Re: [PATCHv5 1/3] dts: socfpga: Add bindings for Altera SoC SDRAM controller

From: Steffen Trumtrar
Date: Wed May 28 2014 - 03:02:33 EST


On Tue, May 27, 2014 at 03:57:47PM -0500, Alan Tull wrote:
> On Tue, May 27, 2014 at 2:42 PM, Steffen Trumtrar
> <s.trumtrar@xxxxxxxxxxxxxx> wrote:
> > On Tue, May 27, 2014 at 02:12:17PM -0500, Alan Tull wrote:
> >> On Tue, May 27, 2014 at 2:11 AM, Steffen Trumtrar
> >> <s.trumtrar@xxxxxxxxxxxxxx> wrote:
> >> > On Wed, May 21, 2014 at 10:38:34AM -0500, Thor Thayer wrote:
> >> >> On Tue, May 20, 2014 at 9:44 AM, Steffen Trumtrar
> >> >> <s.trumtrar@xxxxxxxxxxxxxx> wrote:
> >> >> > Hi!
> >> >> >
> >> >> > On Tue, May 20, 2014 at 09:31:06AM -0500, Alan Tull wrote:
> >> >> >> On Mon, May 19, 2014 at 2:37 PM, Thor Thayer <tthayer.linux@xxxxxxxxx> wrote:
> >> >> >>
> >> >> >> >>> >> diff --git a/Documentation/devicetree/bindings/arm/altera/socfpga-sdram.txt b/Documentation/devicetree/bindings/arm/altera/socfpga-sdram.txt
> >> >> >> >>> >> new file mode 100644
> >> >> >> >>> >> index 0000000..8f8746b
> >> >> >> >>> >> --- /dev/null
> >> >> >> >>> >> +++ b/Documentation/devicetree/bindings/arm/altera/socfpga-sdram.txt
> >> >> >> >>> >> @@ -0,0 +1,11 @@
> >> >> >> >>> >> +Altera SOCFPGA SDRAM Controller
> >> >> >> >>> >> +
> >> >> >> >>> >> +Required properties:
> >> >> >> >>> >> +- compatible : "altr,sdr-ctl";
> >> >> >> >>> >> +- reg : Should contain 1 register ranges(address and length)
> >> >> >> >>> >> +
> >> >> >> >>> >> +Example:
> >> >> >> >>> >> + sdrctl@ffc25000 {
> >> >> >> >>> >> + compatible = "altr,sdr-ctl";
> >> >> >> >>> >> + reg = <0xffc25000 0x1000>;
> >> >> >> >>> >> + };
> >> >> >> >>> >> diff --git a/arch/arm/boot/dts/socfpga.dtsi b/arch/arm/boot/dts/socfpga.dtsi
> >> >> >> >>> >> index df43702..6ce912e 100644
> >> >> >> >>> >> --- a/arch/arm/boot/dts/socfpga.dtsi
> >> >> >> >>> >> +++ b/arch/arm/boot/dts/socfpga.dtsi
> >> >> >> >>> >> @@ -676,6 +676,11 @@
> >> >> >> >>> >> clocks = <&l4_sp_clk>;
> >> >> >> >>> >> };
> >> >> >> >>> >>
> >> >> >> >>> >> + sdrctl@ffc25000 {
> >> >> >> >>> >> + compatible = "altr,sdr-ctl", "syscon";
> >> >> >> >>> > ^^^^^^^^^^
> >> >> >> >>> >
> >> >> >> >>> > Get rid of that, too, please.
> >> >> >> >>>
> >> >> >> >>> Hi Steffan,
> >> >> >> >>>
> >> >> >> >>> I believe I need to keep the "syscon". The same register (ctrlcfg)
> >> >> >> >>> that has the ECC enable bitfield also includes the DRAM configuration
> >> >> >> >>> bitfields that other drivers will want to access (specifically the
> >> >> >> >>> FPGA bridge needs this information). Since this register will be
> >> >> >> >>> shared between drivers, syscon seems like the best solution.
> >> >> >> >>>
> >> >> >> >>
> >> >> >> >> Hm, from looking at the documentation of the ctrlcfg I can't really
> >> >> >> >> understand which bits you would need for the FPGA brigde and why.
> >> >> >>
> >> >> >> Hi Steffen,
> >> >> >>
> >> >> >> Offset 0x80 in the sdr-ctl is the "fpgaportrst" register. 14 bits
> >> >> >> wide, defaults to 0. When appropriate bits set to 1 in that reg, it
> >> >> >> allows an FPGA port to come out of reset (enables that port). Has no
> >> >> >> other effect on SDRAM configuration.
> >> >> >>
> >> >> >> >> That all sounds like stuff you would want to set for the specific
> >> >> >> >> RAM you are dealing with on a specific board.
> >> >> >> >> What bridge are you talking about? The SDRAM bridge?
> >> >> >>
> >> >> >> Yes, the port allows the FPGA a direct path to the SDRAM. This one
> >> >> >> register the only register in the sdr that the bridge driver needs.
> >> >> >>
> >> >> >
> >> >> > So, what I suggested down ...
> >> >> >
> >> >> >> >>
> >> >> >> >> I can see the problem with the ECC enable, though.
> >> >> >> >>
> >> >> >> >> Regards,
> >> >> >> >> Steffen
> >> >> >> >>
> >> >> >> >
> >> >> >> >>> > sdrctl@ffc25000 {
> >> >> >> >>> > compatible = "altr,sdr-ctl";
> >> >> >> >>> > reg = <0xffc25000 0x1000>;
> >> >> >> >>> > ranges;
> >> >> >> >>> >
> >> >> >> >>> > edac@ffc2502c {
> >> >> >> >>> > compatible = "altr,sdram-edac";
> >> >> >> >>> > reg = <0xffc2502c 0x50>;
> >> >> >> >>> > interrupts = <0 39 4>;
> >> >> >> >>> > };
> >> >> >> >>> > };
> >> >> >> >>> >
> >> >> >> >>> > Then we can later add:
> >> >> >> >>> >
> >> >> >> >>> > sdr-ports: ports@ffc2507c {
> >> >> >> >>> > #reset-cells = <1>;
> >> >> >> >>> > compatible = "altr,sdr-ports";
> >> >> >> >>> > reg = <0xffc2507c 0x10>;
> >> >> >> >>> > clocks = <&ddr_dqs_clk>;
> >> >> >> >>> > ...
> >> >> >> >>> > };
> >> >> >>
> >> >> >
> >> >> > ... here should work. No?! Just a simple driver that registers with the
