Re: [9fans] Crazy idea... or a new project?

Re: [9fans] Crazy idea... or a new project?

[rob pike]

> Another way could be to be able to link the kernel
> with device drivers before loading it, and keep
> the kernel just with devices used. I think that
> could be done by making each driver provide a
> (user) program to check if the device is there,
> and linking a second kernel depending on what's
> there. 

You don't have a user program running when you're trying
to boot the kernel.

-rob

Re: [9fans] Crazy idea... or a new project?

[Francisco J Ballesteros]

I was talking about booting a kernel with
the bare minimum to be able to run a couple of
user programs to link a new kernel. Then reboot.
But perhaps not a good idea. Because
at least IDE and ether drivers should be in place.

rob pike wrote:
> 
> > Another way could be to be able to link the kernel
> > with device drivers before loading it, and keep
> > the kernel just with devices used. I think that
> > could be done by making each driver provide a
> > (user) program to check if the device is there,
> > and linking a second kernel depending on what's
> > there.
> 
> You don't have a user program running when you're trying
> to boot the kernel.
> 
> -rob

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

From: Francisco J Ballesteros <nemo@gsyc.escet.urjc.es>


> I was talking about booting a kernel with
> the bare minimum to be able to run a couple of
> user programs to link a new kernel. Then reboot.
> But perhaps not a good idea. Because
> at least IDE and ether drivers should be in place.
> 

that is a ghastly idea.  boot twice?  look, one of
the _features_ of a plan 9 terminal is that it
boots in about a second, well i'm talking about
a gnot.

christ, a vax 11/780 booted faster than most modern
systems and it was i/o bound on the la120.

given inferno has loadable modules, it wouldn't be
too hard to add loadable device drivers to it.  the
who point of a loadable device driver is _not_ to
reboot.

IDE?  very general.

Re: [9fans] Crazy idea... or a new project?

[Francisco J Ballesteros]

Boyd Roberts wrote:
> that is a ghastly idea.  boot twice?  look, one of
> the _features_ of a plan 9 terminal is that it
> boots in about a second, well i'm talking about
> a gnot.

you only boot twice each time you change your hardware.
Next time you already have a complete kernel in place.

If you boot from the net, I think I prefer the dhcp idea.

And in any case, the real problem is that of "p-n-p" (as
Presotto said), and
I have no clue about how to do it right on a PC.

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

From: Francisco J Ballesteros <nemo@gsyc.escet.urjc.es>
>
> you only boot twice each time you change your hardware.
> Next time you already have a complete kernel in place.
> 

maybe, but i think you have forgotton a fundemental
design principle:

    no disk, no fan

you _always_ boot across the wire and no you don't
do it like NFS diskless boots.

boot with a generic and then load in the other pieces,
on the fly.

you wouldn't happen to use linux, by chance?

Re: [9fans] Crazy idea... or a new project?

[Francisco J Ballesteros]

Boyd Roberts wrote:
> 
>     no disk, no fan
> 
> you _always_ boot across the wire and no you don't
> do it like NFS diskless boots.

for a local boot, I think I could afford a second
boot if I changed something.
but, again, for a regular terminal booting from
the net I agree it´s best to let the server link
the kernel (if you changed hw).

> 
> boot with a generic and then load in the other pieces,
> on the fly.

I just do not like dynamic loading because in the end,
I found I (most of the times) had to rebuild my unix
kernel whenever I changed hw---you know, forgot to
configure an (static) option needed for the new board, or just
got the module out of date wrt the kernel, or just the
hardware was new and there was no module before, ...

> 
> you wouldn't happen to use linux, by chance?

I use plan9 booting from a local disk for almost everything,
and I also use linux. You are right in that
I was thinking along the way I boot the kernel and not
along the way a terminal should boot---otherwise it would
have been kind of obvious to do the link in the server instead
of booting the terminal twice; but I never thought about it.

have a good night. zzZ zz

PS: still laughing because of the p-n-p + n-p-n :-)

Re: [9fans] Crazy idea... or a new project?

