Hi,

On Tue, Sep 16, 2014 at 11:15:43AM -0700, Jack Pham wrote:

Hi Andy,

On Fri, Sep 12, 2014 at 02:28:06PM -0500, Andy Gross wrote:

quoted

+Example device nodes:
+
+		hs_phy: phy@100f8800 {
+			compatible = "qcom,dwc3-hs-usb-phy";
+			reg = <0x100f8800 0x30>;

Just wanted to point out that in our downstream code, the glue
device/driver, i.e. "qcom,dwc3", does claim an entire register region
that encompasses the same registers used by these PHYs. I noticed you're
not adding a reg resource to the glue node below in this patchset. In
order to allow multiple devices to use the overlapping regions, we avoid
calling devm_ioremap_resource() and instead call devm_ioremap_nocache()
directly which bypasses the request_mem_region() call, which I don't
think is an entirely correct thing to do.

On the other hand I'm trying to think of use cases on our other SOCs
(MSM8974, APQ8084) where the glue actually would need access to the same
or adjacent IO registers that couldn't just be handled directly by these
PHY drivers. Should we accommodate for the potential of overlapping
regions or should we just hold the line here and maybe only expose with
finer granularity the registers that will actually be needed by the PHYs
& glue respectively?

I prefer to keep things separated. Glue only needs to access whatever
belongs to glue. If a register related to PHY settings, it should be in
the PHY address space.

cheers

-- 
balbi