Re: via-rhine weirdness with via kt8235 Southbridge

Re: via-rhine weirdness with via kt8235 Southbridge

[Christian Guggenberger]

> On Wed, 23 Oct 2002 15:49:31 +0200, Christian Guggenberger wrote:
>> This concerns both 2.4 and 2.5 kernels  (testet with 2.4.20pre*aa series,
>> and with 2.5.43, 2.5.44 and 2.5.44-ac1):
>> 
>> When I enable APIC in the Bios, the via-rhine module will insert
>> properly, but I won't get a link... With APIC disabled, link is ok.  Ok,
>> this could be caused by buggy bios, so I'll try again, when a new
>> biosversion is available.
> 
> Yeah, it seems there's a problem with IO-APICs. I currently don't have a
> machine with IO-APIC for testing, though, so...
> 

A new Biosversion is installed on my mobo now, but that APIC problem is still
there.
Are there some dumps I could post to get some light on that topic?

Maybe some outputs of via-diag, mii-diag, lspci, dmesg ...?
If they could help, what options should I pass to mii-diag and via-diag ?


Christian

Re: via-rhine weirdness with via kt8235 Southbridge

[Erik Hensema]

Christian Guggenberger (christian.guggenberger@physik.uni-regensburg.de) wrote:
>> On Wed, 23 Oct 2002 15:49:31 +0200, Christian Guggenberger wrote:
>>> This concerns both 2.4 and 2.5 kernels  (testet with 2.4.20pre*aa series,
>>> and with 2.5.43, 2.5.44 and 2.5.44-ac1):
>>> 
>>> When I enable APIC in the Bios, the via-rhine module will insert
>>> properly, but I won't get a link... With APIC disabled, link is ok.  Ok,
>>> this could be caused by buggy bios, so I'll try again, when a new
>>> biosversion is available.
>> 
>> Yeah, it seems there's a problem with IO-APICs. I currently don't have a
>> machine with IO-APIC for testing, though, so...
>> 
> 
> A new Biosversion is installed on my mobo now, but that APIC problem is still
> there.
> Are there some dumps I could post to get some light on that topic?
> 
> Maybe some outputs of via-diag, mii-diag, lspci, dmesg ...?
> If they could help, what options should I pass to mii-diag and via-diag ?

I've had the exact same problems with via-rhine and APIC. I donated (well,
actually sold, but not voluntarily ;-)) my card to Roger Luethi so he can
look into the bug. And yes, it's an actual PCI card, which seems to be rare
with via-rhines.

-- 
Erik Hensema (erik@hensema.net)

Yet another IO-APIC problem (was Re: via-rhine weirdness with via kt8235 Southbridge)

[Jeff Garzik]

Christian Guggenberger wrote:

> >On Wed, 23 Oct 2002 15:49:31 +0200, Christian Guggenberger wrote:
> >
> >>This concerns both 2.4 and 2.5 kernels  (testet with 2.4.20pre*aa 
> series,
> >>and with 2.5.43, 2.5.44 and 2.5.44-ac1):
> >>
> >>When I enable APIC in the Bios, the via-rhine module will insert
> >>properly, but I won't get a link... With APIC disabled, link is ok.  Ok,
> >>this could be caused by buggy bios, so I'll try again, when a new
> >>biosversion is available.
> >
> >Yeah, it seems there's a problem with IO-APICs. I currently don't have a
> >machine with IO-APIC for testing, though, so...
> >
>
>
> A new Biosversion is installed on my mobo now, but that APIC problem 
> is still
> there.
> Are there some dumps I could post to get some light on that topic?
>
> Maybe some outputs of via-diag, mii-diag, lspci, dmesg ...?
> If they could help, what options should I pass to mii-diag and via-diag ?



Yet Another IO-APIC problem.  From what I see quoted above, it has 
nothing at all to do with the via-rhine, and everything to do with IO-APIC.

Almost universally, in 2.4 and 2.5, running IO-APIC on uniprocessor is 
badly broken, and causes all sorts of bogus bug reports for unrelated 
subsystems.  Sometimes IO-APIC is busted even on SMP.

