[9fans] SATA controller found by kernel but not 9load

[9fans] SATA controller found by kernel but not 9load

[Anthony Sorace]

I'm trying to install a new system and I've run into a problem: the
SATA controller is recognized fine by the kernel, but not by 9load.
The kernel tells me:

#S/sdE: ahci: port 0xe0000000: hba sss 0; ncs 31; coal 1; mports 3;
led 1; clo 1; ems 0
#S/sdE: sb600: sata-II with 4 ports

but 9load has nothing at all to say about it. Once I'm up with a
kernel (say, loaded off the CD which is still attached to (p)ATA and
works fine with 9load), I can access it without issue. In the normal
kernel, this is handled by sdiahci.c, which seems to be an older
version in 9load.

How straight-forward is it to move the kernel version of the driver
into 9load? I don't suppose anyone's already looked at that for this
device?

As an aside: how space-constrained are the kernels used for install
currently? It would be very nice to have fs(3) in there so an install
can be done with mirrored disks, rather than having to retrofit it
after the fact.
Anthony

Re: [9fans] SATA controller found by kernel but not 9load

[Gorka Guardiola]

On Nov 20, 2007 4:35 AM, Anthony Sorace <anothy@gmail.com> wrote:
> How straight-forward is it to move the kernel version of the driver
> into 9load? I don't suppose anyone's already looked at that for this
> device?

Does your bios recognize it?. You can try booting with the bios device
in 9load.

-- 
- curiosity sKilled the cat

Re: [9fans] SATA controller found by kernel but not 9load

[Anthony Sorace]

Geoff asked about *noahciload in my plan9.ini; no, it is not set. None
of the *no variables are.

Gorka asked about my bios: yes, my bios knows about it. I've not got
any of the legacy compatibility stuff turned on, as my initial
experiments showed it made them not detected by the kernel, either
(although there are several modes, and I didn't try them all).

I tried the bios loading (sdB0), but that was unproductive. No example
is given, but from the description in the man page I'd expect it to
parallel sdC0 and look like sdB0!9fat!9pcf. When 9load(8) says about
sdB0 "partitions are understood", I'm assuming it means in the same
way sdC0 understands them.

What's interesting, though, is that giving it sdC0!cdboot!9pccd.gz has
no effect when I use the on-disk 9load, either (which works fine using
the one on the cd), with the distributed plan9.ini (which does have
some *no variables set). More poking to do.

Re: [9fans] SATA controller found by kernel but not 9load

[erik quanstrom]

the problem with 9load is that it was written before i did
support for the sb600 chipset.  i now have a machine with
that northbridge, but i haven't finished with 9load.

i'm working on it right now.

(p.s. if you feel brave, /n/sources/contrb/quanstro/9loadsb600)

- erik

> Geoff asked about *noahciload in my plan9.ini; no, it is not set. None
> of the *no variables are.
> 
> Gorka asked about my bios: yes, my bios knows about it. I've not got
> any of the legacy compatibility stuff turned on, as my initial
> experiments showed it made them not detected by the kernel, either
> (although there are several modes, and I didn't try them all).
> 
> I tried the bios loading (sdB0), but that was unproductive. No example
> is given, but from the description in the man page I'd expect it to
> parallel sdC0 and look like sdB0!9fat!9pcf. When 9load(8) says about
> sdB0 "partitions are understood", I'm assuming it means in the same
> way sdC0 understands them.
> 
> What's interesting, though, is that giving it sdC0!cdboot!9pccd.gz has
> no effect when I use the on-disk 9load, either (which works fine using
> the one on the cd), with the distributed plan9.ini (which does have
> some *no variables set). More poking to do.

Re: [9fans] SATA controller found by kernel but not 9load

[Anthony Sorace]

I'm not sure if 9loadsb600 is far enough along for bug reports to be
useful, but here you go.

It gets significantly further than it had been. Far enough that it
looks as though 9load's work might be done: it finds my plan9.ini and
loads the kernel indicated (both on a 9fat partition on sdE0).

Now, however, my booted kernels can't find their root partition on the
disk. They display the expected message about finding the sb600, but
attempts to boot off local!#S/sdE0/fossil return errors about not
being able to find the server. I'm certain that there's a valid fossil
partition there (I can mount it while booted off the CD). More
digging.
Anthony

Re: [9fans] SATA controller found by kernel but not 9load

[erik quanstrom]

On Tue Nov 20 20:14:16 EST 2007, anothy@gmail.com wrote:
> I'm not sure if 9loadsb600 is far enough along for bug reports to be
> useful, but here you go.
> 
> It gets significantly further than it had been. Far enough that it
> looks as though 9load's work might be done: it finds my plan9.ini and
> loads the kernel indicated (both on a 9fat partition on sdE0).
> 
> Now, however, my booted kernels can't find their root partition on the
> disk. They display the expected message about finding the sb600, but
> attempts to boot off local!#S/sdE0/fossil return errors about not
> being able to find the server. I'm certain that there's a valid fossil
> partition there (I can mount it while booted off the CD). More
> digging.

hey!  good deal.  i think you were having a problem with the serial
controller.  i don't understand why this is, but i was having the
same problem on my guinea pig machine.

i would think that 9load has done it's job successfully.  but somehow,
fossil is not configured correctly.  i would be interested in the output
of fossil/flconf when booted from the cd.  can you mount fossil when
you boot from cd?

one final question. how many sata drives do you have (including cdroms)?

- erik

Re: [9fans] SATA controller found by kernel but not 9load

[Anthony Sorace]

I can mount my fossil when booted from the cdrom. The output of
fossil/conf /dev/sdE0/fossil is as follows:

fsys main config /dev/sdE0/fossil
fsys main open -AWVP
fsys other config /dev/sdE0/other
fsys other open -AWVP
srv -p fscons
srv fossil

Running fossil/fossil -f /dev/sdE0/fossil while booted form the CD
returns no errors, and subsequent mounts of of both main and other
work as expected.

