NetBSD on Sun's Motorola 68020 (sun3) and 68030 (sun3x) based machines (NetBSD/sun3)

John Carr | 21 Jan 2009 19:46

Port help wanted/needed?

Hey, all...                                                                     

Since I've been running NetBSD/sun3 for 5+ years now, I'd like to throw         
my hat in the ring to help out the port where I can. While I know of the        
list at http://www.netbsd.org/contrib, I'd like to get an idea where I          
can help out on the port that will be the most effective based on what          
other folks are doing.                                                          

While I can't code (sadly, never had time to learn), I have an admin            
and security background and can, if nothing else, run -current, run 5.0,        
document, etc. I have a 3/60 and access to a 3/50 and can, of course,           
run tme somewhere, if needed.                                                   

If there's a specific area I can help out, please let me know. Thanks!

headers

Izumi Tsutsui | 23 Jan 2009 11:38

Re: Port help wanted/needed?

John Carr wrote:

> Since I've been running NetBSD/sun3 for 5+ years now, I'd like to throw
> my hat in the ring to help out the port where I can. While I know of the
> list at http://www.netbsd.org/contrib, I'd like to get an idea where I
> can help out on the port that will be the most effective based on what
> other folks are doing.
> 
> While I can't code (sadly, never had time to learn), I have an admin
> and security background and can, if nothing else, run -current, run 5.0,
> document, etc. I have a 3/60 and access to a 3/50 and can, of course,
> run tme somewhere, if needed.

It's appreciated to simply try 5.0_BETA GENERIC kernel on your 3/60.
ftp://ftp.NetBSD.org/pub/NetBSD-daily/netbsd-5/200901210002Z/sun3/

(it will work even on 4.0.1 userland, without full upgrade)

After 4.0, I added bus_dma(9) support to sun3 and made si(4) SCSI
use bus_dma(9) API, but it was only tested on si(4) at at vme
on TME emulating 3/1x0, and AFAIK there is no report about
si(4) at obio on 3/60.
http://mail-index.NetBSD.org/port-sun3/2007/02/oindex.html

(no one has noticed that even 4.0.1 GENERIC didn't boot on 3/60
 until you reported it...)
---
Izumi Tsutsui

(Continue reading)

headers

John Carr | 26 Jan 2009 16:29

Re: Port help wanted/needed?

On Fri, Jan 23, 2009 at 07:38:20PM +0900, Izumi Tsutsui wrote:
> It's appreciated to simply try 5.0_BETA GENERIC kernel on your 3/60.
> ftp://ftp.NetBSD.org/pub/NetBSD-daily/netbsd-5/200901210002Z/sun3/
> (it will work even on 4.0.1 userland, without full upgrade)
> After 4.0, I added bus_dma(9) support to sun3 and made si(4) SCSI
> use bus_dma(9) API, but it was only tested on si(4) at at vme
> on TME emulating 3/1x0, and AFAIK there is no report about
> si(4) at obio on 3/60.
> http://mail-index.NetBSD.org/port-sun3/2007/02/oindex.html
> (no one has noticed that even 4.0.1 GENERIC didn't boot on 3/60
>  until you reported it...)

The new kernel boots just fine on the 3/60 and works without issue. I'm
currently running a few bonnie++s in tandem to put it under load to see
if the si(4) to bus_dma(9) is stable on the 3/60. So far it's looking
good!

headers

Izumi Tsutsui | 29 Jan 2009 14:46

Re: Port help wanted/needed?

John Carr wrote:

> The new kernel boots just fine on the 3/60 and works without issue. I'm
> currently running a few bonnie++s in tandem to put it under load to see
> if the si(4) to bus_dma(9) is stable on the 3/60. So far it's looking
> good!

Thanks.

The major changes on sun3 port after 4.0 is the bus_dma(9) and
device_t/softc (internal structures in device drivers) split,
but it should also be okay since GENERIC just works.

(it means no other port specific changes since 4.0, though 
---
Izumi Tsutsui

headers

John Carr | 23 Jan 2009 19:21

Re: Port help wanted/needed?