Workaround solutions:  (1) boot with "noapic" on command line, (2) 
disable IO-APIC in kernel config (if uniprocessor), and/or (3) make sure 
that "MP 1.1" is listed in BIOS setup instead of "MP 1.4".

Man, this must be the fourth or fifth bug report I've gotten in the past 
week or so, where the IO-APIC can be blamed.  ;-(  In all recent cases 
it has been uniprocessor IOAPIC, not SMP IOAPIC, that was problematic. 
IMO we should just take out UP IOAPIC support in the kernel, or put a 
big fat warning in the kernel config _and_ at boot...

	Jeff

Re: Yet another IO-APIC problem (was Re: via-rhine weirdness with viakt8235 Southbridge)

[Andrew Morton]

Jeff Garzik wrote:
> 
> ...
> IMO we should just take out UP IOAPIC support in the kernel, or put a
> big fat warning in the kernel config _and_ at boot...
> 

It would be nice to get it working, because oprofile needs it.

(Well, oprofile can use the rtc, but then it doesn't profile
ints-off code)

Re: Yet another IO-APIC problem (was Re: via-rhine weirdness with viakt8235 Southbridge)

[Christian Guggenberger]

On 15 Nov 2002 02:04:31 Andrew Morton wrote:
> Jeff Garzik wrote:
> >
> > ...
> > IMO we should just take out UP IOAPIC support in the kernel, or put a
> > big fat warning in the kernel config _and_ at boot...
> >
> 
> It would be nice to get it working, because oprofile needs it.
> 
> (Well, oprofile can use the rtc, but then it doesn't profile
> ints-off code)
> 

so, is there anything I could do for you in that IO-APIC case ? Especially in 
my via-rhine case? As I mentioned in my previous mail, would some via-diag, 
mii-diag outputs (what options) be helpful for you?

Christian

Re: Yet another IO-APIC problem (was Re: via-rhine weirdness with viakt8235 Southbridge)

[John Levon]

On Thu, Nov 14, 2002 at 05:04:31PM -0800, Andrew Morton wrote:

> It would be nice to get it working, because oprofile needs it.

That's the local APIC (CONFIG_X86_UP_APIC) not IO-APIC
(CONFIG_X86_UP_IOAPIC). They should be separate, and it's been a while
since I've tested it, but CONFIG_X86_UP_IOAPIC=n should still work OK

regards
john
-- 
Khendon's Law: If the same point is made twice by the same person,
the thread is over.

Re: Yet another IO-APIC problem (was Re: via-rhine weirdness with viakt8235 Southbridge)

[Jeff Garzik]

Andrew Morton wrote:

> Jeff Garzik wrote:
>
> >...
> >IMO we should just take out UP IOAPIC support in the kernel, or put a
> >big fat warning in the kernel config _and_ at boot...
> >
>
>
> It would be nice to get it working, because oprofile needs it.
>
> (Well, oprofile can use the rtc, but then it doesn't profile
> ints-off code)



I don't see it happening, when uniprocessor mobo vendors (a) don't wire 
it up, or (b) put buggy data in their MP tables...   :(

	Jeff

Re: Yet another IO-APIC problem (was Re: via-rhine weirdness withviakt8235 Southbridge)

[Andrew Morton]

Jeff Garzik wrote:
> 
> Andrew Morton wrote:
> 
> > Jeff Garzik wrote:
> >
> > >...
> > >IMO we should just take out UP IOAPIC support in the kernel, or put a
> > >big fat warning in the kernel config _and_ at boot...
> > >
> >
> >
> > It would be nice to get it working, because oprofile needs it.
> >
> > (Well, oprofile can use the rtc, but then it doesn't profile
> > ints-off code)
> 
> I don't see it happening, when uniprocessor mobo vendors (a) don't wire
> it up, or (b) put buggy data in their MP tables...   :(
> 

OK.  Actually, rtc-based profiling gives perfectly grand results for
profiling userspace, and those people who want to profile their
interrupt handlers presumably know where to buy a decent motherboard.