[PATCH 2/2] board: tbs2910: Convert to DM_SERIAL

Simon Glass sjg at chromium.org
Mon Mar 14 23:20:43 CET 2022


Hi Sören,

On Mon, 14 Mar 2022 at 15:51, Sören Moch <smoch at web.de> wrote:
>
> Hi Simon,
>
> On 14.03.22 20:37, Simon Glass wrote:
> > Hi Soeren,
> >
> > On Mon, 14 Mar 2022 at 13:22, Soeren Moch <smoch at web.de> wrote:
> >>
> >> On 14.03.22 19:28, Tom Rini wrote:
> >>> On Mon, Mar 14, 2022 at 12:24:36PM -0600, Simon Glass wrote:
> >>>> Hi Soeren,
> >>>>
> >>>> On Mon, 14 Mar 2022 at 02:26, Soeren Moch <smoch at web.de> wrote:
> >>>>> ... to get rid of the build warning.
> >>>>> Unfortunately we still need the board specific serial pin init code.
> >>>>> Otherwise the first boot messages over the serial console are lost.
> >>>>>
> >>>>> Signed-off-by: Soeren Moch <smoch at web.de>
> >>>>> ---
> >>>>> Cc: Stefano Babic <sbabic at denx.de>
> >>>>> Cc: Fabio Estevam <festevam at gmail.com>
> >>>>> Cc: Tom Rini <trini at konsulko.com>
> >>>>> Cc: Simon Glass <sjg at chromium.org>
> >>>>> Cc: u-boot at lists.denx.de
> >>>>>
> >>>>> The whole purpose of DM is somewhat defeated when we still need board
> >>>>> specific initializations. Any ideas how we can get all boot messages
> >>>>> without board specific inits? 'u-boot,dm-pre-reloc;' in the uart device
> >>>>> tree node did not help.
> >>>> You can put that in your serial driver, perhaps? Or in the initial SoC
> >>>> init code?
> >> Why should I do so? The whole point of DM is initializing devices from
> >> DT. And when I wish to do so pre-relocation, it is advertised in DM to
> >> add 'u-boot,dm-pre-reloc;' for this purpose. I tried, it did not work.
> >> And this is nothing closely related to the serial driver itself, I just
> >> want the pin setup running pre-relocation and not as late as it is
> >> running now under DM_SERIAL.
> > If you have a pinctrl driver it will be used. I don't really
> > understand your problem.
> My problem is that pin initializations come too late (just before the
> "Core" boot message).
> Apparently I have a pinctrl driver, otherwise the pin init would not be
> done at all, I guess.

Who knows, why don't you check?

> >> I also do not want to run this pin setup twice (first in board or SoC
> >> code and again by DM_SERIAL later). Maybe I miss something obvious, but
> >> duplication of the setup code cannot be a proper solution.
> > Well the pinctrl will be triggered before relocation and after, if
> > enabled. We could solve that but have not tried.
> My problem is not runtime, if initialization is done twice from the same
> code this is probably OK. In my setup pins are _not_ initialized before
> relocation, when not done in board_early_init_f() "by hand", which I
> would like to avoid since this results in code duplication.
> Do I need to enable the before-relocation part somewhere? When yes, how
> exactly? 'u-boot,dm-pre-reloc;' in the uart DT node (as documented) did
> not work.

You need your driver to be bound before relocation (so needs the tag
as you say). The infrastructure is all there and works on other
boards. It is strange that you don't use SPL, though. How do you init
the DRAM?

You could enable the debug UART as a starting point, if you don't have
JTAG debugging, since that will allow you to figure out why your
pinctrl driver is not being run.

In the unlikely event that it helps, see the diff below that was
enough to get the serial going on an mx6 board in SPL about 2 years
ago (so the tags should work the same for U-Boot proper before
relocation).