[okamoto]

nclude: /mail/fs/mbox/41/raw

>Yup. Everyone knows the right answer is to combine
>p-n-p with n-p-n.

I rather love to listen music from oldest glass tubes.  :-)

Kenji

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

From: <okamoto@granite.cias.osakafu-u.ac.jp>
> 
> I rather love to listen music from oldest glass tubes.  :-)
> 

dunno, i got this marshall valvestate practice amp [solid state]
and they seem to have analysed the problem pretty well.  with
a les paul standard and enough pre-amp overdrive it sounds
pretty good.

Re: [9fans] Crazy idea... or a new project?

[Andy Newman]

>dunno, i got this marshall valvestate practice amp [solid state]
>and they seem to have analysed the problem pretty well.  with
>a les paul standard and enough pre-amp overdrive it sounds
>pretty good.

I was going to mention guitar amps.  I reckon the solid-state
ones sound okay at low volumes but get someplace where you can
crank 'em and they're still shit.   And pre-amp overdrive. Bah!
You want them Celestions jumping out of their frames to get the
real break up sound :))))))


--
Andy

Re: [9fans] Crazy idea... or a new project?

[rob pike]

Yup. Everyone knows the right answer is to combine
p-n-p with n-p-n.

-rob

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

> Yup. Everyone knows the right answer is to combine
> p-n-p with n-p-n.

plug and pray [that it boots].

could chuck in the odd FET.

Re: [9fans] Crazy idea... or a new project?

[Fariborz 'Skip' Tavakkolian]

It took me a minute. good one.
I can't resist suggesting the obvious solution
which is to t-n-t the p-n-p.

At 06:47 PM 11/24/00 -0500, rob pike wrote:
>Yup. Everyone knows the right answer is to combine
>p-n-p with n-p-n.
>
>-rob
>
>

Re: [9fans] Crazy idea... or a new project?

[presotto]

[-- Attachment #1: Type: text/plain, Size: 1035 bytes --]

I seem to be missing messages from this stream so
excuse me if this has already been suggested:

I guess it depends on what problem you are trying to
solve.  We have lots (> 100) of systems that boot off
of LAN's.  We end up with kernels stuffed full of
drivers because otherwise, we'ld have to keep a database
lying around somewhere for each system and spend time
making lots of kernels each time we change something.
Granted we could write enough software to make this
painless, but it's one more thing to screw up.

I was actually fond of an idea I've heard from Rob or
Ken, can't remember which.  To address the problem, he
suggested the dhcp daemon link the kernel on the fly.
It could use a hardware vector passed in the dhcp request
(generated from plan9.ini perhaps) and return the file
name of the result so that the system could tftp it over.

Unfortunately, I find this all a lot less of a problem than
figuring out what IRQ's and IO ports to assign devices or that
of getting a new VGA board working.

[-- Attachment #2: Type: message/rfc822, Size: 2314 bytes --]

From: "Boyd Roberts" <boyd@planete.net>
To: <9fans@cse.psu.edu>
Subject: Re: [9fans] Crazy idea... or a new project?
Date: Fri, 24 Nov 2000 23:24:09 +0100
Message-ID: <01e701c05665$44417500$0ab9c6d4@cybercable.fr>

From: Scott Schwartz <schwartz@bio.cse.psu.edu>

> Just to digress for a moment, I've never liked the idea of loadable
> device drivers, because they seeem like such a stopgap measure.

maybe you could 'plumb' them.  after boot the kernel gets
handed up a file with rules/whatever to load device drivers
based on interrupts or other device ready 'events'.

i was thinking about digital uda-50 interfaces.  you could
just plug 'em in and ultrix would autoconf them at run
(ie. any) time.  of course, the driver was already loaded.

probably hard to generalise, but it might work out.

Re: [9fans] Crazy idea... or a new project?

