Skip to contentSkip to conversation

board: tbs2910: Enable Link Time Optimizations in defconfig

PATCH In review Soeren Moch 4 participants 16 messages Mar 14, 2022 – Mar 19, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 board: tbs2910: Enable Link Time Optimizations in defconfig

This saves about 12 kBytes image size and helps to stay within the size limit.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Unfortunately we already are at the size limit for U-Boot v2022.04-rc3. Maybe Tom can pick up this patch directly, otherwise it's also fine for me to take this through the imx tree.

---
 configs/tbs2910_defconfig | 1 +
 1 file changed, 1 insertion(+)
configs/tbs2910_defconfig +1 -0
index e1278f2e70..8a33160564 100644
@@ -14,6 +14,7 @@ CONFIG_DEFAULT_DEVICE_TREE="imx6q-tbs2910"
 CONFIG_PRE_CON_BUF_ADDR=0x7c000000
 CONFIG_CMD_HDMIDETECT=y
 CONFIG_AHCI=y
+CONFIG_LTO=y
 CONFIG_SUPPORT_RAW_INITRD=y
 CONFIG_BOOTDELAY=3
 CONFIG_USE_BOOTCOMMAND=y

--
2.25.1

#1 lore raw

PATCH 2/2 board: tbs2910: Convert to DM_SERIAL

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

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.

---
 configs/tbs2910_defconfig | 2 ++
 1 file changed, 2 insertions(+)
configs/tbs2910_defconfig +2 -0
index 8a33160564..79454053b0 100644
@@ -81,6 +81,8 @@ CONFIG_PINCTRL=y
 CONFIG_PINCTRL_IMX6=y
 CONFIG_DM_RTC=y
 CONFIG_RTC_DS1307=y
+CONFIG_DM_SERIAL=y
+CONFIG_SERIAL_RX_BUFFER=y
 CONFIG_MXC_UART=y
 CONFIG_DM_THERMAL=y
 CONFIG_IMX_THERMAL=y

--
2.25.1

Reviewed-by on v1 2/2
#2 lore raw

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

Hi Soeren,

Quoted 13 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 4 lines

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?

Another recent way (in -next) is to use events to monitor the EVT_DM_PRE_PROBE event for the serial driver.

Quoted 3 lines
---
 configs/tbs2910_defconfig | 2 ++
 1 file changed, 2 insertions(+)
Quoted 17 lines
configs/tbs2910_defconfig +2 -1
index 8a33160564..79454053b0 100644
@@ -81,6 +81,8 @@ CONFIG_PINCTRL=y
 CONFIG_PINCTRL_IMX6=y
 CONFIG_DM_RTC=y
 CONFIG_RTC_DS1307=y
+CONFIG_DM_SERIAL=y
+CONFIG_SERIAL_RX_BUFFER=y
 CONFIG_MXC_UART=y
 CONFIG_DM_THERMAL=y
 CONFIG_IMX_THERMAL=y
--

2.25.1

#3 lore raw

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

Quoted 21 lines · Simon Glass

Hi Soeren,

Quoted 13 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 4 lines

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.

Quoted 5 lines

You can put that in your serial driver, perhaps? Or in the initial SoC init code?

Another recent way (in -next) is to use events to monitor the EVT_DM_PRE_PROBE event for the serial driver.

It's just the same thing every single imx platform is doing.

Signature · 1 line
Tom
#4 lore raw

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

Quoted 16 lines · Tom Rini
Quoted 15 lines · Simon Glass

Hi Soeren,

Quoted 12 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 6 lines
Quoted 6 lines
Quoted 4 lines

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.

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.

Quoted 3 lines
Quoted 3 lines

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.

Quoted 2 lines

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?

Thanks,
Soeren

#5 lore raw

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

Quoted 41 lines · Soeren Moch
Quoted 16 lines · Tom Rini
Quoted 15 lines · Simon Glass

Hi Soeren,

Quoted 12 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 6 lines
Quoted 6 lines
Quoted 4 lines

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.

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.

Quoted 3 lines
Quoted 3 lines

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.

Quoted 2 lines

It's just the same thing every single imx platform is doing.

Quoted 9 lines

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?

