Skip to contentSkip to conversation

serial: mxc: get the clock frequency from the used clock for the device

RFC In review Heiko Thiery 6 participants 17 messages Mar 17, 2022 – Mar 24, 2022

lore
Share & tools
lore
#0 lore raw

RFC 1/1 serial: mxc: get the clock frequency from the used clock for the device

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
 drivers/serial/serial_mxc.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +13 -2
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
  * (c) 2007 Sascha Hauer <[email protected]>
  */
 
+#include <clk.h>
 #include <common.h>
 #include <dm.h>
 #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device *default_serial_console(void)
 int mxc_serial_setbrg(struct udevice *dev, int baudrate)
 {
 	struct mxc_serial_plat *plat = dev_get_plat(dev);
-	u32 clk = imx_get_uartclk();
+	u32 rate = 0;
+
+	if (IS_ENABLED(CONFIG_CLK)) {
+		struct clk clk;
+		if(!clk_get_by_name(dev, "ipg", &clk))
+			rate = clk_get_rate(&clk);
+	}
+
+	/* as fallback we try to get the clk rate that way */
+	if (rate == 0)
+		rate = imx_get_uartclk();
 
-	_mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+	_mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
 	return 0;
 }
Signature · 1 line
2.30.2
#1 lore raw

Hi Heiko,

Quoted 27 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
 drivers/serial/serial_mxc.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +1 -0
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
  * (c) 2007 Sascha Hauer <[email protected]>
  */
 
+#include <clk.h>
 #include <common.h>
 #include <dm.h>
 #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device 
 *default_serial_console(void)

drivers/serial/serial_mxc.c __weak struct serial_device · line 276

Quoted 10 lines
drivers/serial/serial_mxc.c +6 -1
 int mxc_serial_setbrg(struct udevice *dev, int baudrate)
 {
 	struct mxc_serial_plat *plat = dev_get_plat(dev);
-	u32 clk = imx_get_uartclk();
+	u32 rate = 0;
+
+	if (IS_ENABLED(CONFIG_CLK)) {
+		struct clk clk;
+		if(!clk_get_by_name(dev, "ipg", &clk))
+			rate = clk_get_rate(&clk);

Is the "ipg" clock the correct name for all of the imx DM boards ?

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 280

Quoted 5 lines
drivers/serial/serial_mxc.c +5 -0
+	}
+
+	/* as fallback we try to get the clk rate that way */
+	if (rate == 0)
+		rate = imx_get_uartclk();

Would it be better to re-write imx_get_uartclk so that both the getting and setting of clocks was correct ?

With DM clocks enabled I don't even think it makes sense to call those older functions.

Angus

Quoted 6 lines
drivers/serial/serial_mxc.c +1 -1
-	_mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+	_mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
 	return 0;
 }
#2 lore raw

Hi Angus,

Am Do., 17. März 2022 um 14:19 Uhr schrieb Angus Ainslie <[email protected]>:

Quoted 37 lines

Hi Heiko,

Quoted 33 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console
output is garbage. To do this correctly the UART frequency is taken from
the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
 drivers/serial/serial_mxc.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +1 -1
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
  * (c) 2007 Sascha Hauer <[email protected]>
  */
 
+#include <clk.h>
 #include <common.h>
 #include <dm.h>
 #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device
 *default_serial_console(void)
 int mxc_serial_setbrg(struct udevice *dev, int baudrate)
 {
      struct mxc_serial_plat *plat = dev_get_plat(dev);
-     u32 clk = imx_get_uartclk();
Quoted 8 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 275

Quoted 6 lines
drivers/serial/serial_mxc.c +6 -0
+     u32 rate = 0;
+
+     if (IS_ENABLED(CONFIG_CLK)) {
+             struct clk clk;
+             if(!clk_get_by_name(dev, "ipg", &clk))
+                     rate = clk_get_rate(&clk);

Is the "ipg" clock the correct name for all of the imx DM boards ?

I checked the dtsi files for all the boards that use the compatible strings from the serial_mxc and see that there are always 2 clocks:
'ipg' and 'per'.

The imx8 dtsi files describe ipg and per always the same clock. The imx7 dtsi files descibe ipg and per always the same clock. The imx6 dtsi files describe ipg and per are 2 different clocks The imx51 dtsi files describeipg and per are 2 different clocks

So I'm not sure if the ipg clock is the right one for the boards that has different clock for ipg and per.

Quoted 8 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 280

Quoted 5 lines
drivers/serial/serial_mxc.c +5 -0
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)
+             rate = imx_get_uartclk();

Would it be better to re-write imx_get_uartclk so that both the getting and setting of clocks was correct ?

I do not understand what you mean with that.

Quoted 2 lines

With DM clocks enabled I don't even think it makes sense to call those older functions.

You mean when DM clocks are available the "new" method should be used and no fallback to the old mechanism?