[rob pike]

Having the window system allocate memory from unpageable ram bothers me too,
and is surely fixable, but putting pixels in user space with user code pumping them
had worse problems.  The last pieces of this design space have not been explored.

-rob

Re: [9fans] Crazy idea... or a new project?

[nemo]

[-- Attachment #1: Type: text/plain, Size: 308 bytes --]

Another way could be to be able to link the kernel
with device drivers before loading it, and keep
the kernel just with devices used. I think that
could be done by making each driver provide a
(user) program to check if the device is there,
and linking a second kernel depending on what's
there. 



[-- Attachment #2: Type: message/rfc822, Size: 4114 bytes --]

[-- Attachment #2.1.1: Type: text/plain, Size: 580 bytes --]

That's a little simplistic.  The x86 is a lousy architecture,
and you're right about the compilers etc.  But all those
lousy drivers, VGAs, USB floppies, it goes on and on.
Are any two the same?  I doubt it.  That's the horror of
PCs, yet is a perhaps unavoidable consequence of the
history that made the PC dominant (c.f. the Mac).

The PC kernel for Plan 9 is huge because of all the drivers
it must have to run on all devices.  Sure we could have
(and perhaps should have) loadable device drivers,
but that is a solution to a problem that shouldn't exist.

-rob


[-- Attachment #2.1.2: Type: message/rfc822, Size: 1463 bytes --]

From: "Russ Cox" <rsc@plan9.bell-labs.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Crazy idea... or a new project?
Date: Thu, 23 Nov 2000 19:05:08 -0500
Message-ID: <20001124000509.EF643199EE@mail.cse.psu.edu>

I don't agree about the x86.
Sure it's an ugly chip.
Sure it and its PC hosts are complicated.
But the bootstrap code and the compiler
are written.  In Plan 9, it's no different
to write applications for the x86 than it
is to write them for any other architecture.

Russ

Re: [9fans] Crazy idea... or a new project?

[rob pike]

[-- Attachment #1: Type: text/plain, Size: 580 bytes --]

That's a little simplistic.  The x86 is a lousy architecture,
and you're right about the compilers etc.  But all those
lousy drivers, VGAs, USB floppies, it goes on and on.
Are any two the same?  I doubt it.  That's the horror of
PCs, yet is a perhaps unavoidable consequence of the
history that made the PC dominant (c.f. the Mac).

The PC kernel for Plan 9 is huge because of all the drivers
it must have to run on all devices.  Sure we could have
(and perhaps should have) loadable device drivers,
but that is a solution to a problem that shouldn't exist.

-rob


[-- Attachment #2: Type: message/rfc822, Size: 1463 bytes --]

From: "Russ Cox" <rsc@plan9.bell-labs.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Crazy idea... or a new project?
Date: Thu, 23 Nov 2000 19:05:08 -0500
Message-ID: <20001124000509.EF643199EE@mail.cse.psu.edu>

I don't agree about the x86.
Sure it's an ugly chip.
Sure it and its PC hosts are complicated.
But the bootstrap code and the compiler
are written.  In Plan 9, it's no different
to write applications for the x86 than it
is to write them for any other architecture.

Russ

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

> ... Sure we could have
> (and perhaps should have) loadable device drivers,
> but that is a solution to a problem that shouldn't exist.
> 

yes, in your environment.

but like you can glue a namespace together, why not device
drivers?

Re: [9fans] Crazy idea... or a new project?

[Scott Schwartz]

| Sure we could have
| (and perhaps should have) loadable device drivers,
| but that is a solution to a problem that shouldn't exist.

Just to digress for a moment, I've never liked the idea of loadable
device drivers, because they seeem like such a stopgap measure.  Either a
real microkernel (ala QNX) or a conventional but pagable kernel would
address the space issue in a (IMHO) cleaner way.  Having the window
system allocate memory from unpageable ram bothers me in a similar way.