My point, largely to Simon, sorry, is that this isn't a "tbs2910" specific problem, but rather something that's not solved for any of the imx platforms as they do some form of what you note above. I suspect it's not solved for anything else either and other platforms are doing some number of what you said above (or relying on something else already having done the pinmux).

Signature · 1 line
Tom
#6 lore raw

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

In reply to Soeren Moch

Hi Soeren,

Quoted 23 lines · Soeren Moch
Quoted 20 lines · Tom Rini
Quoted 19 lines · Simon Glass

Hi Soeren,

Quoted 16 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

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.

Quoted 8 lines
Quoted 2 lines
Quoted 2 lines

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.

Quoted 4 lines

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.

Quoted 7 lines
Quoted 3 lines
Quoted 3 lines

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.

Quoted 2 lines

It's just the same thing every single imx platform is doing.

Quoted 9 lines

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.

You could look at firefly-rk3288 (or bob/coral/jerry) which I believe is done fully with driver model.

Perhaps Tom has a better handle on the problem.

Regards,
Simon

#7 lore raw

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

Hi Simon,

Quoted 27 lines · Simon Glass

Hi Soeren,

Quoted 24 lines · Soeren Moch
Quoted 22 lines · Tom Rini
Quoted 21 lines · Simon Glass

Hi Soeren,

Quoted 12 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 4 lines

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?

Quoted 8 lines
Quoted 6 lines

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.

Quoted 5 lines
Quoted 3 lines

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.

Quoted 15 lines
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines

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.

Quoted 2 lines

It's just the same thing every single imx platform is doing.

Quoted 9 lines

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?

Quoted 4 lines

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.

Quoted 3 lines

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.

Quoted 1 line

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.

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.

Regards,
Soeren

Quoted 3 lines

Regards,
Simon

#8 lore raw

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

Hi Sören,

Quoted 37 lines · Sören Moch

Hi Simon,

Quoted 33 lines · Simon Glass

Hi Soeren,

Quoted 30 lines · Soeren Moch
Quoted 16 lines · Tom Rini
Quoted 15 lines · Simon Glass

Hi Soeren,

Quoted 12 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 6 lines
Quoted 6 lines
Quoted 4 lines

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.

Quoted 6 lines
Quoted 2 lines

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?

Quoted 5 lines
Quoted 5 lines
Quoted 3 lines

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.

Quoted 7 lines

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

arch/arm/dts/imx6q-snappermx6-u-boot.dtsi new file +22 -0
@@ -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;
+};
arch/arm/dts/imx6qdl.dtsi +2 -2
index e4daf150881..33e636b2d31 100644
@@ -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@2100000 { /* AIPS2 */
+        aips2: aips-bus@2100000 { /* AIPS2 */
             compatible = "fsl,aips-bus", "simple-bus";
             #address-cells = <1>;
             #size-cells = <1>;
Quoted 23 lines
Quoted 15 lines
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines

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.

Quoted 2 lines

It's just the same thing every single imx platform is doing.

Quoted 9 lines

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?

Quoted 4 lines

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.

Quoted 3 lines

You could look at firefly-rk3288 (or bob/coral/jerry) which I believe is done fully with driver model.

Quoted 7 lines

I want a proper solution without SPL, see above.

Quoted 1 line

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.

Quoted 9 lines

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?

Quoted 9 lines

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

#9 lore raw

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

Quoted 62 lines · Simon Glass

Hi Sören,

Quoted 37 lines · Sören Moch

Hi Simon,

Quoted 33 lines · Simon Glass

Hi Soeren,

Quoted 30 lines · Soeren Moch
Quoted 16 lines · Tom Rini
Quoted 15 lines · Simon Glass

Hi Soeren,

Quoted 12 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 6 lines
Quoted 6 lines
Quoted 4 lines

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.

Quoted 6 lines
Quoted 2 lines

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?

Quoted 5 lines
Quoted 5 lines
Quoted 3 lines

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.

Quoted 7 lines

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.

Quoted 4 lines

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?

Probably through the DCD script that CONFIG_IMX_CONFIG sets.