Quoted 8 lines

Angus

Quoted 6 lines
drivers/serial/serial_mxc.c +1 -1
-     _mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+     _mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
      return 0;
 }
#3 lore raw
Quoted 53 lines · Heiko Thiery

Hi Angus,

Am Do., 17. März 2022 um 14:19 Uhr schrieb Angus Ainslie <[email protected]>:

Quoted 37 lines

Hi Heiko,

Quoted 33 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console
output is garbage. To do this correctly the UART frequency is taken from
the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
 drivers/serial/serial_mxc.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +1 -1
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
  * (c) 2007 Sascha Hauer <[email protected]>
  */
 
+#include <clk.h>
 #include <common.h>
 #include <dm.h>
 #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device
 *default_serial_console(void)
 int mxc_serial_setbrg(struct udevice *dev, int baudrate)
 {
      struct mxc_serial_plat *plat = dev_get_plat(dev);
-     u32 clk = imx_get_uartclk();
Quoted 8 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 275

Quoted 6 lines
drivers/serial/serial_mxc.c +6 -0
+     u32 rate = 0;
+
+     if (IS_ENABLED(CONFIG_CLK)) {
+             struct clk clk;
+             if(!clk_get_by_name(dev, "ipg", &clk))
+                     rate = clk_get_rate(&clk);

Is the "ipg" clock the correct name for all of the imx DM boards ?

I checked the dtsi files for all the boards that use the compatible strings from the serial_mxc and see that there are always 2 clocks:
'ipg' and 'per'.

Quoted 7 lines

The imx8 dtsi files describe ipg and per always the same clock. The imx7 dtsi files descibe ipg and per always the same clock. The imx6 dtsi files describe ipg and per are 2 different clocks The imx51 dtsi files describeipg and per are 2 different clocks

So I'm not sure if the ipg clock is the right one for the boards that has different clock for ipg and per.

So I only looked at imx6qdl.dtsi where the clocks are different

clocks = <&clks IMX6QDL_CLK_UART_IPG>,

         <&clks IMX6QDL_CLK_UART_SERIAL>;

clock-names = "ipg", "per";

And from that file it looks like the per clock would be the correct one.

Should the clock be looked up by id instead of by name and then have a different code path for each imx board type ?

Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
drivers/serial/serial_mxc.c +5 -0
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)
+             rate = imx_get_uartclk();
Quoted 5 lines
Quoted 2 lines

Would it be better to re-write imx_get_uartclk so that both the getting and setting of clocks was correct ?

I do not understand what you mean with that.

There are other places in the code that imx_get_uartclk gets called. If an index was added to imx_get_uartclk(int index) then you wouldn't need the code above in the mxc_serial_setbrg function. That would also make all of the places where imx_get_uartclk gets called return the correct value.

It might make sense to create a new function imx7_get_uartclk that gets called on newer SOCs so that the imx6 and earlier code doesn't need to get changed.

Quoted 6 lines
Quoted 2 lines

With DM clocks enabled I don't even think it makes sense to call those older functions.

You mean when DM clocks are available the "new" method should be used and no fallback to the old mechanism?

For older boards it makes sense to fallback to the single clock. On newer boards if it returned an error instead it would have been easier for you to figure out where the serial console failed.

Angus

Quoted 8 lines
Quoted 8 lines

Angus

Quoted 6 lines
drivers/serial/serial_mxc.c +1 -1
-     _mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+     _mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
      return 0;
 }
#4 lore raw

Hi Angus,

[snip]

Quoted 10 lines
Quoted 2 lines

So I'm not sure if the ipg clock is the right one for the boards that has different clock for ipg and per.

So I only looked at imx6qdl.dtsi where the clocks are different

clocks = <&clks IMX6QDL_CLK_UART_IPG>,

          <&clks IMX6QDL_CLK_UART_SERIAL>;

clock-names = "ipg", "per";

And from that file it looks like the per clock would be the correct one.

Yes, 'per' seems to be the right one.

Quoted 2 lines

Should the clock be looked up by id instead of by name and then have a different code path for each imx board type ?

But how to get the right clk id? The id's for all the implementations are different. Or not?

Quoted 12 lines
Quoted 12 lines
Quoted 11 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 280

Quoted 7 lines
drivers/serial/serial_mxc.c +5 -0
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)
+             rate = imx_get_uartclk();

Would it be better to re-write imx_get_uartclk so that both the getting
and setting of clocks was correct ?

Quoted 8 lines
Quoted 2 lines

I do not understand what you mean with that.

There are other places in the code that imx_get_uartclk gets called. If an index was added to imx_get_uartclk(int index) then you wouldn't need the code above in the mxc_serial_setbrg function. That would also make all of the places where imx_get_uartclk gets called return the correct value.

By index do you mean the clk id?

