[9fans] Virtual PC 6 PCI device probe failure

[9fans] Virtual PC 6 PCI device probe failure

[Lyndon Nerenberg]

I've done some more fiddling with the VPC 6 install. Video is working
fine, but the kernel does not see any PCI (or PNP) devices during the
bus scan. This explains why the virtual 21140 ethernet isn't getting
configured.

Was there any magic required to get the kernel to work under vmware?

Meanwhile I'm going to install a FreeBSD and see what it thinks about
PCI devices.

--lyndon

Re: [9fans] Virtual PC 6 PCI device probe failure

[jmk]

On Tue Jun 24 23:48:27 EDT 2003, lyndon@orthanc.ab.ca wrote:
> I've done some more fiddling with the VPC 6 install. Video is working
> fine, but the kernel does not see any PCI (or PNP) devices during the
> bus scan. This explains why the virtual 21140 ethernet isn't getting
> configured.
>
> Was there any magic required to get the kernel to work under vmware?
>
> Meanwhile I'm going to install a FreeBSD and see what it thinks about
> PCI devices.
>
> --lyndon

This came up a year or so ago, so maybe things have changed, but at that time
the VirtualPC docs claimed it emulated an Intel Triton chipset, but it was discovered
it didn't emulate the PCI config registers in that chipset. This matters on Plan9
because it does raw accesses to probe the PCI space whereas other operating systems
do BIOS calls to obtain that information.

Re: [9fans] Virtual PC 6 PCI device probe failure

[Lyndon Nerenberg]

On Wednesday, June 25, 2003, at 09:06  AM, jmk@plan9.bell-labs.com
wrote:

> This came up a year or so ago, so maybe things have changed, but at
> that time
> the VirtualPC docs claimed it emulated an Intel Triton chipset, but it
> was discovered
> it didn't emulate the PCI config registers in that chipset.

Technical documentation on the emulated hardware is sparse to
non-existent from what I can see. I'm loath to contact their technical
support folks, since that's guaranteed to degenerate into a "we don't
support the Plan9 release of Windows" conversation. First let's see if
the FreeBSD kernel has any luck finding bits ...

--lyndon

Re: [9fans] Virtual PC 6 PCI device probe failure

[Lyndon Nerenberg]

> This matters on Plan9
> because it does raw accesses to probe the PCI space whereas other
> operating systems
> do BIOS calls to obtain that information.

And FreeBSD finds the hardware just fine, using the BIOS interface.

I now vaguely remember a discussion here a long while back about why
the BIOS wasn't used for this. Was it to do with broken BIOS code?

--lyndon

Re: [9fans] Virtual PC 6 PCI device probe failure

[David Presotto]

more to do with lack of neurons/time on our part.

Re: [9fans] Virtual PC 6 PCI device probe failure

[andrey mirtchovski]

Ron has the greatest desire to fix this, but keeps getting distracted..
Maybe someday...

I don't know enough low-level hardware details to do it myself (they don't
teach these things in school anymore).

It comes down to configuring IRQs on peripheral devices using the BIOS (or
whatever's there) instead of trusting what the device thinks it has for an
IRQ.

andrey

On Wed, 25 Jun 2003, David Presotto wrote:

> more to do with lack of neurons/time on our part.
>

Re: [9fans] Virtual PC 6 PCI device probe failure

[jmk]

On Wed Jun 25 16:21:29 EDT 2003, mirtchov@cpsc.ucalgary.ca wrote:
> Ron has the greatest desire to fix this, but keeps getting distracted..
> Maybe someday...
>
> I don't know enough low-level hardware details to do it myself (they don't
> teach these things in school anymore).
>
> It comes down to configuring IRQs on peripheral devices using the BIOS (or
> whatever's there) instead of trusting what the device thinks it has for an
> IRQ.
>
> andrey
>
> On Wed, 25 Jun 2003, David Presotto wrote:
>
> > more to do with lack of neurons/time on our part.
> >

None of this really hits the target. The Plan9 code doesn't use the BIOS calls
because it is sometimes used on hardware which doesn't have a BIOS, it just does
the work itself.

Re: [9fans] Virtual PC 6 PCI device probe failure

[Lyndon Nerenberg]

On Wednesday, June 25, 2003, at 02:10  PM, David Presotto wrote:

> more to do with lack of neurons/time on our part.

Okay, I'll dig out my PCI spec's and get to work ...

