Skip to contentSkip to conversation

clk: fixed-rate: Use "clock-output-names" to name fixed clocks

PATCH In review Fabio Estevam 3 participants 20 messages Apr 17, 2025 – Apr 25, 2025

Subsystems (4)
lore
Share & tools
lore

Fabio Estevam

author reviewer
#0 lore raw

PATCH v2 1/1 clk: fixed-rate: Use "clock-output-names" to name fixed clocks

From: Fabio Estevam <[email protected]>

Currently, fixed-rate clocks in U-Boot are named based on their devicetree node names. For example, given the following node:

osc_24m: clock-osc-24m {
	compatible = "fixed-clock";
	#clock-cells = <0>;
	clock-frequency = <24000000>;
	clock-output-names = "osc_24m";
};

U-Boot registers the clock as "clock-osc-24m", derived from the node name, ignoring the clock-output-names property.

This differs from Linux, which uses the clock-output-names property when assigning clock names. As a result, clock consumers expecting names like "osc_24m" (as defined in the property) may fail to resolve clocks correctly in U-Boot.

Update the fixed-rate clock driver to check for the clock-output-names property and use it as the clock name if present. If not, the fallback remains the node name. This makes U-Boot behavior consistent with Linux.

One concrete impact is on i.MX8MP, where USB clock lookup failed following commit b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*").

With this change applied, fixed-clocks are correctly registered with their expected names:

u-boot=> clk dump
Rate Usecnt Name
------------------------------------------

 32768                0        |-- osc_32k
 24000000             5        |-- osc_24m

This change restores i.MX8MP USB functionality by ensuring the proper clock names are used.

Fixes b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*")

---
Changes since v1:
- Included <dm/read.h> that defines dev_read_string_index() instead of relying on indirect inclusion via <dm.h>.
- Removed superflous parenthesis around dev_read_string_index().
- Added Adam's Reviewed-by tag.

 drivers/clk/clk_fixed_rate.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)
drivers/clk/clk_fixed_rate.c +13 -0
index d1da05cc18a5..648e31b4fe14 100644
@@ -9,6 +9,7 @@
 #include <dm.h>
 #include <log.h>
 #include <dm/device-internal.h>
+#include <dm/read.h>
 #include <linux/clk-provider.h>
 
 #define UBOOT_DM_CLK_FIXED_RATE "fixed_rate_clock"
@@ -35,6 +36,8 @@ void clk_fixed_rate_ofdata_to_plat_(struct udevice *dev,
 				    struct clk_fixed_rate *plat)
 {
 	struct clk *clk = &plat->clk;
+	const char *clk_name;
+
 	if (CONFIG_IS_ENABLED(OF_REAL))
 		plat->fixed_rate = dev_read_u32_default(dev, "clock-frequency",
 							0);
@@ -45,6 +48,16 @@ void clk_fixed_rate_ofdata_to_plat_(struct udevice *dev,
 
 	clk->dev = dev;
 	clk->enable_count = 0;
+
+	/*
+	 * If the "clock-output-names" property is present, use it
+	 * as the clock name, like it is done in Linux.
+	 * Otherwise, the original behavior will be preserved: the fixed clock
+	 * name will be its node name.
+	 *
+	 */
+	if (!dev_read_string_index(dev, "clock-output-names", 0, &clk_name))
+		dev->name = clk_name;
 }
 
 static ulong clk_fixed_rate_raw_get_rate(struct clk *clk)
Signature · 1 line
2.34.1
#1 lore raw
Quoted 15 lines · Fabio Estevam

From: Fabio Estevam <[email protected]>

Currently, fixed-rate clocks in U-Boot are named based on their devicetree node names. For example, given the following node:

osc_24m: clock-osc-24m {
	compatible = "fixed-clock";
	#clock-cells = <0>;
	clock-frequency = <24000000>;
	clock-output-names = "osc_24m";
};

U-Boot registers the clock as "clock-osc-24m", derived from the node name, ignoring the clock-output-names property.

