Hi Stephen,
On 1 August 2014 15:50, Stephen Warren swarren@wwwdotorg.org wrote:
On 08/01/2014 03:32 PM, Simon Glass wrote:
Hi Stephen,
On 1 August 2014 00:06, Stephen Warren swarren@wwwdotorg.org wrote:
On 07/31/2014 04:10 PM, Simon Glass wrote:
Hi Stephen,
On 31 July 2014 21:16, Stephen Warren swarren@wwwdotorg.org wrote:
On 07/30/2014 03:49 AM, Simon Glass wrote:
Some Tegra device tree files do not include information about the serial ports. Add this and also add information about the input clock speed.
The console alias needs to be set up to indicate which port is used for the console.
Also add a binding file since this is missing.
diff --git a/arch/arm/dts/tegra114-dalmore.dts b/arch/arm/dts/tegra114-dalmore.dts index 435c01e..e2426ef 100644 --- a/arch/arm/dts/tegra114-dalmore.dts +++ b/arch/arm/dts/tegra114-dalmore.dts @@ -7,6 +7,7 @@ compatible = "nvidia,dalmore", "nvidia,tegra114";
aliases {
console = &uart_d;I don't think that's a standard alias name. There was some recent discussion in the devicetree mailing list re: using some property in /chosen for this purpose instead. U-Boot and the kernel should use the same representation here.
This is U-Boot's approach at present,
That's not the U-Boot approach on Tegra boards before this patch. I do not want Tegra U-Boot do adopt any more U-Boot-specific not-really-DT-but-pretending-to-be bindings.
if we change it then we should
change it everywhere. I worry that 'chosen' is for Linux rather than U-Boot and we might get very confused about what chosen is for?
That discussion should be had on the devicetree mailing list.
Please go ahead if you wish, but this is not a Linux issue. The aliases are used for U-Boot and have been for some time.
No, it's not a Linux issue, it's DT issue.
OK agreed on that then.
DT schemas/bindings MUST be identical between U-Boot, Linux, FreeBSD, Barebox, ... (all of which use DT). As such, all the DT bindings MUST be discussed on the devicetree mailing list.
Since you're the author of the patch, it's your responsibility to have that discussion.
Are you referring to the linux,stdout-path discussion, or something more DT-generic?I suppose we could have a 'u-boot,console' for our part. But in any case you are talking about code and a convention that is already in mainline U-Boot. While I accept that we might change to something DT-generic if Linux points the way to something better, I don't want to stop using it just because Linux hasn't decided yet. The early console stuff and early debug UART stuff in Linux is not yet a shining example of perfection.
That said, it's a good time to adopt 'u-boot,console' if that's what we need?
diff --git a/arch/arm/dts/tegra114.dtsi b/arch/arm/dts/tegra114.dtsi
uart_a: serial@70006000 {compatible = "nvidia,tegra20-uart";reg = <0x70006000 0x40>;reg-shift = <2>;clock-frequency = <408000000>;This isn't a property that's defined by the Tegra serial binding. This information should be obtained by looking up the relevant clock, and querying its rate.
We can't do that in the ns16550 driver as yet since there is no generic U-Boot clock infrastructure. I suspect that will come with time.
The solution here is to put the clock infra-structure in place first. One thing I've learned from the kernel DT experience is that a lot of time would have been saved by putting the correct infra-structure in place first, then using it, rather than hacking around things the wrong way, then putting the infra-structure in place, then converting to it. That's a lot more work, and rather painful. Equally, if we don't just do the infra-structure right, there's really little guarantee that we'll ever convert to the correct approach. Just look at all the DT content in use in U-Boot that don't match the real DT bindings, even after it's been around years.
OK, is there a plan to put this in place? Who is working on it?
I don't know whether anyone is or not. However, the fact that nobody is working on a clock driver is no excuse for using the incorrect DT bindings in order to hack around its lack of existence.
Actually the bindings are correct, you are referring to a single new addition which is the input clock speed of the UART. This information is needed in U-Boot because it does not have generic clock infrastructure (and since no one is working on clock infrastructure I guess we can ignore it for this discussion). See below.
For reference, here's the DT node for this UART in the kernel DT, which complies with the relevant binding document:
uarta: serial@70006000 { compatible = "nvidia,tegra114-uart","nvidia,tegra20-uart";
reg = <0x70006000 0x40>; reg-shift = <2>; interrupts = <GIC_SPI 36 IRQ_TYPE_LEVEL_HIGH>; clocks = <&tegra_car TEGRA114_CLK_UARTA>; resets = <&tegra_car 6>; reset-names = "serial"; dmas = <&apbdma 8>, <&apbdma 8>; dma-names = "rx", "tx"; status = "disabled"; };All the comment above apply to all the files in this patch.
My intent was to make this work with a more generic binding for now - ns16550 is a pretty standard thing and I thought I could avoid making the driver Tegra-specific. Then we could allow many SoCs to use it. Why does Tegra have its own binding in the kernel for this standard UART?
The HW is not a PC-style UART where all you care about is the 16550 registers, and clocks/resets/DMA/... can be ignored or deferred to firmware to set up before DT-driven SW runs.
I'm not sure what you are referring to here. Are you saying that U-Boot needs to support these things? I agree it would be great to add a generic clock/reset framework and have made considerable efforts in Tegra towards this myself. But we don't have it yet.
Yes.
Part of the decision to use DT is to use *the* DT bindings, not something that looks like DT.
If you want to move the serial port information into DT, and part of doing so requires extracting clock-related information from DT, and that in turn requires a clock driver to do so, then yes, U-Boot needs to have that clock driver implemented before you can move the serial port information into DT.
Again, your objection is purely about the clock-frequency property I think. The binding is otherwise correct, right?
As an aside, /almost/ all reviewed DT bindings use DT properties of the form:
clocks = <&provider parameters>;
rather than:
clock-rate = <number>;
So, that aspect of the Tegra UART binding isn't anything remotely unusual/non-standard.
The great is the enemy of the good. In this case, I think I might just leave the clock as a #define.
That probably makes sense.
To be honest, I'm not sure that using DT is going to buy U-Boot much, or even the kernel in retrospect.
It's not that bad :-)
The only option I can think of given your objection to the clock-frequency property is to come up with a tegra driver which mostly uses ns16550 but provides the clock frequency from a CONFIG. I think that will work. I'll update and resend.
Regards, Simon