6.4/hppa regression: no LAN or UWSCSI on C360

Next Topic
 
classic Classic list List threaded Threaded
7 messages Options
Reply | Threaded
Open this post in threaded view
|

6.4/hppa regression: no LAN or UWSCSI on C360

Tom Lane
I've just been trying to install OpenBSD on an HP C360, and running into
some problems.  OpenBSD 6.4 boots okay, but it does not recognize either
the machine's LAN interface or its internal ultra-wide SCSI bus, making
it pretty much useless except as a doorstop.

I then tried installing older versions to see if this was a regression,
and it is.  OpenBSD 6.2 recognizes those devices just fine.

I don't have the patience to manually transcribe the full dmesg output
from the HP's screen (too bad the ftp client in lif62.fs is lobotomized,
or I could transfer it somewhere).  But the relevant part of 6.2's dmesg
output seems to be, plus or minus a typo or so,

dino1 at uturn1 offset 2000000 irq 25: Dino V3.1
com1 at dino1 irq 11: ns16550a, 16 byte fifo
com1: probed fifo depth: 0 bytes
pci1 at dino1
sti3 at pci1 dev 1 function 0 "Hewlett-Packard Visualize EG" rev 0x03
[ a couple more lines describing sti3 here ]
siop0 at pci1 dev 19 function 0 "Symbios Logic 53c875" rev 0x04: irq 4, using 4K of on-board RAM
scsibus1 at siop0: 16 targets, initiator 7
sd0 at scsibus1 targ 5 lun 0: <SEAGATE, ST336607LW, 0007> SCSI3 0/direct fixed serial.SEAGATE_ST336607LW_3JA9ZSFP00007522DLQB
sd0: 35003MB, 512 bytes/sector, 71687372 sectors
dc0 at pci1 dev 20 function 0 "DEC 21442/3" rev 0x30: irq 1, address 00:60:b0:d0:6b:73
lxtphy0 at dc0 phy 1: LXT970 10/100 PHY, rev. 1

OpenBSD 6.3 also recognizes these devices, but it has some weird problem
with the PS/2 keyboard: it keeps repeating whatever key I last pressed.

6.4's not got the keyboard problem, but its dmesg output doesn't show
any of the above lines, so it seems like the root problem may be failure
to recognize the "dino1" interface.  (It does report "uturn1".)

Happy to probe further, but I haven't done kernel-level hacking in
decades, so my skillz for this are pretty limited.

                        regards, tom lane

Reply | Threaded
Open this post in threaded view
|

Re: 6.4/hppa regression: no LAN or UWSCSI on C360

Miod Vallat

> I've just been trying to install OpenBSD on an HP C360, and running into
> some problems.  OpenBSD 6.4 boots okay, but it does not recognize either
> the machine's LAN interface or its internal ultra-wide SCSI bus, making
> it pretty much useless except as a doorstop.
>
> I then tried installing older versions to see if this was a regression,
> and it is.  OpenBSD 6.2 recognizes those devices just fine.

Argh. This is a regression in the installation kernel only, which was
introduced by changes I wrote in the 6.4 development cycle. I
incorrectly removed the attachment rule for dino at uturn1. Sorry about
that.

Unfortunately, there is no easy way to repair this from UKC; but if you
are able to use a serial console during the installation or upgrade, you
may use UKC to swap uturn0 and uturn1; instead of losing the on-board
devices (Ethernet, SCSI), this will lose the PCI cards (and thus, the
display).

To achieve this, boot the installation media and, at the "boot>" prompt,
enter "-c". Then, at the "UKC>" prompt, enter "change uturn0". This will
continue in the following dialog, where you need to answer the questions
as follows:

  UKC> change uturn0
  100 uturn0 at mainbus0 offset -1 irq -1 flags 0x0
  change (y/n) ? y
  offset [-1] ? 0xff88000
  irq [-1] ? (press return)
  flags [0] ? (press return)
  100 uturn0 changed
  100 uturn0 at mainbus0 offset 0xff88000 irq -1 flags 0x0
  UKC>

Then when you're back at the "UKC>" prompt, enter "quit". The kernel
will resume booting, and will find the Ethernet and SCSI controllers,
but not the sti(4) display.

