armeb removal before 12
Hi.
Post by Mori Hiroki
----- Original Message -----
Date: 2018/6/14, Thu 01:55
Subject: Armeb removal before 12
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD
to
Post by Mori Hiroki
run in.
sys/arm/ralink target is only 16M of RAM. I use bridge on this target
work fine.
Even with current? It seems to have grown a lot. Even FreeBSD 11 on my
atmel boards, the 32MB ones can't really do much of anything (though a
custom /etc/rc.d might be usable). The 64MB ones I have troubles logging
into without disabling a ton of stuff. And FreeBSD-current is even larger.
Are you using a custom rc.d in this setup?
Of course CURRENT.
https://gist.github.com/yamori813/88224f1c96c9c592fb611b12a15e4ab5
I use custom rc.d by ZRouter build system.
Cool! You have 10MB available. what all runs in that little space? I didn't
see too much starting up

Warner
Post by Mori Hiroki
Thanks
Hiroki Mori
I have FWIXP422BB target(Planex BRC-14VG) with 32M of RAM. I never build
Post by Mori Hiroki
FreeBSD for this target.
I want build for FWIXP422BB but clang cross build is very slow. It take
7 hour on my machine.
I'm sorry to hear that. Even on my fast machines, it takes long time.
Warner
Post by Mori Hiroki
The hardware was last made almost a decade ago.
It uses non-standard non-mainstream boot loaders (boot technology has
moved
Post by Mori Hiroki
on from redboot).
The cost of doing API sweeps, make universe runs, etc exceeds the
benefit
Post by Mori Hiroki
to the project.
So, given all these factors, it sounds like a good candidate for
retirement.
Post by Mori Hiroki
Therefore, I'd like to remove it on July 15th.
Comments? (please keep them on topic to this specific thing: there's
other
Post by Mori Hiroki
things that may also be past their freshness date, we'll discuss those
in a
Post by Mori Hiroki
separate thread).
Warner
______________________________ _________________
https://lists.freebsd.org/ mailman/listinfo/freebsd-arm
"
Gary Jennejohn
2018-06-14 11:41:01 UTC
Permalink
On Wed, 13 Jun 2018 16:57:02 -0600
Post by Warner Losh
Post by Mori Hiroki
Hi.
----- Original Message -----
Date: 2018/6/14, Thu 01:55
Subject: Armeb removal before 12
I'd like to remove armeb before 12.0.
It's poorly supported today.
Nobody has tested the concurrency kit changes on it. And ck is now
mandatory. We don't even know if it works or not.
Last time we asked, it took quite a while to find users.
It maxes out at 256MB of RAM. This is barely large enough for FreeBSD to
run in.
sys/arm/ralink target is only 16M of RAM. I use bridge on this target
work fine.
Even with current? It seems to have grown a lot. Even FreeBSD 11 on my
atmel boards, the 32MB ones can't really do much of anything (though a
custom /etc/rc.d might be usable). The 64MB ones I have troubles logging
into without disabling a ton of stuff. And FreeBSD-current is even larger.
As someone wrote in an earlier mail to one of the lists, it's
amazing how bloated UNIX has become in the last few decades.

In 1986 I was developing on a 68010-based SYS V machine for Morrow
Designs which had a whopping 1MB of RAM. This was considered to
be a very well endowed machine!

It was packed with all kinds of controller boards and I was doing
custom drivers for Morrow Designs.

It even had one of the very first Syquest removable-disk drives in
it, I think it was the 5MB model with MFM encoding.

Unfortunately, 1986 was also the year in which Morrow Designs
went tits up. I then owned the only UNIX-based machine which
Morrow Designs had ever produced. Eventually I donated it to
the VCF.
--
Gary Jennejohn
John-Mark Gurney
2018-07-05 16:21:17 UTC
Permalink
Post by Warner Losh
I'd like to remove armeb before 12.0.
As probably the only person still running armeb, I do not object to
it's removal... Heck, even my board is still running a much older
release...
Post by Warner Losh
It's poorly supported today.
And I probably won't be putting in any additional work, mainly as
those old boards are terribly slow...

I can make a board available if someone has a desire to hack on it..
--
John-Mark Gurney Voice: +1 415 225 5579

"All that I will do, has been done, All that I have, has not."
Warner Losh
2018-07-05 16:57:21 UTC
Permalink
Post by John-Mark Gurney
Post by Warner Losh
I'd like to remove armeb before 12.0.
As probably the only person still running armeb, I do not object to
it's removal... Heck, even my board is still running a much older
release...
Post by Warner Losh
It's poorly supported today.
And I probably won't be putting in any additional work, mainly as
those old boards are terribly slow...
I can make a board available if someone has a desire to hack on it..
I still have 2 of them. Newer boards are so much faster...

Warner
Loading...