I'm beginning to suspect that the kernels on the CD in /386 don't
match their descriptions in /sys/src/9/pc. Maybe it's simply timing as
things die, but I can't get them to show me there ever being a fossil
process.

I have three SATA hard disks attached. Plan 9 identifies them as
sdE[013]; the BIOS identifies them as Channel 2 Master, Channel 3
Master, and Channel 3 Slave (which would lead me to expect sdE[023] or
sdE0+sdF[01]). I also have an IDE cdrom on sdC0 (Channel 0 Master in
the BIOS).
Anthony

Re: [9fans] SATA controller found by kernel but not 9load

[erik quanstrom]

> I'm beginning to suspect that the kernels on the CD in /386 don't
> match their descriptions in /sys/src/9/pc. Maybe it's simply timing as
> things die, but I can't get them to show me there ever being a fossil
> process.
> 
> I have three SATA hard disks attached. Plan 9 identifies them as
> sdE[013]; the BIOS identifies them as Channel 2 Master, Channel 3
> Master, and Channel 3 Slave (which would lead me to expect sdE[023] or
> sdE0+sdF[01]). I also have an IDE cdrom on sdC0 (Channel 0 Master in
> the BIOS).
> Anthony

master/slave/channel doesn't make asense for sata.  it's all
point-to-point.  all sata drives attached to a single hba will
show up as the same drive letter, in your case E, the first sata
drive letter.

in any case, i think you are right.  it is a timing issue.  since ahci drives
are hot-pluggable, the drive might not be available immediately on boot.
i'll take a look.  i'm not sure if fossil needs more patience or sdiahci
needs to force disks up sooner.  perhaps a combination of the two.

- erik

Re: [9fans] SATA controller found by kernel but not 9load

[erik quanstrom]

> 
> I'm beginning to suspect that the kernels on the CD in /386 don't
> match their descriptions in /sys/src/9/pc. Maybe it's simply timing as
> things die, but I can't get them to show me there ever being a fossil
> process.
> 

hopefully you can connect to sources and download
	/n/sources/contrib/quanstro/sb600/9pccpuf 
or
	/n/sources/contrib/quanstro/sb600/9pcf
(whichever you're using) from the livecd.  if you're
having a timing problem, this kernel ought to fix it.
(assuming i didn't break anything else.)  i'll submit
the patch if it works for you.

unfortunately, i can't test yet myself as the livecd can't
find itself if you boot from a sata cdrom.

- erik

Re: [9fans] SATA controller found by kernel but not 9load

[Anthony Sorace]

well, the failure mode is now quite different. i got a panic which i
assumed meant the kernel had given up, but as i was typing this i got
a timeout message from ipconfig. something's still running, clearly.
unfortunately, most of the content scrolled off the top when i got a
dumpstack from whatever it was that crashed ("panic: fault: 0x11c"
mean anything to anyone?) - i'll get a serial console hooked up
tomorrow (while not cooking) and capture the full print.

there's still no fossil running, though; perhaps that's what crashed.
anthony

Re: [9fans] SATA controller found by kernel but not 9load

[erik quanstrom]

On Wed Nov 21 23:31:36 EST 2007, anothy@gmail.com wrote:
> well, the failure mode is now quite different. i got a panic which i
> assumed meant the kernel had given up, but as i was typing this i got
> a timeout message from ipconfig. something's still running, clearly.
> unfortunately, most of the content scrolled off the top when i got a
> dumpstack from whatever it was that crashed ("panic: fault: 0x11c"
> mean anything to anyone?) - i'll get a serial console hooked up
> tomorrow (while not cooking) and capture the full print.
> 
> there's still no fossil running, though; perhaps that's what crashed.
> anthony

Re: [9fans] SATA controller found by kernel but not 9load

[erik quanstrom]

On Wed Nov 21 23:31:36 EST 2007, anothy@gmail.com wrote:
> well, the failure mode is now quite different. i got a panic which i
> assumed meant the kernel had given up, but as i was typing this i got
> a timeout message from ipconfig. something's still running, clearly.
> unfortunately, most of the content scrolled off the top when i got a
> dumpstack from whatever it was that crashed ("panic: fault: 0x11c"
> mean anything to anyone?) - i'll get a serial console hooked up
> tomorrow (while not cooking) and capture the full print.
> 
> there's still no fossil running, though; perhaps that's what crashed.
> anthony

Re: [9fans] SATA controller found by kernel but not 9load

[Anthony Sorace]

okay. between the holiday and the fact that this is the first cpu
server i've had to set up without benefit of a serial console, this
took much longer to confirm than expected, but it does seem to have
worked. i've now got my cpu server booting off my local fossil on
sdE0. i also tested the terminal kernel, which also worked great.

very much thanks for your help. now on to the next bit 'o fun, getting
it to boot off a fs(3)-backed venti.
anthony

[9fans] SATA controller found by kernel but not 9load

[erik quanstrom]

> How straight-forward is it to move the kernel version of the driver
> into 9load? I don't suppose anyone's already looked at that for this
> device?

i missed this question earlier.  sometimes it's straightforward.  but
generally it's difficult.  there are two reasons for this.  first, 9load
doesn't have any procs, tasks or any other multiprogramming constructs.
so sleep and wakeup are out the window.  the other reason is that
the problem of writing a boot-loading driver is conceptually different
than writing a kernel driver.  performance just isn't an issue.  getting
the job done with as few resources as possible is the goal.

- erik