Quoted 120 lines

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

arch/arm/dts/imx6q-snappermx6-u-boot.dtsi new file +22 -0
@@ -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;
+};
arch/arm/dts/imx6qdl.dtsi +2 -2
index e4daf150881..33e636b2d31 100644
@@ -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@2100000 { /* AIPS2 */
+        aips2: aips-bus@2100000 { /* AIPS2 */
             compatible = "fsl,aips-bus", "simple-bus";
             #address-cells = <1>;
             #size-cells = <1>;
Quoted 25 lines
Quoted 17 lines
Quoted 8 lines
Quoted 4 lines
Quoted 4 lines

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.

Quoted 2 lines

It's just the same thing every single imx platform is doing.

Quoted 9 lines

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?

Quoted 4 lines

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.

Quoted 3 lines

You could look at firefly-rk3288 (or bob/coral/jerry) which I believe is done fully with driver model.

Quoted 7 lines

I want a proper solution without SPL, see above.

Quoted 1 line

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.

Quoted 9 lines

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?

Quoted 9 lines

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.

Quoted 8 lines

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.

Again, DM_SERIAL can be enabled on the board as he's already shown by setting two options, which silences the warning, increases the size and doesn't make anything better. That's likely what the other boards are doing, or they're just having some console output be missing. What would need to be set where, so that serial output isn't missing, without having to manually call setup_iomux_uart() like so many imx platforms are doing.

Signature · 1 line
Tom
#10 lore raw

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

Hi Tom,

Quoted 200 lines · Tom Rini
Quoted 62 lines · Simon Glass

Hi Sören,

Quoted 37 lines · Sören Moch

Hi Simon,

Quoted 33 lines · Simon Glass

Hi Soeren,

Quoted 30 lines · Soeren Moch
Quoted 16 lines · Tom Rini
Quoted 15 lines · Simon Glass

Hi Soeren,

Quoted 12 lines · Soeren Moch

... 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.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 6 lines
Quoted 6 lines
Quoted 4 lines

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.

Quoted 6 lines
Quoted 2 lines

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?

Quoted 5 lines
Quoted 5 lines
Quoted 3 lines

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.

Quoted 7 lines

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.

Quoted 4 lines

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?

Probably through the DCD script that CONFIG_IMX_CONFIG sets.

Quoted 120 lines

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

arch/arm/dts/imx6q-snappermx6-u-boot.dtsi new file +22 -0
@@ -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;
+};
arch/arm/dts/imx6qdl.dtsi +2 -2
index e4daf150881..33e636b2d31 100644
@@ -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@2100000 { /* AIPS2 */
+        aips2: aips-bus@2100000 { /* AIPS2 */
             compatible = "fsl,aips-bus", "simple-bus";
             #address-cells = <1>;
             #size-cells = <1>;
Quoted 25 lines
Quoted 17 lines
Quoted 8 lines
Quoted 4 lines
Quoted 4 lines

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.

Quoted 2 lines

It's just the same thing every single imx platform is doing.

Quoted 9 lines

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?

Quoted 4 lines

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.

Quoted 3 lines

You could look at firefly-rk3288 (or bob/coral/jerry) which I believe is done fully with driver model.

Quoted 7 lines

I want a proper solution without SPL, see above.

Quoted 1 line

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.

Quoted 9 lines

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?

Quoted 9 lines

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.

Quoted 8 lines

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.

Quoted 7 lines

Again, DM_SERIAL can be enabled on the board as he's already shown by setting two options, which silences the warning, increases the size and doesn't make anything better. That's likely what the other boards are doing, or they're just having some console output be missing. What would need to be set where, so that serial output isn't missing, without having to manually call setup_iomux_uart() like so many imx platforms are doing.

Probably missing soc/clock node as per my patch above. But it needs investigation.

Regards,
Simon

reviewer
#11 lore raw

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

In reply to Tom Rini

Hi Soeren,

Quoted 7 lines · Tom Rini

Again, DM_SERIAL can be enabled on the board as he's already shown by setting two options, which silences the warning, increases the size and doesn't make anything better. That's likely what the other boards are doing, or they're just having some console output be missing. What would need to be set where, so that serial output isn't missing, without having to manually call setup_iomux_uart() like so many imx platforms are doing.

Does the attached patch work for you?

I tested a similar patch on warp7 and it works fine here and the early U-Boot messages are shown.