Quoted 4 lines

It might make sense to create a new function imx7_get_uartclk that gets called on newer SOCs so that the imx6 and earlier code doesn't need to get changed.

At what point should this be called instead of the imx_get_uartclk() function?

Quoted 11 lines
Quoted 7 lines
Quoted 3 lines

With DM clocks enabled I don't even think it makes sense to call those older functions.

You mean when DM clocks are available the "new" method should be used and no fallback to the old mechanism?

For older boards it makes sense to fallback to the single clock. On newer boards if it returned an error instead it would have been easier for you to figure out where the serial console failed.

--
Heiko

#5 lore raw
Quoted 15 lines · Heiko Thiery

Hi Angus,

[snip]

Quoted 10 lines
Quoted 2 lines

So I'm not sure if the ipg clock is the right one for the boards that has different clock for ipg and per.

So I only looked at imx6qdl.dtsi where the clocks are different

clocks = <&clks IMX6QDL_CLK_UART_IPG>,

          <&clks IMX6QDL_CLK_UART_SERIAL>;

clock-names = "ipg", "per";

And from that file it looks like the per clock would be the correct one.

Quoted 8 lines

Yes, 'per' seems to be the right one.

Quoted 2 lines

Should the clock be looked up by id instead of by name and then have a different code path for each imx board type ?

But how to get the right clk id? The id's for all the implementations are different. Or not?

Yeah you're correct that won't work.

Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
Quoted 11 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 280

Quoted 7 lines
drivers/serial/serial_mxc.c +5 -0
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)
+             rate = imx_get_uartclk();

Would it be better to re-write imx_get_uartclk so that both the getting
and setting of clocks was correct ?

I do not understand what you mean with that.

Quoted 8 lines
Quoted 5 lines

There are other places in the code that imx_get_uartclk gets called. If an index was added to imx_get_uartclk(int index) then you wouldn't need the code above in the mxc_serial_setbrg function. That would also make all of the places where imx_get_uartclk gets called return the correct value.

By index do you mean the clk id?

No I was thinking number of the device uart[0-3].

Thinking about it some more it's probably not useful as you already have the udevice pointer. I was just trying to think of ways to reduce the amount of code change for the older SOCs. Using the 'per' clock is a better solution.

Quoted 7 lines
Quoted 4 lines

It might make sense to create a new function imx7_get_uartclk that gets called on newer SOCs so that the imx6 and earlier code doesn't need to get changed.

At what point should this be called instead of the imx_get_uartclk() function?

I was thinking a CONFIG_IS_ENBLED switch between the old version and the new one based on SOC.

Angus

#6 lore raw
Quoted 68 lines · Angus Ainslie
Quoted 16 lines · Heiko Thiery

Hi Angus,

[snip]

Quoted 11 lines
Quoted 2 lines

So I'm not sure if the ipg clock is the right one for the boards that has different clock for ipg and per.

So I only looked at imx6qdl.dtsi where the clocks are different

clocks = <&clks IMX6QDL_CLK_UART_IPG>,

          <&clks IMX6QDL_CLK_UART_SERIAL>;

clock-names = "ipg", "per";

And from that file it looks like the per clock would be the correct one.

Quoted 8 lines

Yes, 'per' seems to be the right one.

Quoted 2 lines

Should the clock be looked up by id instead of by name and then have a different code path for each imx board type ?

But how to get the right clk id? The id's for all the implementations are different. Or not?

Yeah you're correct that won't work.

Quoted 23 lines
Quoted 15 lines
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
drivers/serial/serial_mxc.c +5 -0
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)
+             rate = imx_get_uartclk();
Quoted 6 lines
Quoted 3 lines

Would it be better to re-write imx_get_uartclk so that both the getting
and setting of clocks was correct ?

I do not understand what you mean with that.

Quoted 7 lines

There are other places in the code that imx_get_uartclk gets called. If
an index was added to imx_get_uartclk(int index) then you wouldn't need
the code above in the mxc_serial_setbrg function. That would also make all of the places where imx_get_uartclk gets called return the correct value.

Quoted 2 lines

By index do you mean the clk id?

No I was thinking number of the device uart[0-3].

Thinking about it some more it's probably not useful as you already have the udevice pointer. I was just trying to think of ways to reduce the amount of code change for the older SOCs. Using the 'per' clock is a better solution.

Quoted 6 lines
Quoted 5 lines

It might make sense to create a new function imx7_get_uartclk that gets
called on newer SOCs so that the imx6 and earlier code doesn't need to get changed.

Quoted 6 lines
Quoted 3 lines

At what point should this be called instead of the imx_get_uartclk() function?

I was thinking a CONFIG_IS_ENBLED switch between the old version and the new one based on SOC.

