On Mon, Mar 03, 2014 at 04:38:10PM -0800, Tim Harvey wrote:

On Mon, Mar 3, 2014 at 3:37 PM, Jason Gunthorpe
[off-list ref] wrote:

quoted

On Mon, Mar 03, 2014 at 11:59:51AM -0800, Tim Harvey wrote:

quoted

A pci fixup for the XIO2001 could at least be placed in
arch/arm/mach-imx/mach-imx6q.c with a check for
of_machine_is_compatible("gw, ventana").  This is currently done in
order to handle GPIO on the PEX890x bridge which is used as PERST# for
downstream slots (see
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/arch/arm/mach-imx/mach-imx6q.c#n87).

That is an odd looking thing .. PERST# of a subordinate port should be
tied to the PCI_BRIDGE_CTL_BUS_RESET bit in the port's bridge
configuration space - is more configuration of the PLX chip required?

PCI_BRIDGE_CTL_BUS_RESET does cause a hot reset on the corresponding
downstream link however the minPCIe card slot has a PERST# pin that is
for a functional reset to the card.  

Ah, I'm getting my PCI versions mixed up, that all sounds right to me
then..

quoted

Also, be aware that PCI specifies a minimum time after
reset-deassertion before accessing the bus, and I don't see a sleep...

there is an msleep(100) there after the de-assertion.

.. and I'm blind, looks good.

Ok - that all makes sense and hopefully when I get the legacy PCI
interrupt mapping via DT working for IMX6 I can prove it out.

What did you think of my two ideas above to resolve this issue on our
add-in card where the bus-topology isn't known ahead of time:

 1. bootloader builds out DT with proper interrupt-map's to handle the
mis-map after enumerating the bus to understand the toplogy.

This is how the OF stuff for PCI has historically been used, so it is
a nice OS-agnostic way to implement this work around, particularly if
you have some kind of daughter card specific information you can rely
on (eg a SPD that could indicate the problematic PCB revision)

If you do that you could also hoist the required fixup for your switch
into firmware too.

 2. a pci fixup for xio2001 that further qualifies the DT machine and
the 'depth' of the xio2001 (ie make sure its the first xio2001 bridge
on the bus) that replaces the swizzle function.  This would be in
arch/arm/mach-imx/mach-imx6q.c

You can probably make this work, but given there is already a DT based
fix I'm not sure if people will like a .c version, especially if it
requires more single-use 'general purpose' interfaces from the core
code...

Jason