Re: Problems with fxp0 on T30 with 5.1-RELEASE

On Thu, Jul 17, 2003 at 09:58:00AM -0600, M. Warner Losh wrote:
> In message: <20030717082918.GA21592_at_speedy.unibe.ch>
>             Tobias Roth <roth_at_iam.unibe.ch> writes: this is caused by
> : an irq conflict. the bug was introduced some time between 5.0 and
> : 5.1.  i have no idea how to solve this, maybe someone else can help
> : here. maybe the ibm ps2 tool offers some help.
> 
> details?

kevin oberman suggested the use of the ps2 utility for thinkpads. i
did not have the chance to try it out myself.

> : another workaround is to free an irq. for me, disabling the pcmcia
> : stuff in the kernel config helped. others reported that disabling
> : the serial port helped for them.
> 
> interesting.

shortly after i discovered this problem, my TP started to have immense
problems with the summer heat and started crashing during buildworlds.
that's why i didn't look into the irq problem any further at this time.

now, i just got my TP back from ibm, and while they did not really fix
the heat problem (i will send it back to them next week), i got around
to do some irq tests.

irqs 0,1,8,12,13,14,15 are set and no subject of manual change.
irq 2, i have no idea about.
irq 3 and 4 are the infrared device and the serial port.
irq 7 is the parallel port
irq 6 can be freed by disabling legacy floppy support in the bios.
however it does not seem be taken by a different device when freed.

this leaves us with irqs 5,9,10,11 to distribute in the pci section
of the bios. after some playing around, i set INT[A-H] PCI IRQ to
5,9,10,11,11,9,10,11. I do not know whether the order matters, but
the distribution does. like this, every irq is assigned to two devices:

5->uhic0 and cbb0, 9->cbb1 and pcm0, 10->uhic2 and wi0, 11->uhic1 and fxp0.

the only drawback i encountered with these settings was a short hang
during boot when probing uhic2. it did not hang there before i started
switching around irqs, but i do not know what is wrong here.

other than that, everything works now.

> : this has to be fixed before 5.2, imho. it renders a default install
> : on thinkpads useless. note that this does NOT happen on all thinkpad
> : systems, i didn't figure out what makes up the difference.
> 
> sounds like a bug that needs to be fixed, but the details are so vague
> as to make that impossible.  chances are very good that someone with a
> clue (like me) will need a machine that fails to fix it.

the only odd thing is that 5.0 did not have this bug, while 5.1 has
it. if it needs fixing is up to you, i am now happy with the current
solution. but then, the default bios settings are so stupid that it may
not be feasable to do anything about it.

hope that helps, t.