--lyndon

Re: [9fans] Virtual PC 6 PCI device probe failure

[ron minnich]

On Wed, 25 Jun 2003, Lyndon Nerenberg wrote:

> > This matters on Plan9
> > because it does raw accesses to probe the PCI space whereas other
> > operating systems
> > do BIOS calls to obtain that information.
>
> And FreeBSD finds the hardware just fine, using the BIOS interface.
>
> I now vaguely remember a discussion here a long while back about why
> the BIOS wasn't used for this. Was it to do with broken BIOS code?

There are lots of reasons not to use the bios, related to broken code,
broken tables, ... the list goes on. The one I care about is I don't
support those bios calls :-)

I prefer what Plan 9 does in this case to what FreeBSD does. More
portable.

ron

Re: [9fans] Virtual PC 6 PCI device probe failure

[Lyndon Nerenberg]

> There are lots of reasons not to use the bios, related to broken code,
> broken tables, ... the list goes on. The one I care about is I don't
> support those bios calls :-)
>
> I prefer what Plan 9 does in this case to what FreeBSD does. More
> portable.

I disagree with the portability argument. FreeBSD boots on every PC I
have, including VPC. Plan9 doesn't boot on VPC.

The BIOS may be ugly, but it works. Since I have to get Plan9 running
under VPC, you can guess which argument wins on this side of the
keyboard.

--lyndon

Re: [9fans] Virtual PC 6 PCI device probe failure

[Christopher Nielsen]

On Wed, Jun 25, 2003 at 02:39:51PM -0600, Lyndon Nerenberg wrote:
>
> I disagree with the portability argument. FreeBSD boots on every PC I
> have, including VPC. Plan9 doesn't boot on VPC.
>
> The BIOS may be ugly, but it works. Since I have to get Plan9 running
> under VPC, you can guess which argument wins on this side of the
> keyboard.

As jmk pointed out, it doesn't work on anything that
doesn't have a bios. Not everything in the world is
a PC.

--
Christopher Nielsen
"They who can give up essential liberty for temporary
safety, deserve neither liberty nor safety." --Benjamin Franklin

Re: [9fans] Virtual PC 6 PCI device probe failure

[boyd, rounin]

> I disagree with the portability argument. FreeBSD boots on every PC I
> have, including VPC. Plan9 doesn't boot on VPC.

yes this is the 'army of programmers' argument -- it just doesn't hold up.

sure, if you spend enough time and effort you'll be able to boot off every
known piece of horrible PC hardware on the face of the planet and that
quickly leads to the 2M SLOC which is the linix kernel.

Re: [9fans] Virtual PC 6 PCI device probe failure

[ron minnich]

On Wed, 25 Jun 2003, Lyndon Nerenberg wrote:

> I disagree with the portability argument. FreeBSD boots on every PC I
> have, including VPC. Plan9 doesn't boot on VPC.

freebsd won't boot on about 1500 of the PCs I have.


ron

Re: [9fans] Virtual PC 6 PCI device probe failure

[Lyndon Nerenberg]

On Wednesday, June 25, 2003, at 03:04  PM, ron minnich wrote:
> freebsd won't boot on about 1500 of the PCs I have.

Point taken. But it still doesn't boot on my gear, so I'll fix that. If
I can do it such that other hardware can take advantage if it, well,
good on it.

--lyndon

Re: [9fans] Virtual PC 6 PCI device probe failure

[jmk]

On Wednesday, June 25 16:33:28 EDT 2003, Lyndon Nerenberg wrote:
>On Wednesday, June 25, 2003, at 02:10  PM, David Presotto wrote:
>
>> more to do with lack of neurons/time on our part.
>
>Okay, I'll dig out my PCI spec's and get to work ...
>
>--lyndon

That won't help if the emulated hardware doesn't emulate the registers
needed to access the PCI config  space.

The best best is to hack the ether driver to assume a device is there
if given a port, irq, etc. in the plan9.ini instead of looking for the
devices in the PCI space, i.e. make the driver look like an ISA device
driver.

Re: [9fans] Virtual PC 6 PCI device probe failure

[Lyndon Nerenberg]

On Wednesday, June 25, 2003, at 02:41  PM, jmk@plan9.bell-labs.com
wrote:
> The best best is to hack the ether driver to assume a device is there
> if given a port, irq, etc. in the plan9.ini instead of looking for the
> devices in the PCI space, i.e. make the driver look like an ISA device
> driver.

Yes, I was going to try that first. Failing that I'll teach the kernel
to use the BIOS to probe (configurable from plan9.ini so the existing
code can stay intact for non-BIOSed hardware).

--lyndon