[PATCH 00/22] Conversion of ARMv4 Gemini to DT and multiplatform

[Linus Walleij]

On Mon, Jan 30, 2017 at 5:35 PM, Hans Ulli Kroll
<ulli.kroll at googlemail.com> wrote:
> On Tue, 24 Jan 2017, Florian Fainelli wrote:
>> On 01/24/2017 12:47 PM, Linus Walleij wrote:

>> >> This is a dual GMAC !
>> >>
>> >> They are two different base address for each GMAC and one shared
>> >> base address for both of them.
>> >> This is all in one address space which remaps the driver.
>> >>
>> >> And to make it worse two interrupt lines, one of them is also shared.
>> >>
>> >> Any ideas to "convert" this into DT ??
>> >> Because I think it's impossible to split both GMAC interfaces ??
>> >
>> > Sounds odd.  Yeah hardware.h lists this:
>> >
>> > #define GEMINI_TOE_BASE         0x60000000
>> > #define GEMINI_GMAC0_BASE       0x6000A000
>> > #define GEMINI_GMAC1_BASE       0x6000E000
>> >
>> > I guess one approach is to make a driver handling all of it in one,
>> > let it ioremap all regions and support one or two GMACs on the same
>> > TOE. More complex solutions is to share the TOE using
>> > syscons regmap mechanism but it looks cumbersome.
>> >
>> > {
>> >    compatible = "cortina,genesis-gmac-ethernet";
>> >    reg = <0x60000000 0xa000>, <0x6000a000 0x4000>, <0x6000e000 0x4000>;
>> >    interrupts = <1 IRQ_TYPE_LEVEL_HIGH>, <2 IRQ_TYPE_LEVEL_HIGH>;
>> > }
>> >
>> > Or something like that?
>
> For addresses, base is here, for both interfaces
> #define GEMINI_TOE_BASE                 0x60000000
> and now for each interface, offset to the above
> code is from the headerfile of the openwrt driver.
> #define TOE_GMAC_DMA_BASE(x)            (0x8000 + 0x4000 * (x))
> #define TOE_GMAC_BASE(x)                (0xA000 + 0x4000 * (x))
>
> maybe some other registers to
> #define TOE_NONTOE_QUE_HDR_BASE         0x2000
> #define TOE_TOE_QUE_HDR_BASE            0x3000
> #define TOE_V_BIT_BASE                  0x4000
> #define TOE_A_BIT_BASE                  0x6000
>
> AFAICR
> The TOE part was "invented" for the gigabit interface.
> They need to speed up the NIC because without TCP offloading you will get
> only fast ethernet speed on a gigabit NIC.

I was suspecting that this ethernet is related to the Faraday ethernet driver
that was merged in drivers/net/ethernet/faraday since so many other
Gemini parts seems to be coming essentially directly from Faraday.

I looked through the header files etc there but the registers seem quite
different ... :/

Still it is using the same "GMAC" naming and I strongly suspect this is
an older Faraday IP that should go into the same driver folder.

Cc:ing some Faraday and Andestech emails to see if we can get some
clarity in this. It would be nice if someone could atleast say "oh it's this
IP FTxxx from Faraday" so we know what to name it.

Yours,
Linus Walleij