I was thinking that we could expand the struct mxc_serial_plat to include both per and igp clocks to cover devices have clocks that are not the same. The configuring of the platdata can happen using mxc_serial_of_to_plat to 'get' both igp and per clocks. The probe would enable both per and igp to ensure they are operating, then the mxc_serial_setbrg could use clk_get_rate(plat->clk_igp) to determine the clock rate.

I am guessing the clock composite would have to be expanded to include the UART clocks because from what I can see, they're not included. However, this could potentially eliminate the need to use some of the functions in arch/arm/mach-imx/imx8m/clock_imx8mm.

The down-side is that a bunch of SoC's might need to be updated to support more and more clocks, so we could potentially use !CONFIG_IS_ENABLED to use the existing functions as a fall-back.

adam

Quoted 2 lines

Angus

#7 lore raw

Hi Adam,

[SNIP]

Quoted 8 lines

I was thinking that we could expand the struct mxc_serial_plat to include both per and igp clocks to cover devices have clocks that are not the same. The configuring of the platdata can happen using mxc_serial_of_to_plat to 'get' both igp and per clocks. The probe would enable both per and igp to ensure they are operating, then the mxc_serial_setbrg could use clk_get_rate(plat->clk_igp) to determine the clock rate.

With your comment I have made the following, does this fit with your suggestion?

drivers/serial/serial_mxc.c +19 -2
index e4970a169b..7f9f7e8383 100644
@@ -3,6 +3,7 @@
  * (c) 2007 Sascha Hauer <[email protected]>
  */
 
+#include <clk.h>
 #include <common.h>
 #include <dm.h>
 #include <errno.h>
@@ -266,9 +267,16 @@ __weak struct serial_device *default_serial_console(void)
 int mxc_serial_setbrg(struct udevice *dev, int baudrate)
 {
        struct mxc_serial_plat *plat = dev_get_plat(dev);
-       u32 clk = imx_get_uartclk();
+       u32 rate = 0;
+
+#if defined(CONFIG_CLK)
+       rate = clk_get_rate(&plat->per_clk);
+#else
+       /* as fallback we try to get the clk rate that way */
+       rate = imx_get_uartclk();
+#endif
 
-       _mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+       _mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
        return 0;
 }
@@ -277,6 +285,11 @@ static int mxc_serial_probe(struct udevice *dev)
 {
        struct mxc_serial_plat *plat = dev_get_plat(dev);
 
+#if defined(CONFIG_CLK)
+       clk_enable(&plat->ipg_clk);
+       clk_enable(&plat->per_clk);
+#endif
+
        _mxc_serial_init(plat->reg, plat->use_dte);
 
        return 0;
@@ -339,6 +352,10 @@ static int mxc_serial_of_to_plat(struct udevice *dev)
 
        plat->use_dte = fdtdec_get_bool(gd->fdt_blob, dev_of_offset(dev),
                                        "fsl,dte-mode");
+#if defined(CONFIG_CLK)
+       clk_get_by_name(dev, "ipg", &plat->ipg_clk);
+       clk_get_by_name(dev, "per", &plat->per_clk);
+#endif
        return 0;
 }
 
(quoted diff) +0 -0

b/include/dm/platform_data/serial_mxc.h
index cc59eeb1dd..330476f816 100644

include/dm/platform_data/serial_mxc.h +4 -0
@@ -10,6 +10,10 @@
 struct mxc_serial_plat {
        struct mxc_uart *reg;  /* address of registers in physical memory */
        bool use_dte;
+#if defined(CONFIG_CLK)
+       struct clk ipg_clk;
+       struct clk per_clk;
+#endif
 };
 
 #endif
Quoted 4 lines

I am guessing the clock composite would have to be expanded to include the UART clocks because from what I can see, they're not included. However, this could potentially eliminate the need to use some of the functions in arch/arm/mach-imx/imx8m/clock_imx8mm.

For the imx8mq I added this and asked Angus to integrate it into his patchset. [1]

Quoted 7 lines

The down-side is that a bunch of SoC's might need to be updated to support more and more clocks, so we could potentially use !CONFIG_IS_ENABLED to use the existing functions as a fall-back.

adam

Quoted 2 lines

Angus

Signature · 3 lines
Heiko

[1] https://lore.kernel.org/u-boot/CAEyMn7bC-p6o6VCu7YWQyXMm+51EcJ5OVXv_625cjDiaR+3TyQ@mail.gmail.com/
#8 lore raw
Quoted 87 lines · Heiko Thiery

Hi Adam,

[SNIP]

Quoted 8 lines

I was thinking that we could expand the struct mxc_serial_plat to include both per and igp clocks to cover devices have clocks that are not the same. The configuring of the platdata can happen using mxc_serial_of_to_plat to 'get' both igp and per clocks. The probe would enable both per and igp to ensure they are operating, then the mxc_serial_setbrg could use clk_get_rate(plat->clk_igp) to determine the clock rate.

