cardbus/pcmcia/pci bridge problems?

cardbus/pcmcia/pci bridge problems?

[Paul Davis]

I recently wrote the ALSA driver for the RME Hammerfall DSP, a
high-end audio interface. The H-DSP comes in a variety of
configurations, one of them involving a "pure" PCI interface, the
other a PCMCIA card. The PCI interface works fine. Users with the
PCMCIA card report major problems, the most representative of which
are:

Case One (failure to make iomem available):
=========
>May 16 09:54:57 satellite kernel: cs: cb_alloc(bus 3): vendor 0x10ee, device 
>0x3fc5
>May 16 09:54:57 satellite kernel: PCI: Failed to allocate resource 
>0(d2400000-d20fffff) for 03:00.0
>May 16 09:54:57 satellite kernel: PCI: Enabling device 03:00.0 (0080 -> 0082)
>May 16 09:54:57 satellite cardmgr[258]: socket 0: CardBus hotplug device


Case Two (failure to allocate IRQ):
=========
>May 16 15:15:32 badass-bukvic kernel: PCI: Found IRQ 5 for device
>02:03.0
>May 16 15:15:33 badass-bukvic kernel: Hammerfall memory allocator:
>buffers allocated for 1 cards
>May 16 15:15:33 badass-bukvic kernel: PCI: No IRQ known for interrupt
>pin A of device . Please try using pci=biosirq.
>May 16 15:15:33 badass-bukvic kernel: PCI: Setting latency timer of
>device  to 64
>May 16 15:15:33 badass-bukvic kernel: ALSA
>../../alsa-kernel/pci/rme9652/hdsp.c:2984: unable to grab memory region
>0x0-0x1447

In both cases, the error comes from cardbus/pcmcia/pci code that runs
"below" and/or before the device driver itself.

The PCI profile of the device looks like:

----------------------------------------------------------------------
03:00.0 Multimedia audio controller: Xilinx, Inc.: Unknown device 3fc5 (rev 
0a)
        Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- 
Stepping+ SERR- FastB2B-
        Status: Cap- 66Mhz- UDF- FastB2B- ParErr- DEVSEL=medium >TAbort- 
<TAbort- <MAbort- >SERR- <PERR-
        Interrupt: pin A routed to IRQ 10
        Region 0: [virtual] Memory at d2400000 (32-bit, non-prefetchable)
----------------------------------------------------------------------

It appears to me that there may be some problems with the current
cardbus/pci bridge code in 2.4.X. can anyone with any involvement in
this code fill me in on how well it is thought to be working, and what
might give rise to either of the error cases shown above?

--p

Re: cardbus/pcmcia/pci bridge problems?

[Alan Cox]

> >May 16 15:15:32 badass-bukvic kernel: PCI: Found IRQ 5 for device
> >02:03.0
> >May 16 15:15:33 badass-bukvic kernel: Hammerfall memory allocator:
> >buffers allocated for 1 cards
> >May 16 15:15:33 badass-bukvic kernel: PCI: No IRQ known for interrupt
> >pin A of device . Please try using pci=biosirq.

The $PIR table in the BIOS provided no useful information on what IRQ
to use, or we didn't know the IRQ router concerned to set it up.

Alan

Re: cardbus/pcmcia/pci bridge problems?

[Paul Davis]

>> >May 16 15:15:32 badass-bukvic kernel: PCI: Found IRQ 5 for device
>> >02:03.0
>> >May 16 15:15:33 badass-bukvic kernel: Hammerfall memory allocator:
>> >buffers allocated for 1 cards
>> >May 16 15:15:33 badass-bukvic kernel: PCI: No IRQ known for interrupt
>> >pin A of device . Please try using pci=biosirq.
>
>The $PIR table in the BIOS provided no useful information on what IRQ
>to use, or we didn't know the IRQ router concerned to set it up.

alan: let me check what you're saying: despite the fact that lspci -vv
shows the relevant information, there is still a problem? 

the problem is either in the BIOS or its an issue with the kernel not
knowing which IRQ router to use when setting up the interrupt? sorry
to be repetitive, but i'm just trying to check that i fully understand
this.

Re: cardbus/pcmcia/pci bridge problems?

[Alan Cox]

> >The $PIR table in the BIOS provided no useful information on what IRQ
> >to use, or we didn't know the IRQ router concerned to set it up.
> 
> alan: let me check what you're saying: despite the fact that lspci -vv
> shows the relevant information, there is still a problem? 
> 
> the problem is either in the BIOS or its an issue with the kernel not
> knowing which IRQ router to use when setting up the interrupt? sorry
> to be repetitive, but i'm just trying to check that i fully understand
> this.

There are two stages to IRQ routing

1.	Knowing how INTA/B/C/D on the slot are wired to the INTA-D lines
	on the PCI bus (its not 1-1)
2.	Knowing how to drive the PCI irq routing hardware to route a
	line to an ISA IRQ or to work out which it already goes to