On Fri, Jan 23, 2009 at 07:38:20PM +0900, Izumi Tsutsui wrote:
> It's appreciated to simply try 5.0_BETA GENERIC kernel on your 3/60.
> ftp://ftp.NetBSD.org/pub/NetBSD-daily/netbsd-5/200901210002Z/sun3/
> (it will work even on 4.0.1 userland, without full upgrade)
> After 4.0, I added bus_dma(9) support to sun3 and made si(4) SCSI
> use bus_dma(9) API, but it was only tested on si(4) at at vme
> on TME emulating 3/1x0, and AFAIK there is no report about
> si(4) at obio on 3/60.
> http://mail-index.NetBSD.org/port-sun3/2007/02/oindex.html

Ok. Will do. I'll start on this today then and begin testing it. I'll
let you know if if it even boots. :)

Work working with -current or _BETA, where's the best place to mention
issues? Here or gnats? I'm not sure how much gnats is used for bug
reports against non-releases.

> (no one has noticed that even 4.0.1 GENERIC didn't boot on 3/60
>  until you reported it...)

Very true. I'll give it a thorough testing.

What's the latest on 3/50 support, by the way? I remember a thread from
about '05 or so that mentioned that either miniroot or the INSTALL kernel 
was too big now to boot. Has anyone even attempted it?

(and yeah, I'm guessing the amount of folks out there trying to use
NetBSD on a 3/50 is probably down to about 1 at the most, but I'm
curious)
to even boot

(Continue reading)

headers

David Brownlee | 23 Jan 2009 23:46

Re: Port help wanted/needed?

On Fri, 23 Jan 2009, John Carr wrote:

> What's the latest on 3/50 support, by the way? I remember a thread from
> about '05 or so that mentioned that either miniroot or the INSTALL kernel
> was too big now to boot. Has anyone even attempted it?
>
> (and yeah, I'm guessing the amount of folks out there trying to use
> NetBSD on a 3/50 is probably down to about 1 at the most, but I'm
> curious)
> to even boot

 	The 3/50 has IIRC its framebuffer at 1MB, so the kernel size is
 	very constrained. Could you try building a kernel with just
 	about everything disabled for the 3/50?

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

headers

John Carr | 9 Feb 2009 20:56

Re: Port help wanted/needed?

On Fri, Jan 23, 2009 at 10:46:04PM +0000, David Brownlee wrote:
>> What's the latest on 3/50 support, by the way? I remember a thread from
>> about '05 or so that mentioned that either miniroot or the INSTALL kernel
>> was too big now to boot. Has anyone even attempted it?
>> (and yeah, I'm guessing the amount of folks out there trying to use
>> NetBSD on a 3/50 is probably down to about 1 at the most, but I'm
>> curious)
>> to even boot
> 	The 3/50 has IIRC its framebuffer at 1MB, so the kernel size is
> 	very constrained. Could you try building a kernel with just
> 	about everything disabled for the 3/50?

By the way, if there's a need, I can get a 3/50 up with serial console
on a routable IP w/a tftp server. I can do some testing with it, or if
anyone with more coding skills wants access to it, I'm offering it up.

Meanwhile, I'll probably have a little time to work with it this week to
see if we can get a kernel booting on it.

headers

David Brownlee | 9 Feb 2009 21:46

Re: Port help wanted/needed?

On Mon, 9 Feb 2009, John Carr wrote:

> By the way, if there's a need, I can get a 3/50 up with serial console
> on a routable IP w/a tftp server. I can do some testing with it, or if
> anyone with more coding skills wants access to it, I'm offering it up.
>
> Meanwhile, I'll probably have a little time to work with it this week to
> see if we can get a kernel booting on it.

 	I'd be very interested to see if the cut down SUN3_50 kernel
 	boots on it... :)

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

headers

David Johnson | 10 Feb 2009 03:57

Re: Port help wanted/needed?

Same here..  If it works I'd like to know what to trim out.  My 3/50M 
has only 4MB ram.

David Brownlee wrote:
> On Mon, 9 Feb 2009, John Carr wrote:
>
>> By the way, if there's a need, I can get a 3/50 up with serial console
>> on a routable IP w/a tftp server. I can do some testing with it, or if
>> anyone with more coding skills wants access to it, I'm offering it up.
>>
>> Meanwhile, I'll probably have a little time to work with it this week to
>> see if we can get a kernel booting on it.
>
>     I'd be very interested to see if the cut down SUN3_50 kernel
>     boots on it... :)
>

Attachment (electronsciences.vcf): text/x-vcard, 184 bytes

headers

David Brownlee | 10 Feb 2009 15:32

Re: Port help wanted/needed?