With your comment I have made the following, does this fit with your suggestion?

drivers/serial/serial_mxc.c +19 -2
index e4970a169b..7f9f7e8383 100644
@@ -3,6 +3,7 @@
  * (c) 2007 Sascha Hauer <[email protected]>
  */
 
+#include <clk.h>
 #include <common.h>
 #include <dm.h>
 #include <errno.h>
@@ -266,9 +267,16 @@ __weak struct serial_device *default_serial_console(void)
 int mxc_serial_setbrg(struct udevice *dev, int baudrate)
 {
        struct mxc_serial_plat *plat = dev_get_plat(dev);
-       u32 clk = imx_get_uartclk();
+       u32 rate = 0;
+
+#if defined(CONFIG_CLK)
+       rate = clk_get_rate(&plat->per_clk);
+#else
+       /* as fallback we try to get the clk rate that way */
+       rate = imx_get_uartclk();
+#endif
 
-       _mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+       _mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
        return 0;
 }
@@ -277,6 +285,11 @@ static int mxc_serial_probe(struct udevice *dev)
 {
        struct mxc_serial_plat *plat = dev_get_plat(dev);
 
+#if defined(CONFIG_CLK)
+       clk_enable(&plat->ipg_clk);
+       clk_enable(&plat->per_clk);
+#endif
+
        _mxc_serial_init(plat->reg, plat->use_dte);
 
        return 0;
@@ -339,6 +352,10 @@ static int mxc_serial_of_to_plat(struct udevice *dev)
 
        plat->use_dte = fdtdec_get_bool(gd->fdt_blob, dev_of_offset(dev),
                                        "fsl,dte-mode");
+#if defined(CONFIG_CLK)
+       clk_get_by_name(dev, "ipg", &plat->ipg_clk);
+       clk_get_by_name(dev, "per", &plat->per_clk);
+#endif
        return 0;
 }
 
include/dm/platform_data/serial_mxc.h +4 -0
index cc59eeb1dd..330476f816 100644
@@ -10,6 +10,10 @@
 struct mxc_serial_plat {
        struct mxc_uart *reg;  /* address of registers in physical memory */
        bool use_dte;
+#if defined(CONFIG_CLK)
+       struct clk ipg_clk;
+       struct clk per_clk;
+#endif
 };
 
 #endif
Quoted 7 lines
Quoted 4 lines

I am guessing the clock composite would have to be expanded to include the UART clocks because from what I can see, they're not included. However, this could potentially eliminate the need to use some of the functions in arch/arm/mach-imx/imx8m/clock_imx8mm.

For the imx8mq I added this and asked Angus to integrate it into his patchset. [1]

That's basically what I was thinking. It probably wouldn't hurt to add some error handling. What's is not clear to me is the impact of every other IMX platform out there. I am guessing the clk_get_by_name calls will fail if the UART clocks haven't been configured or added to the SoC's respective clock driver. We also might want to check for the presence of plat->ipg_clk and plat->per_clk before requesting their clock rate. If either the UART clocks don't exist or ipg or per don't exist, you'll likely need to fallback on the older functions as well to avoid crashing.

adam

Quoted 13 lines
Quoted 8 lines

The down-side is that a bunch of SoC's might need to be updated to support more and more clocks, so we could potentially use !CONFIG_IS_ENABLED to use the existing functions as a fall-back.

adam

Quoted 2 lines

Angus

--
Heiko

[1] https://lore.kernel.org/u-boot/CAEyMn7bC-p6o6VCu7YWQyXMm+51EcJ5OVXv_625cjDiaR+3TyQ@mail.gmail.com/

#9 lore raw
In reply to Heiko Thiery

Hi Heiko,

Quoted 26 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
  drivers/serial/serial_mxc.c | 15 +++++++++++++--
  1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +1 -0
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
   * (c) 2007 Sascha Hauer <[email protected]>
   */
 
+#include <clk.h>
  #include <common.h>
  #include <dm.h>
  #include <errno.h>

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 272

Quoted 7 lines
drivers/serial/serial_mxc.c +3 -1
@@ -266,9 +267,19 @@ __weak struct serial_device *default_serial_console(void)
  int mxc_serial_setbrg(struct udevice *dev, int baudrate)
  {
  	struct mxc_serial_plat *plat = dev_get_plat(dev);
-	u32 clk = imx_get_uartclk();
+	u32 rate = 0;
+
+	if (IS_ENABLED(CONFIG_CLK)) {

CONFIG_IS_ENABLED?

mx6ull at least does not have CONFIG_SPL_CLK enabled.

