[9fans] EPIA CPU node success

[9fans] EPIA CPU node success

[ron minnich]

OK, it's a hack, don't ask how I did it, but EPIA is coming up as a CPU
node with nvram in CMOS (for now). Linuxbios on the EPIA, 9pccpu in the CF
slot.

takes 10-11 seconds to boot to a useful state (rc prompt) from ^T^Tr. It's
amazing how nice it is to debug when the cycle is that short. This is way
faster boot time than the geode cluster I had at usenix. And, I can speed
it up a bit!

Also,
echo reboot 9pccpu > /dev/reboot
is only 6 seconds. Also wonderful.

So, this shows to me that you can have a 9pccpu in the EPIA in Compact
Flash, and that 9pccpu makes a useful bootstrap, and even nicer, is faster
than 9load to do the job.

I have one thing I'd like to do, subject to someone telling me it's a bad
idea. I'm thinking of writing a small C program that does the equivalent
set of steps as the 'disk/fdisk -p blah blah ; disk/prep -p blah blah'
that are currently done in scripts. This would make it so I could move
nvram to the CF instead of CMOS. Comments?

ron

Re: [9fans] EPIA CPU node success

[Russ Cox]

> I have one thing I'd like to do, subject to someone telling me it's a bad
> idea. I'm thinking of writing a small C program that does the equivalent
> set of steps as the 'disk/fdisk -p blah blah ; disk/prep -p blah blah'
> that are currently done in scripts. This would make it so I could move
> nvram to the CF instead of CMOS. Comments?

9load does this for you.  just copy the relevant code out of 9load
(look for "partition") and drop it into /sys/src/9/boot or something.

Re: [9fans] EPIA CPU node success

[Richard Miller]

It's good to see Plan 9 running on EPIA - I think the small size and silent
running should make it a useful board.  Can the rhine driver (and video if
it's working) be made available?

Is 10-11 seconds to boot really that remarkable?  How long does the normal
BIOS and 9load take on the same setup?  After making two small tweaks to
9load, I find my thinkpad t21 goes from power-off to rc prompt in 14 seconds.
Adding grub (to allow multi booting) only takes it to 15.

-- Richard

Re: [9fans] EPIA CPU node success

[Geoff Collyer]

It depends a lot on how much (usually pointless and wasted) work your
BIOSes do.  Memory sizing/ecc-bit-setting/wasted-time can take minutes
on older machines with 512MB or more of RAM (e.g., choline).  SCSI
BIOSes, at least the Mylex ones, can take a while to probe the bus(es)
and wait to see if anybody responds after quite a few seconds.  It's
mostly a lot of several-second delays that together can make getting
to the point of starting your bootstrap, never mind your kernel, take
an annoyingly long time.  Ron and jmk can probably correct me if I'm
wrong, but I suspect that almost none of the time-consuming stuff that
the various PC BIOSes do is necessary when running a competent OS;
they may have been a help to DOS twenty years ago.

Re: [9fans] EPIA CPU node success

[ron minnich]

On Fri, 7 Nov 2003, Geoff Collyer wrote:

> Ron and jmk can probably correct me if I'm wrong, but I suspect that
> almost none of the time-consuming stuff that the various PC BIOSes do is
> necessary when running a competent OS; they may have been a help to DOS
> twenty years ago.


yep. And they still have to run dos 1.0, and that's why all the work is
done. A lot of the BIOS has to work behind DOS's back, which explains a
lot about why APM/ACMPI are so weird, not to mention all the weirdness in
the hardware that also dates back to 1980, such as shadow registers. Don't
get me started :-)

ron

Re: [9fans] EPIA CPU node success

[ron minnich]

On Fri, 7 Nov 2003, Richard Miller wrote:

> Is 10-11 seconds to boot really that remarkable?  How long does the normal
> BIOS and 9load take on the same setup?

I think longer.

> After making two small tweaks to 9load, I find my thinkpad t21 goes from
> power-off to rc prompt in 14 seconds. Adding grub (to allow multi
> booting) only takes it to 15.

That's impressive. My is not quite that fast.

But overall plan 9 sure boots faster than linux nowadays ...

ron

Re: [9fans] EPIA CPU node success

[Vester Thacker]

On Fri, Nov 07, 2003 at 09:01:37AM +0000, Richard Miller wrote:
> It's good to see Plan 9 running on EPIA - I think the small size and silent
> running should make it a useful board.  Can the rhine driver (and video if
> it's working) be made available?

Aki Nyrhinen's rhine driver can be found at
http://www.tp9ug.jp/archive/src/drivers/index.html

-- Vester Thacker

Re: [9fans] EPIA CPU node success

[Charles Forsyth]

>>Aki Nyrhinen's rhine driver can be found at

how appropriate that he should write the driver!

Re: [9fans] EPIA CPU node success

[jmk]

On Fri Nov  7 16:38:37 EST 2003, vthacker@0xfe.org wrote:
> On Fri, Nov 07, 2003 at 09:01:37AM +0000, Richard Miller wrote:
> > It's good to see Plan 9 running on EPIA - I think the small size and silent
> > running should make it a useful board.  Can the rhine driver (and video if
> > it's working) be made available?
>
> Aki Nyrhinen's rhine driver can be found at
> http://www.tp9ug.jp/archive/src/drivers/index.html
>
> -- Vester Thacker

Oops. I've had that for a while and I was sure I'd put it on sources,
it's in our local tree. I've pushed it out now (it's slightly different
to remove some compiler warnings).

Re: [9fans] EPIA CPU node success

[Scott Schwartz]

> 9load does this for you.  just copy the relevant code out of 9load
> (look for "partition") and drop it into /sys/src/9/boot or something.

Or even, factor it out, put the code in a library, and have one copy
for everyone to share.

Re: [9fans] EPIA CPU node success

[ron minnich]

On 6 Nov 2003, Scott Schwartz wrote:

> Or even, factor it out, put the code in a library, and have one copy
> for everyone to share.


Sold.

ron