From: David Johnson <electronsciences <at> bellsouth.net>

> Same here..  If it works I'd like to know what to trim out.  My 3/50M 
> has only 4MB ram.

Are you in a position to test boot a kernel? I can easily make one 
available...

headers

John Carr | 24 Jan 2009 00:29

Re: Port help wanted/needed?

On Fri, Jan 23, 2009 at 10:46:04PM +0000, David Brownlee wrote:
>> What's the latest on 3/50 support, by the way? I remember a thread from
>> about '05 or so that mentioned that either miniroot or the INSTALL kernel
>> was too big now to boot. Has anyone even attempted it?
> 	The 3/50 has IIRC its framebuffer at 1MB, so the kernel size is
> 	very constrained. Could you try building a kernel with just
> 	about everything disabled for the 3/50?

Sure thing. I lent it away a few years ago, but I can get it back over
the weekend to give it a try. This one is a stock 3/50 minus the extra 8Mb
board. We'll see what we can do with only 3Mb to work with. I remember
the last time I did get a kernel booting on it (GENERIC, no less) a few 
years ago, it was swapping by the time it hit the login prompt.

headers

David Brownlee | 24 Jan 2009 00:37

Re: Port help wanted/needed?

On Fri, 23 Jan 2009, John Carr wrote:

> On Fri, Jan 23, 2009 at 10:46:04PM +0000, David Brownlee wrote:
>>> What's the latest on 3/50 support, by the way? I remember a thread from
>>> about '05 or so that mentioned that either miniroot or the INSTALL kernel
>>> was too big now to boot. Has anyone even attempted it?
>> 	The 3/50 has IIRC its framebuffer at 1MB, so the kernel size is
>> 	very constrained. Could you try building a kernel with just
>> 	about everything disabled for the 3/50?
>
> Sure thing. I lent it away a few years ago, but I can get it back over
> the weekend to give it a try. This one is a stock 3/50 minus the extra 8Mb
> board. We'll see what we can do with only 3Mb to work with. I remember
> the last time I did get a kernel booting on it (GENERIC, no less) a few
> years ago, it was swapping by the time it hit the login prompt.

 	I think its worth adding a SUN350 kernel config to the tree.
 	I'm curious to see how much free memory we can get :)

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

headers

Izumi Tsutsui | 24 Jan 2009 02:51

Re: Port help wanted/needed?

abs <at> NetBSD.org wrote:

> >> 	The 3/50 has IIRC its framebuffer at 1MB, so the kernel size is
> >> 	very constrained. Could you try building a kernel with just
> >> 	about everything disabled for the 3/50?
> >
> > Sure thing. I lent it away a few years ago, but I can get it back over
> > the weekend to give it a try. This one is a stock 3/50 minus the extra 8Mb
> > board. We'll see what we can do with only 3Mb to work with. I remember
> > the last time I did get a kernel booting on it (GENERIC, no less) a few
> > years ago, it was swapping by the time it hit the login prompt.
> 
>  	I think its worth adding a SUN350 kernel config to the tree.
>  	I'm curious to see how much free memory we can get :)

The problem is that sun3 ports requires three kernels for installation:
RAMDISK*: kernel with minimum binaries to write miniroot in md_root
INSTALL*: kernel in miniroot for installation
GENERIC*: GENERIC kernel

so we should consider if we will provide all these variants for 3/50.
(I guess single 3_50 would be enough by assuming network boot install)

I've tried to build a shrinked kernel for 3/50 and
it's still ~1.5MB, but NetBSD/sun2 works with 1.7MB
FOURMEG kernel with 4MB RAM, so it will also work on 4MB 3/50.

http://www.ceres.dti.ne.jp/~tsutsui/netbsd/netbsd-sun3-3_50-5.0_BETA-20090124.gz

(IIRC NetBSD 1.1 kernel was ~700KB and OS is ever growing

(Continue reading)

headers

David Brownlee | 25 Jan 2009 01:29

Re: Small kernel for sun3

 	I've been playing around with Izumi's kernel config, looking
 	where to save space, and came up with an interesting issue
 	- specifically how to determine kernel size.