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 279

Quoted 7 lines
drivers/serial/serial_mxc.c +7 -0
+		struct clk clk;
+		if(!clk_get_by_name(dev, "ipg", &clk))
+			rate = clk_get_rate(&clk);
+	}
+
+	/* as fallback we try to get the clk rate that way */
+	if (rate == 0)

!rate || IS_ERR_VALUE(rate)

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 285

Quoted 8 lines
drivers/serial/serial_mxc.c +2 -1
+		rate = imx_get_uartclk();
 
-	_mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+	_mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
  	return 0;
  }

--Sean

#10 lore raw
Quoted 10 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 279

Quoted 8 lines
drivers/serial/serial_mxc.c +7 -0
+		struct clk clk;
+		if(!clk_get_by_name(dev, "ipg", &clk))
+			rate = clk_get_rate(&clk);
+	}
+
+	/* as fallback we try to get the clk rate that way */
+	if (rate == 0)

!rate || IS_ERR_VALUE(rate)

This looked so weird I had to actually look at clk_get_rate() in u-boot.

/**
* clk_get_rate() - Get current clock rate.
* @clk: A clock struct that was previously successfully requested by
* clk_request/get_by_*().
*
* Return: clock rate in Hz on success, 0 for invalid clock, or -ve
error code
* for other errors.
*/
ulong clk_get_rate(struct clk *clk);

How can an ulong return a negative error value? What if the clock speed happens to be the same as -errno?

-michael

#11 lore raw
Quoted 24 lines · Michael Walle
Quoted 10 lines
Quoted 8 lines

+        struct clk clk;
+        if(!clk_get_by_name(dev, "ipg", &clk))
+            rate = clk_get_rate(&clk);
+    }
+
+    /* as fallback we try to get the clk rate that way */ +    if (rate == 0)

!rate || IS_ERR_VALUE(rate)

This looked so weird I had to actually look at clk_get_rate() in u-boot.

/**
 * clk_get_rate() - Get current clock rate.
 * @clk:    A clock struct that was previously successfully requested by *        clk_request/get_by_*().
 *
 * Return: clock rate in Hz on success, 0 for invalid clock, or -ve error code *       for other errors.
 */
ulong clk_get_rate(struct clk *clk);

Quoted 1 line

How can an ulong return a negative error value?

It can't. Which is why we need IS_ERR_VALUE and not < 0.

Quoted 1 line

What if the clock speed happens to be the same as -errno?

Due to physical constraints, this isn't an issue on 64 bit systems. And you could run into trouble if you have a 32-bit system with a 4GHz processor.

I don't think it's the best interface. In e.g. Linux clk_get_rate returns 0 on failure, which I think is reasonable. If you are interested, please send a patch series :)

--Sean

#12 lore raw
In reply to Sean Anderson

Hi Sean,

Quoted 40 lines

Hi Heiko,

Quoted 27 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
  drivers/serial/serial_mxc.c | 15 +++++++++++++--
  1 file changed, 13 insertions(+), 2 deletions(-)
(quoted diff) +0 -0

b/drivers/serial/serial_mxc.c index e4970a169b..6fdb2b2397 100644

drivers/serial/serial_mxc.c +1 -0
@@ -3,6 +3,7 @@
   * (c) 2007 Sascha Hauer <[email protected]>
   */
 
+#include <clk.h>
  #include <common.h>
  #include <dm.h>
  #include <errno.h>

drivers/serial/serial_mxc.c __weak struct serial_device · line 272

Quoted 7 lines
drivers/serial/serial_mxc.c +3 -1
@@ -266,9 +267,19 @@ __weak struct serial_device
 *default_serial_console(void) int mxc_serial_setbrg(struct udevice
 *dev, int baudrate) {
  	struct mxc_serial_plat *plat = dev_get_plat(dev);
-	u32 clk = imx_get_uartclk();
+	u32 rate = 0;
+
+	if (IS_ENABLED(CONFIG_CLK)) {  

CONFIG_IS_ENABLED?

Quoted 1 line

mx6ull at least does not have CONFIG_SPL_CLK enabled.

The problem with serial is that not all boards support DM clock in SPL. I'm wondering if this patch has passed the CI tests for all boards.

Quoted 13 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 279

Quoted 8 lines
drivers/serial/serial_mxc.c +7 -0
+		struct clk clk;
+		if(!clk_get_by_name(dev, "ipg", &clk))
+			rate = clk_get_rate(&clk);
+	}
+
+	/* as fallback we try to get the clk rate that way */
+	if (rate == 0)  

!rate || IS_ERR_VALUE(rate)

Quoted 2 lines
drivers/serial/serial_mxc.c +1 -0
+		rate = imx_get_uartclk();
 
Quoted 8 lines
Quoted 7 lines
drivers/serial/serial_mxc.c +1 -1
-	_mxc_serial_setbrg(plat->reg, clk, baudrate,
 plat->use_dte);
+	_mxc_serial_setbrg(plat->reg, rate, baudrate,
 plat->use_dte); 
  	return 0;
  }

--Sean

Best regards,

Lukasz Majewski

--

DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: [email protected]

#13 lore raw
In reply to Sean Anderson

Hi Sean,

Am Do., 17. März 2022 um 15:38 Uhr schrieb Sean Anderson <[email protected]>:

Quoted 25 lines

Hi Heiko,

Quoted 21 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
  drivers/serial/serial_mxc.c | 15 +++++++++++++--
  1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +0 -0
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
   * (c) 2007 Sascha Hauer <[email protected]>
   */
 
Quoted 14 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 272

Quoted 12 lines
drivers/serial/serial_mxc.c +4 -1
+#include <clk.h>
  #include <common.h>
  #include <dm.h>
  #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device *default_serial_console(void)
  int mxc_serial_setbrg(struct udevice *dev, int baudrate)
  {
      struct mxc_serial_plat *plat = dev_get_plat(dev);
-     u32 clk = imx_get_uartclk();
+     u32 rate = 0;
+
+     if (IS_ENABLED(CONFIG_CLK)) {

CONFIG_IS_ENABLED?

This should be correct. The CONFIG_IS_ENABLED is a preprocessor macro and the IS_ENABLED can be used in c code. Or do you mean something else?

Quoted 2 lines

mx6ull at least does not have CONFIG_SPL_CLK enabled.

I expect that in that case it will fallback to the old behavior ... not?

Quoted 20 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 279

Quoted 8 lines
drivers/serial/serial_mxc.c +7 -0
+             struct clk clk;
+             if(!clk_get_by_name(dev, "ipg", &clk))
+                     rate = clk_get_rate(&clk);
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)

!rate || IS_ERR_VALUE(rate)

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 285

Quoted 8 lines
drivers/serial/serial_mxc.c +2 -1
+             rate = imx_get_uartclk();
 
-     _mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+     _mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
      return 0;
  }

--Sean

Signature · 1 line
Heiko
#14 lore raw
Quoted 43 lines · Heiko Thiery

Hi Sean,

Am Do., 17. März 2022 um 15:38 Uhr schrieb Sean Anderson <[email protected]>:

Quoted 25 lines

Hi Heiko,

Quoted 21 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
   drivers/serial/serial_mxc.c | 15 +++++++++++++--
   1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +0 -0
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
    * (c) 2007 Sascha Hauer <[email protected]>
    */
 
Quoted 14 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 272

Quoted 12 lines
drivers/serial/serial_mxc.c +4 -1
+#include <clk.h>
   #include <common.h>
   #include <dm.h>
   #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device *default_serial_console(void)
   int mxc_serial_setbrg(struct udevice *dev, int baudrate)
   {
       struct mxc_serial_plat *plat = dev_get_plat(dev);
-     u32 clk = imx_get_uartclk();
+     u32 rate = 0;
+
+     if (IS_ENABLED(CONFIG_CLK)) {

CONFIG_IS_ENABLED?

Quoted 3 lines

This should be correct. The CONFIG_IS_ENABLED is a preprocessor macro and the IS_ENABLED can be used in c code. Or do you mean something else?

I mean you should be using CONFIG_IS_ENABLED(CLK) so that this code is correct for both SPL and U-Boot proper. But it is also fine to "try and fail" (making this check unnecessary).

Quoted 4 lines
Quoted 2 lines

mx6ull at least does not have CONFIG_SPL_CLK enabled.

I expect that in that case it will fallback to the old behavior ... not?

Yes, if you handle -ENOSYS correctly.

Quoted 4 lines
Quoted 4 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 275

Quoted 4 lines
drivers/serial/serial_mxc.c +3 -0
+             struct clk clk;
+             if(!clk_get_by_name(dev, "ipg", &clk))
+                     rate = clk_get_rate(&clk);

You may also need to enable this clock.

Quoted 17 lines
Quoted 16 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 279

Quoted 4 lines
drivers/serial/serial_mxc.c +4 -0
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)

!rate || IS_ERR_VALUE(rate)

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 285

Quoted 8 lines
drivers/serial/serial_mxc.c +2 -1
+             rate = imx_get_uartclk();
 
-     _mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+     _mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
       return 0;
   }

--Sean

#15 lore raw

Hi Sean,

Am Fr., 18. März 2022 um 03:19 Uhr schrieb Sean Anderson <[email protected]>:

Quoted 58 lines · Heiko Thiery
Quoted 43 lines

Hi Sean,

Am Do., 17. März 2022 um 15:38 Uhr schrieb Sean Anderson <[email protected]>:

Quoted 25 lines

Hi Heiko,

Quoted 21 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
   drivers/serial/serial_mxc.c | 15 +++++++++++++--
   1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +0 -0
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
    * (c) 2007 Sascha Hauer <[email protected]>
    */
 
Quoted 14 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 272

Quoted 12 lines
drivers/serial/serial_mxc.c +4 -1
+#include <clk.h>
   #include <common.h>
   #include <dm.h>
   #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device *default_serial_console(void)
   int mxc_serial_setbrg(struct udevice *dev, int baudrate)
   {
       struct mxc_serial_plat *plat = dev_get_plat(dev);
-     u32 clk = imx_get_uartclk();
+     u32 rate = 0;
+
+     if (IS_ENABLED(CONFIG_CLK)) {

CONFIG_IS_ENABLED?