If the error checking is working correctly and people have not just
make drivers return 0 when something goes wrong, you can normally
figure out which driver is missing.

new file mode 100644
index 00000000000..b83881780c3
--- /dev/null
+++ b/arch/arm/dts/imx6q-snappermx6-u-boot.dtsi
@@ -0,0 +1,22 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Copyright 2020 Designa Electronics Ltd
+ */
+
+/ {
+    chosen {
+        stdout-path = &uart5;
+    };
+};
+
+&aips2 {
+    u-boot,dm-pre-reloc;
+};
+
+&soc {
+    u-boot,dm-pre-reloc;
+};
+
+&uart5 {
+    u-boot,dm-pre-reloc;
+};
diff --git a/arch/arm/dts/imx6qdl.dtsi b/arch/arm/dts/imx6qdl.dtsi
index e4daf150881..33e636b2d31 100644
--- a/arch/arm/dts/imx6qdl.dtsi
+++ b/arch/arm/dts/imx6qdl.dtsi
@@ -139,7 +139,7 @@
         interrupts = <0 94 IRQ_TYPE_LEVEL_HIGH>;
     };

-    soc {
+    soc: soc {
         #address-cells = <1>;
         #size-cells = <1>;
         compatible = "simple-bus";
@@ -913,7 +913,7 @@
             };
         };

-        aips-bus at 2100000 { /* AIPS2 */
+        aips2: aips-bus at 2100000 { /* AIPS2 */
             compatible = "fsl,aips-bus", "simple-bus";
             #address-cells = <1>;
             #size-cells = <1>;


> >>>> Another recent way (in -next) is to use events to monitor the
> >>>> EVT_DM_PRE_PROBE event for the serial driver.
> >> I can monitor the probe event, OK. But how can this solve my problem?
> >> Again, maybe I miss something obvious, please tell me when I do so.
> >>> It's just the same thing every single imx platform is doing.
> >>>
> >> Sorry, I don't understand what you mean here. The reference platform for
> >> my board is mx6sabresd. This is not converted to DM_SERIAL yet. Most (?)
> >> imx boards use SPL, pin setup is different there.
> >> I looked into imx boards with DM_SERIAL. They either removed the
> >> board-specific setup code (which results in missing early boot messages:
> >> u-boot version, board name, DDR size, ...) or they are playing tricks in
> >> SPL (not the clean and easy solution that DM promises). Maybe I missed a
> >> better reference for the DM_SERIAL conversion without SPL. Can you point
> >> me to such board?
> > If you want to use pinctrl in SPL, you can do all of this cleanly. If
> > you have code-size constraints, then you may want to do something like
> > rockchip, where only specific peripherals are supported in pinctrl in
> > SPL.
> I do not use SPL, only U-Boot proper.
> >
> > You could look at firefly-rk3288 (or bob/coral/jerry) which I believe
> > is done fully with driver model.
> I want a proper solution without SPL, see above.
> > Perhaps Tom has a better handle on the problem.
> "Great." You are forcing everyone in DM conversions with deadlines. This
> conversion does not work as documented. When asked for help you only
> provide answers to different questions than what was asked. And you
> conclude with "create your own solution or ask someone else", at least
> this is as I understand this.

Your expectations are way out of whack. U-Boot has supported driver
model for serial for 8 years. U-Boot relies on people digging in and
figuring out their own problems. I have converted more than a dozen
boards to driver model (I am not actually sure how many) but I am just
one person and there are over a thousand boards.

>
> All this while DM conversions only bring additional work for maintainers
> of existing boards, DM conversions always come with increased code size,
> and only in the best case everything works like before.
>
> On the other hand you complain about slow conversions and maintainers
> that wait for the very end of the deadline. Do you see the relation?
>
>
> So I ask you again, Simon. How is this DM_SERIAL conversion supposed to
> be done properly? In general and especially for imx boards without SPL?
> Or is all this as incomplete as it looks like? In this case the
> conversion probably will again last until the end of the real deadline,
> about 3 years from now. And it will be done as in this patch (with your
> Reviewed-by blessing), papering over the fact that all the old code is
> still active, only the build warning is silenced. Exactly what we want
> to avoid, bigger code not better code. I hope we can clean this up in a
> follow-up patch.

I would suggest that instead of complaining and accusing people of
things, you would be better to sit down and take a hard look at it. It
is not that difficult to figure out, if you have a debug UART or JTAG.
It looks like 50% of iMX6 boards enable DM_SERIAL so it cannot be
impossible. There really is nothing magical about driver model. It is
just a model for how drivers fit together. It still runs the same
code, just organised in a somewhat more rational way, at the cost of
some extra complexity and code size.

Regards,
Simon


More information about the U-Boot mailing list