You can then proceed with the installation or upgrade as usual.


Note that, once you have installed or upgraded your system, the regular
kernel does not suffer from this problem and will recognize all your
devices.


The source code fix can be found below.

Index: sys/arch/hppa/conf/RAMDISK
===================================================================
RCS file: /OpenBSD/src/sys/arch/hppa/conf/RAMDISK,v
retrieving revision 1.111
diff -u -p -r1.111 RAMDISK
--- sys/arch/hppa/conf/RAMDISK 20 Sep 2018 15:19:38 -0000 1.111
+++ sys/arch/hppa/conf/RAMDISK 24 Dec 2018 19:41:12 -0000
@@ -59,7 +59,7 @@ wax* at phantomas0 # Wax on [AB]*
 wax* at uturn? # Wax on C*
 mongoose* at mainbus0 irq 17 # EISA Bus Adapter (i82350 or TI???)
 dino* at phantomas? # PCI bus bridge on [AB]*
-dino* at uturn0 # PCI bus bridge on [CJ]*
+dino* at uturn? # PCI bus bridge on [CJ]*
 pci* at dino?
 com1 at dino? irq 11
 

Reply | Threaded
Open this post in threaded view
|

Re: 6.4/hppa regression: no LAN or UWSCSI on C360

Mark Kettenis
> From: Miod Vallat <[hidden email]>
> Date: Mon, 24 Dec 2018 20:05:02 +0000 (UTC)
>
> > I've just been trying to install OpenBSD on an HP C360, and running into
> > some problems.  OpenBSD 6.4 boots okay, but it does not recognize either
> > the machine's LAN interface or its internal ultra-wide SCSI bus, making
> > it pretty much useless except as a doorstop.
> >
> > I then tried installing older versions to see if this was a regression,
> > and it is.  OpenBSD 6.2 recognizes those devices just fine.
>
> Argh. This is a regression in the installation kernel only, which was
> introduced by changes I wrote in the 6.4 development cycle. I
> incorrectly removed the attachment rule for dino at uturn1. Sorry about
> that.
>
> Unfortunately, there is no easy way to repair this from UKC; but if you
> are able to use a serial console during the installation or upgrade, you
> may use UKC to swap uturn0 and uturn1; instead of losing the on-board
> devices (Ethernet, SCSI), this will lose the PCI cards (and thus, the
> display).
>
> To achieve this, boot the installation media and, at the "boot>" prompt,
> enter "-c". Then, at the "UKC>" prompt, enter "change uturn0". This will
> continue in the following dialog, where you need to answer the questions
> as follows:
>
>   UKC> change uturn0
>   100 uturn0 at mainbus0 offset -1 irq -1 flags 0x0
>   change (y/n) ? y
>   offset [-1] ? 0xff88000
>   irq [-1] ? (press return)
>   flags [0] ? (press return)
>   100 uturn0 changed
>   100 uturn0 at mainbus0 offset 0xff88000 irq -1 flags 0x0
>   UKC>
>
> Then when you're back at the "UKC>" prompt, enter "quit". The kernel
> will resume booting, and will find the Ethernet and SCSI controllers,
> but not the sti(4) display.
>
> You can then proceed with the installation or upgrade as usual.
>
>
> Note that, once you have installed or upgraded your system, the regular
> kernel does not suffer from this problem and will recognize all your
> devices.
>
>
> The source code fix can be found below.

Should I commit that fix to -current?