Quoted 8 lines

This differs from Linux, which uses the clock-output-names property when assigning clock names. As a result, clock consumers expecting names like "osc_24m" (as defined in the property) may fail to resolve clocks correctly in U-Boot.

Update the fixed-rate clock driver to check for the clock-output-names property and use it as the clock name if present. If not, the fallback remains the node name. This makes U-Boot behavior consistent with Linux.

This part above ^ is fine.

This part below v is wrong .

Quoted 6 lines

One concrete impact is on i.MX8MP, where USB clock lookup failed following commit b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*").

With this change applied, fixed-clocks are correctly registered with their expected names:

Quoted 8 lines

u-boot=> clk dump
Rate Usecnt Name
------------------------------------------

  32768                0        |-- osc_32k
  24000000             5        |-- osc_24m

This change restores i.MX8MP USB functionality by ensuring the proper clock names are used.

This is wrong, because this is NOT how the clock look up works at all.

The lookup works this way. Take for example:

drivers/clk/imx/clk-imx8mp.c

240 clk_dm(IMX8MP_ARM_PLL_REF_SEL, imx_clk_mux(dev, "arm_pll_ref_sel", base + 0x84, 0, 2, pll_ref_sels, ARRAY_SIZE(pll_ref_sels)));

pll_ref_sels:

21 static const char * const pll_ref_sels[] = { "osc_24m", "dummy",
"dummy", "dummy", };

the "osc_24m" is looked up in the clock-controller@30380000 clock-names property:

dts/upstream/src/arm64/freescale/imx8mp.dtsi

745 clk: clock-controller@30380000 {
...
751 clocks = <&osc_32k>, <&osc_24m>, <&clk_ext1>, <&clk_ext2>,
752 <&clk_ext3>, <&clk_ext4>;
753 clock-names = "osc_32k", "osc_24m", ...

                                       ^^^^^^^

The offset of osc_24m in that clock-names property is 1 , so the phandle in clocks property array at offset 1 is used to look up those clock:

745 clk: clock-controller@30380000 {
...
751 clocks = <&osc_32k>, <&osc_24m>,

                                   ^^^^^^^^

Which leads to this clock node in DT:

 188         osc_24m: clock-osc-24m {
             ^^^^^^^^

189 compatible = "fixed-clock";
190 #clock-cells = <0>;

The "clock-output-names" is NEVER used for any clock look up.

...

I am fine with the first half of the description, but the second half is wrong and this actually does not fix anything. Simply remove clock-output-names from the osc_24m and the code will be broken again.

The clk_resolve_parent_clk() is meant to perform the aforementioned resolution(), that is what "[PATCH v2 00/24] clk: Add clk_resolve_parent_clk() and fix up iMX clock drivers" actually fixed. It seems there is something still missing and clk_resolve_parent_clk() does not do the resolution properly for some clock, but that is what needs to be understood and fixed, not worked around this way.

#2 lore raw
Quoted 80 lines · Marek Vasut
Quoted 15 lines · Fabio Estevam

From: Fabio Estevam <[email protected]>

Currently, fixed-rate clocks in U-Boot are named based on their devicetree node names. For example, given the following node:

osc_24m: clock-osc-24m {
      compatible = "fixed-clock";
      #clock-cells = <0>;
      clock-frequency = <24000000>;
      clock-output-names = "osc_24m";
};

U-Boot registers the clock as "clock-osc-24m", derived from the node name, ignoring the clock-output-names property.

Quoted 8 lines

This differs from Linux, which uses the clock-output-names property when assigning clock names. As a result, clock consumers expecting names like "osc_24m" (as defined in the property) may fail to resolve clocks correctly in U-Boot.

Update the fixed-rate clock driver to check for the clock-output-names property and use it as the clock name if present. If not, the fallback remains the node name. This makes U-Boot behavior consistent with Linux.

This part above ^ is fine.

This part below v is wrong .

Quoted 6 lines

One concrete impact is on i.MX8MP, where USB clock lookup failed following commit b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*").