I'm sure rob's thought about these issues more than I, so I look forward
to being set straight on the topic. :-)

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

From: Scott Schwartz <schwartz@bio.cse.psu.edu>

> Just to digress for a moment, I've never liked the idea of loadable
> device drivers, because they seeem like such a stopgap measure.

maybe you could 'plumb' them.  after boot the kernel gets
handed up a file with rules/whatever to load device drivers
based on interrupts or other device ready 'events'.

i was thinking about digital uda-50 interfaces.  you could
just plug 'em in and ultrix would autoconf them at run
(ie. any) time.  of course, the driver was already loaded.

probably hard to generalise, but it might work out.

Re: [9fans] Crazy idea... or a new project?

[Russ Cox]

I don't agree about the x86.
Sure it's an ugly chip.
Sure it and its PC hosts are complicated.
But the bootstrap code and the compiler
are written.  In Plan 9, it's no different
to write applications for the x86 than it
is to write them for any other architecture.

Russ

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

From: Russ Cox <rsc@plan9.bell-labs.com>

> I don't agree about the x86.
> Sure it's an ugly chip.

that's my point.

> Sure it and its PC hosts are complicated.
> But the bootstrap code and the compiler
> are written.  In Plan 9, it's no different
> to write applications for the x86 than it
> is to write them for any other architecture.

violent agreement.

[9fans] Crazy idea... or a new project?

[Andrew Zubinski]

I think that I'm correct in next sentences:

1. Plan9 support community (both developers and users) is small
2. Any OS (even so beautifull like Plan9) whithout x86 platform support
is dead OS
3. x86 platform is evil for small community
4. Because small community must support all spectre of PC-compatible
garbage
5. So, there are no free time to develop application software
6. And without application software even great OS will be no more than
"very
interesting thing" for very small community

IMHO, there is only one way to realize all Plan9 power - new hardware
platform:
low-cost, open at specifications level, non-x86.
Why it must be open and cheap?
Are you remember the stories of NeXT and Be?

Let it be wearable Plan9 terminal especially designed with requirements
of small
electronics companies in mind (simple and small PCB, chips with small
pin-outs). And for a CPU & file servers we must create portable and
POSIX-OS-independent implementations (it is hard but possible).

As engineer I see the prototype of such "Plan9Wear":
------------------
System CPU - low-cost ($25-30 in sample quantity) MIPS-clone
(with frequency about 100-133 MHz and power ~1-3 W)
like IDT R/RC4640 or NEC). And we have the MIPS kernel...
------------------
System DSP - for the bitblt implementation, speech controls (why not?)
and etc.
------------------
System Bus Controller - FPGA based or low-cost single chip
PCI bridge (DRAM controller)
------------------
System Memory - 128 MB standard DIMM + Flash BOOT
------------------
Interfaces - USB 2.0 + IEEE-802.11a high-speed wireless
------------------
Grayscale microdisplay (Kopin 1280x1024, 256 gray levels), $100-150
(???)

IMHO, the wearable Plan9 will be really great thing. And main system
ideas are very
suitable for wireless and wearable applications.

Maybe it is time to start development ????

P.S.
Sorry for my ugly English... :-)

[9fans] Crazy idea... or a new project?

[Richard]

Andrew Zubinski writes:

>I think that I'm correct in next sentences:
>
>1. Plan9 support community (both developers and users) is small
>2. Any OS (even so beautifull like Plan9) whithout x86 platform support
>is dead OS
>3. x86 platform is evil for small community
>4. Because small community must support all spectre of PC-compatible
>garbage
>5. So, there are no free time to develop application software
>6. And without application software even great OS will be no more than
>"very interesting thing" for very small community

agree.  and let me add that

7. the more users an OS has, the more value it offers to the average user,
   and hence, the easier it is for the OS to attract new users.

Re: [9fans] Crazy idea... or a new project?

[Boyd Roberts]

From: Andrew Zubinski <andrew@itc.kiev.ua>

> 3. x86 platform is evil for small community

3. x86 platform is evil.