> Index: sys/arch/hppa/conf/RAMDISK
> ===================================================================
> RCS file: /OpenBSD/src/sys/arch/hppa/conf/RAMDISK,v
> retrieving revision 1.111
> diff -u -p -r1.111 RAMDISK
> --- sys/arch/hppa/conf/RAMDISK 20 Sep 2018 15:19:38 -0000 1.111
> +++ sys/arch/hppa/conf/RAMDISK 24 Dec 2018 19:41:12 -0000
> @@ -59,7 +59,7 @@ wax* at phantomas0 # Wax on [AB]*
>  wax* at uturn? # Wax on C*
>  mongoose* at mainbus0 irq 17 # EISA Bus Adapter (i82350 or TI???)
>  dino* at phantomas? # PCI bus bridge on [AB]*
> -dino* at uturn0 # PCI bus bridge on [CJ]*
> +dino* at uturn? # PCI bus bridge on [CJ]*
>  pci* at dino?
>  com1 at dino? irq 11
>  
>
>

Reply | Threaded
Open this post in threaded view
|

Re: 6.4/hppa regression: no LAN or UWSCSI on C360

Miod Vallat
> Should I commit that fix to -current?

I have not tested it yet, still baking the muild. But it should be TRT
as it matches GENERIC.

Reply | Threaded
Open this post in threaded view
|

Re: 6.4/hppa regression: no LAN or UWSCSI on C360

Tom Lane
In reply to this post by Miod Vallat
Miod Vallat <[hidden email]> writes:
>> I've just been trying to install OpenBSD on an HP C360, and running into
>> some problems.  OpenBSD 6.4 boots okay, but it does not recognize either
>> the machine's LAN interface or its internal ultra-wide SCSI bus, making
>> it pretty much useless except as a doorstop.

> Argh. This is a regression in the installation kernel only, which was
> introduced by changes I wrote in the 6.4 development cycle. I
> incorrectly removed the attachment rule for dino at uturn1. Sorry about
> that.

Thanks for the quick response!

> Unfortunately, there is no easy way to repair this from UKC; but if you
> are able to use a serial console during the installation or upgrade, you
> may use UKC to swap uturn0 and uturn1; instead of losing the on-board
> devices (Ethernet, SCSI), this will lose the PCI cards (and thus, the
> display).

Meh --- think I'll just wait for the fix to reach the snapshot build,
and then install from that.  I'm waiting on a replacement disk to
show up before I can really go past the "does it boot" stage, anyway.

Again, thanks for the quick assistance.

                        regards, tom lane

Reply | Threaded
Open this post in threaded view
|

Re: 6.4/hppa regression: no LAN or UWSCSI on C360

Miod Vallat
In reply to this post by Miod Vallat

Tested to work (and successfully attach two dino devices) on C240, which
was suffering from the same issue as the OP's C360.

Reply | Threaded
Open this post in threaded view
|

Re: 6.4/hppa regression: no LAN or UWSCSI on C360

Tom Lane
Miod Vallat <[hidden email]> writes:
> Tested to work (and successfully attach two dino devices) on C240, which
> was suffering from the same issue as the OP's C360.

I can confirm successful install and operation on my C360 with the
snapshot of 2019-01-01.  Thanks for fixing that!

I do notice a couple of oddities:

1. Very early in boot, before the kernel copyright message comes out,
I see some worrisome messages:

booting dk6a:/bsd: ...
SPID bits: 0x0, error = -2
pdc_coproc: 0xc0, 0xc0; model 10 rev 1
WARNING: BTLB purge failed
WARNING: cannot block map kernel text
[ using 813416 bytes of bsd ELF symbol table ]
Copyright ...

The dmesg output starts with the "[ using ]" line, so I had to
transcribe these from a screen video.  Things seem to work fine,
so maybe I should disregard these, but if so the all-caps WARNING
is over-enthusiastic.

2. This machine has a soft power button: when you pop it out,
the OS is supposed to do a clean shutdown before letting the
machine power-off.  This works in HPUX, and it seems to work
with the OpenBSD install kernel, but with the regular runtime
kernel the power just drops immediately.  I'd have just said
"oh that's not implemented" had I not seen the install kernel
doing the right thing.

                        regards, tom lane