With this change applied, fixed-clocks are correctly registered with their expected names:

Quoted 8 lines

u-boot=> clk dump
Rate Usecnt Name
------------------------------------------

  32768                0        |-- osc_32k
  24000000             5        |-- osc_24m

This change restores i.MX8MP USB functionality by ensuring the proper clock names are used.

This is wrong, because this is NOT how the clock look up works at all.

The lookup works this way. Take for example:

drivers/clk/imx/clk-imx8mp.c

240 clk_dm(IMX8MP_ARM_PLL_REF_SEL, imx_clk_mux(dev,
"arm_pll_ref_sel", base + 0x84, 0, 2, pll_ref_sels, ARRAY_SIZE(pll_ref_sels)));

pll_ref_sels:

21 static const char * const pll_ref_sels[] = { "osc_24m", "dummy",
"dummy", "dummy", };

the "osc_24m" is looked up in the clock-controller@30380000 clock-names property:

dts/upstream/src/arm64/freescale/imx8mp.dtsi

745 clk: clock-controller@30380000 {
...
751 clocks = <&osc_32k>, <&osc_24m>, <&clk_ext1>, <&clk_ext2>,
752 <&clk_ext3>, <&clk_ext4>;
753 clock-names = "osc_32k", "osc_24m", ...

                                        ^^^^^^^

The offset of osc_24m in that clock-names property is 1 , so the phandle in clocks property array at offset 1 is used to look up those clock:

745 clk: clock-controller@30380000 {
...
751 clocks = <&osc_32k>, <&osc_24m>,

                                    ^^^^^^^^
Quoted 8 lines

Which leads to this clock node in DT:

  188         osc_24m: clock-osc-24m {
              ^^^^^^^^
  189                 compatible = "fixed-clock";
  190                 #clock-cells = <0>;

The "clock-output-names" is NEVER used for any clock look up.

For what it's worth, I remove the "clock-out-names" from the device tree in Linux, and the Linux driver properly coped with it, but I am curious to know what the point of this entry is.

Quoted 7 lines

...

I am fine with the first half of the description, but the second half is wrong and this actually does not fix anything. Simply remove clock-output-names from the osc_24m and the code will be broken again.

Quoted 6 lines

The clk_resolve_parent_clk() is meant to perform the aforementioned resolution(), that is what "[PATCH v2 00/24] clk: Add clk_resolve_parent_clk() and fix up iMX clock drivers" actually fixed. It seems there is something still missing and clk_resolve_parent_clk() does not do the resolution properly for some clock, but that is what needs to be understood and fixed, not worked around this way.

I added some debug code to clk_resolve_parent_clk it does return osc_24m as the name of the clock, but when searching for index, it appears to fail. Do you have any suggestion on how searching for the index can query up the chain to the parent CCM driver on how to return the reference to proper clock? The clock-names property is at the CCM parent and not inside the children, and the number of levels vary depending on the child clocks, you can't assume the parent's parent has the 'clock-name's property.

adam

#3 lore raw
Quoted 88 lines · Adam Ford
Quoted 87 lines · Marek Vasut
Quoted 15 lines · Fabio Estevam

From: Fabio Estevam <[email protected]>

Currently, fixed-rate clocks in U-Boot are named based on their devicetree node names. For example, given the following node:

osc_24m: clock-osc-24m {
       compatible = "fixed-clock";
       #clock-cells = <0>;
       clock-frequency = <24000000>;
       clock-output-names = "osc_24m";
};

U-Boot registers the clock as "clock-osc-24m", derived from the node name, ignoring the clock-output-names property.

Quoted 8 lines

This differs from Linux, which uses the clock-output-names property when assigning clock names. As a result, clock consumers expecting names like "osc_24m" (as defined in the property) may fail to resolve clocks correctly in U-Boot.

Update the fixed-rate clock driver to check for the clock-output-names property and use it as the clock name if present. If not, the fallback remains the node name. This makes U-Boot behavior consistent with Linux.