> >> >> > reset-framework. I would add 0x7c to that driver and than that driver could
> >> >> > "configure" the port and let it out of reset.
> >> >> >
> >> >> > I have done the same thing for the other 3 bridges, but am not finished yet.
> >> >> > Especially the GPV-stuff needs to at least be able to be added later if not now.
> >> >> >
> >> >
> >> > Hi Thor!
> >> >
> >> >> I'm not clear on how the EDAC driver will interact with the registers
> >> >> allocated to the SDRAM controller. If the group of registers from
> >> >> 0xffc25000 to 0xffc26000 is exclusively allocated to the SDRAM
> >> >> controller, how does the EDAC driver cleanly access that single
> >> >> register inside this range?
> >> >>
> >> >
> >> > The compatible in the example is wrong. I didn't mean to map the whole address space
> >> > to some driver.
> >> > I think for the configuration register syscon is the right approach. It is a
> >> > bag of bits that don't necessitate an own driver, so syscon is perfect.
> >> >
> >> > So, let me change my proposal to
> >> >
> >> > sdr-ctl: sdram@ffc25000 {
> >> > compatible = "altr,sdr-ctl", "syscon";
> >> > reg = <0xffc25000 0x1>;
> >> > };
> >> >
> >> > edac: edac@ffc2502c {
> >> > compatible = "altr,sdram-edac";
> >> > reg = <0xffc2502c 0x50>;
> >> > interrupts = <0 39 4>;
> >> > config = <&sdr-ctl>;
> >> > ...
> >> > };
> >> >
> >> > sdr-ports: ports@ffc2507c {
> >> > compatible = "altr,sdr-ports";
> >> > reg = <0xffc2507c 0x10>;
> >> > clocks = <&ddr_dqs_clk>;
> >> > ...
> >> > };
> >> >
> >> > Maybe we can just skip the outer node that combines all the others.
> >> > So, if we do it like that, you can still use syscon, but only for the register
> >> > that needs it. And the EDAC definitely needs access to the config register, so
> >> > all is good.
> >> >
> >> >> Is the solution that I don't use request_region() (and therefore not
> >> >> request exclusive access) when setting up the SDRAM controller?
> >> >>
> >> >> If you could point me to your drivers for the other bridges that you
> >> >> reference, your code may answer my question.
> >> >>
> >> >
> >> > The other bridges don't need access to any SDRAM controller registers and
> >> > I haven't even started to implement the GPV-stuff with the lw2hps-bridge :-(
> >> >
> >> > Regards,
> >> > Steffen
> >> >
> >> > --
> >> > Pengutronix e.K. | |
> >> > Industrial Linux Solutions | http://www.pengutronix.de/ |
> >> > Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
> >> > Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
> >>
> >> OK now I understand. For register offset 0x0, that's all stuff set up
> >> in the bootloader. We won't be touching that register except for the
> >> EDAC. So syscon isn't needed here.
> >>
> >> Each driver that uses some sdr registers can specify which registers
> >> it uses in the device tree. So far we don't have any cases of two
> >> drivers that share a register.
> >>
> >> The ECC driver will need two ranges: offset 0x00 and 0x38 through 0x50.
> >> The fpga bridges just needs offset 0x80.
> >>
> >
> > Well, almost ;-)
> > Use syscon for 0x0 and reference that in the ECC driver, which only is
> > responsible for 0x38 to 0x50. Then flip the two EDAC bits via the syscon-phandle.
> > Then a bootloader can probe with the same dts and have a driver match on the
> > config-register and one for all the DRAM setup (< 0x38), which we don't need in
> > the kernel. On the other hand the EDAC seems unneccessary for the bootloader.
> > And this all matches the partitioning of the SDR register space, so I think it
> > is also a correct hardware description.
> >
> > Regards,
> > Steffen
> >
> > --
> > Pengutronix e.K. | |
> > Industrial Linux Solutions | http://www.pengutronix.de/ |
> > Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
> > Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
>
> We could do that. But is syscon a concept the bootloader will know about?
>

Depends on the bootloader. I guess u-boot doesn't, but barebox does.
Alternative would be to have the bootloader know that it has to look for an
ECC node when it instead is looking for DRAM timing/width/... setup.
Sounds wrong to me.

> Actually it is a bit messier than what I originally said:
> ECC driver needs two ranges: offset 0x00 and 0x38 through 0x50.
> Machine layer needs control of 0x10, 0x14, 0x54, and 0x58 for putting
> SDRAM into self-refresh.

I didn't get that. Does the ECC driver need access to this "machine layer"?

Regards,
Steffen


--
Pengutronix e.K. | |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/