ATI SB400 PCI bridge fallback to substractive decode

[Thierry Deval]

Hi,

When I put a CF to PCCard adapter (not CardBus) in my laptop to work on a CF 
boot image, I was surprised by this kernel message :

** wdc2 at pcmcia0 function 0 "SanDisk, SDP, 5/3 0.6": can't handle card info

And the card was not working at all, as weren't any other PCCard I tried 
afterwards.

After digging and enabling as much debugging as I could, I found that the 
cardbus bridge (TI PCI7xx1 CardBus) couldn't allocate any IO or mem range 
for the cards.

Digging deeper to understand how the allocation should work, I noticed 
a comment in dev/pci/ppb.c talking about handling substractive (or is it 
really 'subtractive' as in the comment ? ) decode scheme for more than the 
Intel 82801 PCI bridge.
So, as a test, I included the ATI SB400 PCI bridge (behind which the TI 
PCI7xx1 CardBus bridge was lying) to the substractive decode treatment. 
I was rewarded by a working CF card.

** ppb2 at pci0 dev 20 function 4 "ATI SB400 PCI" rev 0x00
** pci3 at ppb2 bus 6
** cbb0 at pci3 dev 9 function 0 "TI PCI7XX1 CardBus" rev 0x00: apic 1 int 23
** cardslot0 at cbb0 slot 0 flags 0
** cardbus0 at cardslot0: bus 7 device 0 cacheline 0x8, lattimer 0x20
** pcmcia0 at cardslot0
** wdc2 at pcmcia0 function 0 "SanDisk, SDP, 5/3 0.6" port 0xa000/16
** wd1 at wdc2 channel 0 drive 0: <SanDisk SDCFB-64>
** wd1: 1-sector PIO, LBA, 61MB, 125440 sectors
** wd1(wdc2:0:0): using BIOS timings
** wd1 detached
** wdc2 detached

I don't know if anybody else have seen the same problem but I would be glad 
to have a confirmation of the fix.

Here's the diff:

Index: ppb.c
===================================================================
RCS file: /cvs/src/sys/dev/pci/ppb.c,v
retrieving revision 1.58
diff -u -p -u -p -r1.58 ppb.c
--- ppb.c       12 Jul 2014 18:48:52 -0000      1.58
+++ ppb.c       14 Sep 2014 17:00:40 -0000
@@ -286,9 +286,11 @@ ppbattach(struct device *parent, struct
         * XXX We probably should handle subtractive decode bridges
         * in general.
         */
-       if (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_INTEL &&
+       if ((PCI_VENDOR(pa->pa_id) == PCI_VENDOR_INTEL &&
            (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82801BA_HPB ||
-           PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82801BAM_HPB)) {
+            PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82801BAM_HPB)) ||
+           (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_ATI &&
+            PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_ATI_SB400_PCI)){
                if (sc->sc_ioex == NULL)
                        sc->sc_ioex = pa->pa_ioex;
                if (sc->sc_memex == NULL)

I hope this could go in, at least as a first step.
I plan on looking further about what this substractive decode means, and 
if there could be a generic way of enabling it where supported...

Thierry