Re: PCIE-to-PCI bridge doesn't pass mem resources downstream [Re: yenta cardbus problem]

[Bjorn Helgaas <bhelgaas@google.com>]

On Wed, Apr 27, 2011 at 4:33 AM, Germán Sanchis <eaglecros@gmail.com> wrote:
> 2011/4/25 Dominik Brodowski <linux@dominikbrodowski.net>:
>> Hey,
>>
>> thanks for the additional debug information. It seems to me to be not a
>> bug with the yenta driver, but with the parent PCI bridge instead.
>> Therefore, I've added Jesse and the linux-pci list as recipients. Germán's
>> problem relates to Ubuntu's 2.6.35-derived kernel; lspci snippets follow.
>>
>> Let's look at the (grand-)parent bridge: It offers some, if not much I/O and
>> memory resources for its childs to be used:
>>
>> 00:1c.5 PCI bridge: Intel Corporation 82801I (ICH9 Family) PCI Express Port 6 (rev 03) (prog-if 00 [Normal decode])
>>        Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx+
>>        Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
>>        Latency: 0
>>        Bus: primary=00, secondary=04, subordinate=09, sec-latency=0
>>        I/O behind bridge: 00001000-00001fff
>>        Memory behind bridge: d6100000-d70fffff
>>        Prefetchable memory behind bridge: 00000000d5100000-00000000d60fffff
>>
>> The PCI bridge, however, does only pass the I/O resources downstream, but
>> _no_ memory at all.
>>
>> 04:00.0 PCI bridge: Texas Instruments XIO2000(A)/XIO2200(A) PCI Express-to-PCI Bridge (rev 03) (prog-if 00 [Normal decode])
>>        Control: I/O+ Mem- BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
>>        Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
>>        Latency: 0
>>        Bus: primary=04, secondary=05, subordinate=09, sec-latency=0
>>        I/O behind bridge: 00001000-00001fff
>>        Memory behind bridge: fff00000-000fffff
>>
>> The CardBus bridge then has I/O resources to use, but cannot enable memory
>> resources:
>>
>> 05:00.0 CardBus bridge: Texas Instruments PCIxx12 Cardbus Controller
>>        Memory window 0: 00000000-00000000 [disabled] (prefetchable)
>>        Memory window 1: 00000000-00000000 [disabled] (prefetchable)
>>        I/O window 0: 00001000-000010ff
>>        I/O window 1: 00001400-000014ff
>>
>> So my question to the PCI folks: why does the PCI bridge fail to pass memory
>> regions downstream, and assign them properly to the CardBus bridge?

The BIOS assigned [mem 0xd6100000-0xd70fffff] (16MB) and [mem
0xd5100000-0xd60fffff 64bit pref] (16MB) to bus 04, but left the
04:00.0 bridge windows to bus 05 disabled.  Linux only enables them if
we find a downstream device that needs resources.  In this case, there
*is* a downstream CardBus bridge (05:00.0) that would like resources.
By default, pci_bus_size_cardbus() requests 64MB for each CardBus
memory window, but there's only 16MB each (prefetchable and
non-prefetchable) available.

I don't think your audio card needs anywhere near 64MB of memory
space.  Can you try booting with "pci=cbmemsize=8M"?

Bjorn