 	For example the two kernels below are very similar in size
 	as reported by ls (the second is 673 bytes larger), but
 	when using 'size' the second is 130620 bytes (just under
 	128K) smaller.

ls -l: 1360 -rwxr-xr-x  1 abs  wheel  1371776 Jan 25 00:07 netbsd
  size: 1055608   26916  207584 1290108  13af7c netbsd

ls -l: 1360 -rwxr-xr-x  1 abs  wheel  1372449 Jan 25 00:20 netbsd
  size: 1056060   26916   76512 1159488  11b140 netbsd

 	The difference is due to using a slighly patched r1.11
 	version of netinet/ip_id.c, which does not include a 128K
 	table for returning a random IP 'id' number.
http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/netinet/ip_id.c?only_with_tag=MAIN

 	The first may look slightly smaller, but the second will use
 	128K less memory when loaded.

 	Its almost worth keeping the old netinet/ip_id.c around for
 	conditional compiling for install and very low memory kernels...

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

(Continue reading)

headers

Christos Zoulas | 25 Jan 2009 01:53

Re: Small kernel for sun3

In article <alpine.NEB.2.00.0901250022130.20410 <at> localhost>,
David Brownlee  <abs <at> NetBSD.org> wrote:
> 	I've been playing around with Izumi's kernel config, looking
> 	where to save space, and came up with an interesting issue
> 	- specifically how to determine kernel size.
>
> 	For example the two kernels below are very similar in size
> 	as reported by ls (the second is 673 bytes larger), but
> 	when using 'size' the second is 130620 bytes (just under
> 	128K) smaller.
>
>ls -l: 1360 -rwxr-xr-x  1 abs  wheel  1371776 Jan 25 00:07 netbsd
>  size: 1055608   26916  207584 1290108  13af7c netbsd
>
>ls -l: 1360 -rwxr-xr-x  1 abs  wheel  1372449 Jan 25 00:20 netbsd
>  size: 1056060   26916   76512 1159488  11b140 netbsd
>
> 	The difference is due to using a slighly patched r1.11
> 	version of netinet/ip_id.c, which does not include a 128K
> 	table for returning a random IP 'id' number.
>http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/netinet/ip_id.c?only_with_tag=MAIN
>
> 	The first may look slightly smaller, but the second will use
> 	128K less memory when loaded.
>
> 	Its almost worth keeping the old netinet/ip_id.c around for
> 	conditional compiling for install and very low memory kernels...
>

Well, you could allocate the array dynamically but using the old ip_id

(Continue reading)

headers

perryh | 25 Jan 2009 07:33

Re: Small kernel for sun3

> > Its almost worth keeping the old netinet/ip_id.c around for
> > conditional compiling for install and very low memory kernels...
>
> Well, you could allocate the array dynamically but using the old
> ip_id code we'll lead to predictable sequence numbers again.

How much randomness is actually needed?  Could it be obtained by
harvesting entropy as needed instead of dedicating a huge array?

headers

David Johnson | 24 Jan 2009 23:51

Re: Port help wanted/needed?

Thanks for the answer on this one.  I too have a 4MB Sun3/50M that I had 
hoped to load NETBSD onto as a diskless workstation.  The monochrome 
monitor, while difficult to move around, nevertheless has a very crisp 
display for B/W editing.  But I was unsuccessful at getting through all 
the steps diskless boot steps and then quit when I learned that the 
kernel was going to be too large in any case.

So thanks for posting this.  Time to dust it off and try again.

David Johnson / Birmingham, AL

Izumi Tsutsui wrote:
> abs <at> NetBSD.org wrote:
>
>   
>>>> 	The 3/50 has IIRC its framebuffer at 1MB, so the kernel size is
>>>> 	very constrained. Could you try building a kernel with just
>>>> 	about everything disabled for the 3/50?
>>>>         
>>> Sure thing. I lent it away a few years ago, but I can get it back over
>>> the weekend to give it a try. This one is a stock 3/50 minus the extra 8Mb
>>> board. We'll see what we can do with only 3Mb to work with. I remember
>>> the last time I did get a kernel booting on it (GENERIC, no less) a few
>>> years ago, it was swapping by the time it hit the login prompt.
>>>       
>>  	I think its worth adding a SUN350 kernel config to the tree.
>>  	I'm curious to see how much free memory we can get :)
>>     
>
> The problem is that sun3 ports requires three kernels for installation:

(Continue reading)

headers

David Brownlee | 25 Jan 2009 00:33

Re: Port help wanted/needed?