From a47c625f30333e45e3836ad797a47a3cc506ed94 Mon Sep 17 00:00:00 2001
From: Fabio Estevam <[email protected]>
Date: Mon, 14 Mar 2022 20:12:02 -0300
Subject: [PATCH] tbs2910: Convert to DM_SERIAL

Conversion to DM_SERIAL is mandatory.

Select DM_SERIAL and add a imx6q-tbs2910-u-boot.dtsi file that describes the nodes that require dm-pre-reloc, which allows the DM model to configure the UART pinctrl early.

Remove the now unneeded board UART initialization.

---
 arch/arm/dts/imx6q-tbs2910-u-boot.dtsi | 16 ++++++++++++++++
 arch/arm/dts/imx6qdl.dtsi              |  4 ++--
 board/tbs/tbs2910/tbs2910.c            | 26 --------------------------
 configs/tbs2910_defconfig              |  2 +-
 4 files changed, 19 insertions(+), 29 deletions(-)
 create mode 100644 arch/arm/dts/imx6q-tbs2910-u-boot.dtsi
arch/arm/dts/imx6q-tbs2910-u-boot.dtsi new file +16 -0
new file mode 100644
index 000000000000..4c164068b016
@@ -0,0 +1,16 @@
+&aips1 {
+	u-boot,dm-pre-reloc;
+};
+
+
+&pinctrl_uart1 {
+	u-boot,dm-pre-reloc;
+};
+
+&soc {
+	u-boot,dm-pre-reloc;
+};
+
+&uart1 {
+	u-boot,dm-pre-reloc;
+};
arch/arm/dts/imx6qdl.dtsi +2 -2
index efd89510d512..d89272039b28 100644
@@ -139,7 +139,7 @@
 		interrupts = <0 94 IRQ_TYPE_LEVEL_HIGH>;
 	};
 
-	soc {
+	soc: soc {
 		#address-cells = <1>;
 		#size-cells = <1>;
 		compatible = "simple-bus";
@@ -283,7 +283,7 @@
 			status = "disabled";
 		};
 
-		bus@2000000 { /* AIPS1 */
+		aips1: bus@2000000 { /* AIPS1 */
 			compatible = "fsl,aips-bus", "simple-bus";
 			#address-cells = <1>;
 			#size-cells = <1>;
board/tbs/tbs2910/tbs2910.c +0 -26
index faf73cc218c9..3a447ca8a93c 100644
@@ -22,32 +22,12 @@
 #include <asm/arch/sys_proto.h>
 DECLARE_GLOBAL_DATA_PTR;
 
-#define UART_PAD_CTRL  (PAD_CTL_PUS_100K_UP |			\
-	PAD_CTL_SPEED_MED | PAD_CTL_DSE_40ohm |			\
-	PAD_CTL_SRE_FAST  | PAD_CTL_HYS)
-
-static iomux_v3_cfg_t const uart1_pads[] = {
-	MX6_PAD_CSI0_DAT10__UART1_TX_DATA | MUX_PAD_CTRL(UART_PAD_CTRL),
-	MX6_PAD_CSI0_DAT11__UART1_RX_DATA | MUX_PAD_CTRL(UART_PAD_CTRL),
-};
-
-static iomux_v3_cfg_t const uart2_pads[] = {
-	MX6_PAD_EIM_D26__UART2_TX_DATA | MUX_PAD_CTRL(UART_PAD_CTRL),
-	MX6_PAD_EIM_D27__UART2_RX_DATA | MUX_PAD_CTRL(UART_PAD_CTRL),
-};
-
 int dram_init(void)
 {
 	gd->ram_size = 2048ul * 1024 * 1024;
 	return 0;
 }
 
-static void setup_iomux_uart(void)
-{
-	imx_iomux_v3_setup_multiple_pads(uart1_pads, ARRAY_SIZE(uart1_pads));
-	imx_iomux_v3_setup_multiple_pads(uart2_pads, ARRAY_SIZE(uart2_pads));
-}
-
 #ifdef CONFIG_FSL_ESDHC_IMX
 /* set environment device to boot device when booting from SD */
 int board_mmc_get_env_dev(int devno)
@@ -150,12 +130,6 @@ static void setup_display(void)
 }
 #endif /* CONFIG_VIDEO_IPUV3 */
 