Quoted 3 lines

This should be correct. The CONFIG_IS_ENABLED is a preprocessor macro and the IS_ENABLED can be used in c code. Or do you mean something else?

I mean you should be using CONFIG_IS_ENABLED(CLK) so that this code is correct for both SPL and U-Boot proper. But it is also fine to "try and fail" (making this check unnecessary).

Quoted 4 lines
Quoted 2 lines

mx6ull at least does not have CONFIG_SPL_CLK enabled.

I expect that in that case it will fallback to the old behavior ... not?

Quoted 8 lines

Yes, if you handle -ENOSYS correctly.

Quoted 4 lines
Quoted 4 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 275

Quoted 4 lines
drivers/serial/serial_mxc.c +3 -0
+             struct clk clk;
+             if(!clk_get_by_name(dev, "ipg", &clk))
+                     rate = clk_get_rate(&clk);

You may also need to enable this clock.

What would be the right place to enable the clock? in mxc_serial_probe()?

Quoted 20 lines
Quoted 18 lines
Quoted 17 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 279

Quoted 5 lines
drivers/serial/serial_mxc.c +4 -0
+     }
+
+     /* as fallback we try to get the clk rate that way */
+     if (rate == 0)

!rate || IS_ERR_VALUE(rate)

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 285

Quoted 8 lines
drivers/serial/serial_mxc.c +2 -1
+             rate = imx_get_uartclk();
 
-     _mxc_serial_setbrg(plat->reg, clk, baudrate, plat->use_dte);
+     _mxc_serial_setbrg(plat->reg, rate, baudrate, plat->use_dte);
 
       return 0;
   }

--Sean

#16 lore raw
Quoted 64 lines · Heiko Thiery

Hi Sean,

Am Fr., 18. März 2022 um 03:19 Uhr schrieb Sean Anderson <[email protected]>:

Quoted 61 lines · Heiko Thiery
Quoted 43 lines

Hi Sean,

Am Do., 17. März 2022 um 15:38 Uhr schrieb Sean Anderson <[email protected]>:

Quoted 25 lines

Hi Heiko,

Quoted 21 lines · Heiko Thiery

With the clock driver enabled for the imx8mq, it was noticed that the frequency used to calculate the baud rate is always taken from the root clock of UART1. This can cause problems if UART1 is not used as console and the settings are different from UART1. The result is that the console output is garbage. To do this correctly the UART frequency is taken from the used device. For the implementations that don't have the igp clock frequency written or can't return it the old way is tried.

---
    drivers/serial/serial_mxc.c | 15 +++++++++++++--
    1 file changed, 13 insertions(+), 2 deletions(-)
drivers/serial/serial_mxc.c +0 -0
index e4970a169b..6fdb2b2397 100644
@@ -3,6 +3,7 @@
     * (c) 2007 Sascha Hauer <[email protected]>
     */
 
Quoted 14 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 272

Quoted 12 lines
drivers/serial/serial_mxc.c +4 -1
+#include <clk.h>
    #include <common.h>
    #include <dm.h>
    #include <errno.h>
@@ -266,9 +267,19 @@ __weak struct serial_device *default_serial_console(void)
    int mxc_serial_setbrg(struct udevice *dev, int baudrate)
    {
        struct mxc_serial_plat *plat = dev_get_plat(dev);
-     u32 clk = imx_get_uartclk();
+     u32 rate = 0;
+
+     if (IS_ENABLED(CONFIG_CLK)) {

CONFIG_IS_ENABLED?

Quoted 3 lines

This should be correct. The CONFIG_IS_ENABLED is a preprocessor macro and the IS_ENABLED can be used in c code. Or do you mean something else?

I mean you should be using CONFIG_IS_ENABLED(CLK) so that this code is correct for both SPL and U-Boot proper. But it is also fine to "try and fail" (making this check unnecessary).

Quoted 4 lines
Quoted 2 lines

mx6ull at least does not have CONFIG_SPL_CLK enabled.

I expect that in that case it will fallback to the old behavior ... not?

Yes, if you handle -ENOSYS correctly.

Quoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
Quoted 3 lines

drivers/serial/serial_mxc.c __weak struct serial_device *default_serial_console(void) · line 275

Quoted 3 lines
drivers/serial/serial_mxc.c +3 -0
+             struct clk clk;
+             if(!clk_get_by_name(dev, "ipg", &clk))
+                     rate = clk_get_rate(&clk);

You may also need to enable this clock.

What would be the right place to enable the clock? in mxc_serial_probe()?

Yes.

--Sean