On Sat, 24 Jan 2009, David Johnson wrote:

> Thanks for the answer on this one.  I too have a 4MB Sun3/50M that I had 
> hoped to load NETBSD onto as a diskless workstation.  The monochrome monitor, 
> while difficult to move around, nevertheless has a very crisp display for B/W 
> editing.  But I was unsuccessful at getting through all the steps diskless 
> boot steps and then quit when I learned that the kernel was going to be too 
> large in any case.

 	What are you using as a server host? Were you using the Diskless
 	NetBSD HOW-TO? http://www.netbsd.org/docs/network/netboot/

 	I suspect it could do with some updating

> So thanks for posting this.  Time to dust it off and try again.

 	Let us know how it goes! :)

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

Attachment (ElectronSciences.vcf): text/x-vcard, 173 bytes

headers

jscottkasten | 21 Jan 2009 20:46

Re: Port help wanted/needed?

Actually, there probably are plenty of places you can help, especially with a strong admin and security background.

Some things might involve doing builds. I know you said you were not a coder, but I'm sure you're no stranger
to running an automated build or launching a make process to configure, build, and install an application.
If you are willing to do that, the doors are wide open.

Many of the ports run behind the i386 port even at the same release number in some areas. For example, i386
kernels ship with packet filtering and cgd enabled. Yet those types of things are not enabled, and often of
unknown status in many ports.

Back in the 1.6 days for Sun3, I had to build packet filtering into my kernel on my own. Even then, I found it
didn't work, but only because there was no magic number allocated for it in the kernel. And because no one
had tried it, even MAKEDEV didn't have an entry in the script to create the device nodes. I submitted a patch
for that and it worked great there after. Then in the 2.0 days, I tried cgd. Again, it wasn't enabled in the
kernel, so I had to build one. Once i built it and added the device nodes, life was good. It worked great.
There was no reason, other than lack of testers, that either of those things couldn't have been enabled in
the release kernels and used by someone.

My Sun 3/80 served as a DNS cache and static web page and ftp server for my DSL line for years quite reliably.
Unfortunately about a year and a half ago, the hard disk went bad, and I've still not gotten around to
replacing it, but that too will happen one day. Thusly...I don't know how the 3.x and 4.x ports stand as I
haven't had the chance to play with them lately.

Another area is pkgsrc builds. Look at what has been done, and see what is needed. I've been working with the
hpcmips port the last year. Who knew that the mesa OpenGL port actually worked.....until it was tried.

Consider also that lack of hardware need not block someone from contributing. I've had great success using
things like GXEMUL and others to run a specific port in an emulator. I've built the bulk of pkgsrc for
hpcmips 4.0 that way.

(Continue reading)

headers

John Carr | 23 Jan 2009 19:30

Re: Port help wanted/needed?

On Wed, Jan 21, 2009 at 11:46:28AM -0800, jscottkasten <at> yahoo.com wrote:
> Actually, there probably are plenty of places you can help, especially with a strong admin and security background.
> Some things might involve doing builds.  I know you said you were not a coder, but I'm sure you're no stranger
to running an automated build or launching a make process to configure, build, and install an application.
> If you are willing to do that, the doors are wide open.
> Many of the ports run behind the i386 port even at the same release number in some areas.  For example, i386
kernels ship with packet filtering and cgd enabled.  Yet those types of things are not enabled, and often of
unknown status in many ports.

All very good points. I just wasn't sure if there were areas that folks
were already concentrating where I'd be just wasting my time!

Very true. I'll admit, I haven't run NetBSD as my primary i386 desktop
in a few years now, so I don't know how big the differences are now. I
think it'd be worth it to get it going in a VM and start comparing. I
can start beating on that pretty soon.

> Another area is pkgsrc builds.  Look at what has been done, and see what is needed.  I've been working with the
hpcmips port the last year.  Who knew that the mesa OpenGL port actually worked.....until it was tried.

Heh. Good point. I can't say that I've done a meta-pkgs build of gnome
on that thing in the past few years. I've never been able to get it to
complete in the past due to different things. I guess it's time to fire
that back up.
(which, of course, begs the question of who would want to run gnome on a
3/60....)

> Consider also that lack of hardware need not block someone from contributing.  I've had great success
using things like GXEMUL and others to run a specific port in an emulator.  I've built the bulk of pkgsrc for
hpcmips 4.0 that way.

(Continue reading)