-int board_early_init_f(void)
-{
-	setup_iomux_uart();
-	return 0;
-}
-
 #ifdef CONFIG_CMD_BMODE
 static const struct boot_mode board_boot_modes[] = {
 	/* 4 bit bus width */
configs/tbs2910_defconfig +1 -1
index 8a3316056408..63a778a45f0b 100644
@@ -23,7 +23,6 @@ CONFIG_USE_PREBOOT=y
 CONFIG_PREBOOT="echo PCI:; pci enum; pci 1; usb start; if hdmidet; then run set_con_hdmi; else run set_con_serial; fi"
 CONFIG_DEFAULT_FDT_FILE="imx6q-tbs2910.dtb"
 CONFIG_PRE_CONSOLE_BUFFER=y
-CONFIG_BOARD_EARLY_INIT_F=y
 CONFIG_HUSH_PARSER=y
 CONFIG_SYS_PROMPT="Matrix U-Boot> "
 # CONFIG_CMD_BDI is not set
@@ -81,6 +80,7 @@ CONFIG_PINCTRL=y
 CONFIG_PINCTRL_IMX6=y
 CONFIG_DM_RTC=y
 CONFIG_RTC_DS1307=y
+CONFIG_DM_SERIAL=y
 CONFIG_MXC_UART=y
 CONFIG_DM_THERMAL=y
 CONFIG_IMX_THERMAL=y
Signature · 1 line
2.25.1
#12 lore raw

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

Hi Fabio,

Quoted 11 lines · Fabio Estevam

Hi Soeren,

Quoted 7 lines · Tom Rini

Again, DM_SERIAL can be enabled on the board as he's already shown by setting two options, which silences the warning, increases the size and doesn't make anything better. That's likely what the other boards are doing, or they're just having some console output be missing. What would need to be set where, so that serial output isn't missing, without having to manually call setup_iomux_uart() like so many imx platforms are doing.

Quoted 4 lines

Does the attached patch work for you?

I tested a similar patch on warp7 and it works fine here and the early U-Boot messages are shown.

This looks quite similar to my first attempt on this, but with additional pre-reloc attributes on aips1, pinctrl_uart1, and soc. Quite promising, probably exactly the right solution that I was looking for.

I promise to test this still this week, but may not be able to do so before the weekend.

Thanks for working on this,
Soeren

#13 lore raw

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

Hi Fabio,

Quoted 19 lines · Soeren Moch

Hi Fabio,

Quoted 12 lines · Fabio Estevam

Hi Soeren,

Quoted 8 lines · Tom Rini

Again, DM_SERIAL can be enabled on the board as he's already shown by setting two options, which silences the warning, increases the size and doesn't make anything better.  That's likely what the other boards are doing, or they're just having some console output be missing. What would need to be set where, so that serial output isn't missing, without
having to manually call setup_iomux_uart() like so many imx platforms are doing.

Quoted 4 lines

Does the attached patch work for you?

I tested a similar patch on warp7 and it works fine here and the early U-Boot messages are shown.

Quoted 7 lines

This looks quite similar to my first attempt on this, but with additional pre-reloc attributes on aips1, pinctrl_uart1, and soc. Quite promising, probably exactly the right solution that I was looking for.

I promise to test this still this week, but may not be able to do so before the weekend.

Your patch works great, I already sent my Tested-by to patchwork. So my patch should be dropped in favour of your one, but I saw that my patch already is marked 'Superseded'.

Regards,
Soeren

Quoted 3 lines

Thanks for working on this,
Soeren

Reviewed-by on v1 1/2
#15 lore raw
In reply to Soeren Moch
Quoted 12 lines · Soeren Moch

This saves about 12 kBytes image size and helps to stay within the size limit.

---
Cc: Stefano Babic <[email protected]>
Cc: Fabio Estevam <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: [email protected]

Quoted 6 lines

Unfortunately we already are at the size limit for U-Boot v2022.04-rc3. Maybe Tom can pick up this patch directly, otherwise it's also fine for me to take this through the imx tree.

---
 configs/tbs2910_defconfig | 1 +
 1 file changed, 1 insertion(+)