Don't use fdt_* functions, just use OF_getencprop here and fail if the address or size is missing.

Use OF_getencprop(OF_parent(node) for this

This doesn't seem right. The alloc_resource routine uses a local rman for memory, I/O, and IRQs. You should be using rman_release_resource() for all all of those. Ideally you would have a helper routine:

static struct rman *
generic_pcie_rman(int type)
{
   switch (type) {
       case SYS_RES_IOPORT:
           return (&sc->io_rman);
   ....
}

And then in release_resource do something like:

struct rman *rm;

rm = generic_pcie_rman(type);
if (rm != NULL) {
    KASSERT(rman_is_region_manager(res, rm), ("rman mismatch"));
    rman_release_resource(res);
}
return (bus_generic_release_resource(dev, child, type, rid, res));

(I prefer using bus_generic_* over BUS_* directly)

You can then use the rman lookup routine in alloc_resource as well.

Please only set bustag/bushandle in activate_resource(). It needs to work that way for NEW_PCIB. Also, I would add an adjust_resource method. It can be fairly short:

adjust_resource(...)
{
   rm = generic_pcie_rman(rman_get_type(res));
   if (rm != NULL)
       return (rman_adjust_resource(res, ...));
   return (bus_generic_adjust_resource(...));
}

If you want a parent device to set the bus tag/handle for memory, then this probably warrants
a comment. Most of the arm upper layers don't work properly for that either since they try to only set bus tag/handle in alloc and not activate. If you don't want to pass this up, you should be using rman_activate_resource() directly instead.

Regardless I think you to assert that resources all use a local rman (similar to the KASSERT I suggested for release). You should also be using the rman methods to activate/deactivate on SYS_RES_IRQ.

Also, I don't see a deactivate resource method. It should undo the effects of activate (deactivate in rman, etc.).

Also, in general in activate routines you should to the rman activate first since that checks for !RF_SHAREABLE violations, etc. and only do MD work like pmap_mapdev if rman_activate_resource succeeds.

I would just use bus_generic_setup_intr() and bus_generic_teardown_intr() directly in the DEVMETHODS table and remove these.

This doesn't look correct, shouldn't it call into the parent for this?