[Qemu-discuss] how to debug PCI passthrough with PLX (Avago) bridge and TI endpoints

[Jeff Brower]

All-

We're trying to get a PCIe card working with PCI passthrough and it's close but 
not yet.  The card has a PLX (Avago)
PEX 8624 bridge chip and multiple Texas Instruments CPU end-points, and there 
is a second card with PEX 8748 and same
TI endpoints.  Both cards are recognized, all TI devices show up as separate 
PCI devices and the PLX chip shows up as
several separate PCI devices.  Per PCI passthrough instructions, we're only 
attempting to pass through endpoints and
not PLX (bridge) devices.  The card drivers load ok, endpoint config space 
reads look ok, but endpoint writes are not
correct, for example we cannot reliably write values to endpoint config space.

My question is about how to debug this.  Neither Avago nor card vendor tech 
guys think this is their problem; in fact,
they're not even clear on what is PCI passthrough and they don't have VMs set 
up to reproduce the problem.

My first thought is to get a reference card from Avago that has the PEX device 
and a couple PCIe slots for arbitrary
endpoint cards, and test with a simple single-endpoint card that we already 
have working with passthrough, with the
objective being to rule out the PEX bridge devices as causing problems.  If 
that passes, then go the other way, and
find a card with the TI endpoint but no PEX device.

Maybe there are other approaches?  Could there be timing issues?  Some 
parameters known to cause variations in host
vs. passthrough operation that we might try adjusting?  Any advice welcome, 
thanks.

-Jeff