This part above ^ is fine.

This part below v is wrong .

Quoted 6 lines

One concrete impact is on i.MX8MP, where USB clock lookup failed following commit b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*").

With this change applied, fixed-clocks are correctly registered with their expected names:

Quoted 8 lines

u-boot=> clk dump
Rate Usecnt Name
------------------------------------------

   32768                0        |-- osc_32k
   24000000             5        |-- osc_24m

This change restores i.MX8MP USB functionality by ensuring the proper clock names are used.

This is wrong, because this is NOT how the clock look up works at all.

The lookup works this way. Take for example:

drivers/clk/imx/clk-imx8mp.c

240 clk_dm(IMX8MP_ARM_PLL_REF_SEL, imx_clk_mux(dev,
"arm_pll_ref_sel", base + 0x84, 0, 2, pll_ref_sels, ARRAY_SIZE(pll_ref_sels)));

pll_ref_sels:

   21 static const char * const pll_ref_sels[] = { "osc_24m", "dummy",

"dummy", "dummy", };

the "osc_24m" is looked up in the clock-controller@30380000 clock-names property:

dts/upstream/src/arm64/freescale/imx8mp.dtsi

   745 clk: clock-controller@30380000 {
...
   751         clocks = <&osc_32k>, <&osc_24m>, <&clk_ext1>, <&clk_ext2>,
   752                  <&clk_ext3>, <&clk_ext4>;
   753         clock-names = "osc_32k", "osc_24m", ...
                                         ^^^^^^^

The offset of osc_24m in that clock-names property is 1 , so the phandle in clocks property array at offset 1 is used to look up those clock:

   745 clk: clock-controller@30380000 {
...
   751         clocks = <&osc_32k>, <&osc_24m>,
                                     ^^^^^^^^

Which leads to this clock node in DT:

   188         osc_24m: clock-osc-24m {
               ^^^^^^^^
   189                 compatible = "fixed-clock";
   190                 #clock-cells = <0>;
Quoted 5 lines
Quoted 1 line

The "clock-output-names" is NEVER used for any clock look up.

For what it's worth, I remove the "clock-out-names" from the device tree in Linux, and the Linux driver properly coped with it, but I am curious to know what the point of this entry is.

DT schema dtschema/schemas/clock/clock.yaml says:

82 clock-output-names:
83 description: |

 84       Recommended to be a list of strings of clock output signal
 85       names indexed by the first cell in the clock specifier.
 86       However, the meaning of clock-output-names is domain
 87       specific to the clock provider, and is only provided to
 88       encourage using the same meaning for the majority of clock
 89       providers.  This format may not work for clock providers
 90       using a complex clock specifier format.  In those cases it
 91       is recommended to omit this property and create a binding
 92       specific names property.

93
94 Clock consumer nodes must never directly reference
95 the provider\'s clock-output-names property.

Quoted 14 lines
Quoted 7 lines

...

I am fine with the first half of the description, but the second half is wrong and this actually does not fix anything. Simply remove clock-output-names from the osc_24m and the code will be broken again.

Quoted 6 lines

The clk_resolve_parent_clk() is meant to perform the aforementioned resolution(), that is what "[PATCH v2 00/24] clk: Add clk_resolve_parent_clk() and fix up iMX clock drivers" actually fixed. It seems there is something still missing and clk_resolve_parent_clk() does not do the resolution properly for some clock, but that is what needs to be understood and fixed, not worked around this way.

Quoted 3 lines

I added some debug code to clk_resolve_parent_clk it does return osc_24m as the name of the clock, but when searching for index, it appears to fail.

What exactly fails and where does it fail ?

Quoted 6 lines

Do you have any suggestion on how searching for the
index can query up the chain to the parent CCM driver on how to return the reference to proper clock? The clock-names property is at the CCM parent and not inside the children, and the number of levels vary depending on the child clocks, you can't assume the parent's parent has the 'clock-name's property.

It is always the clock-controller@30380000 node to which the drivers/clk/imx/clk-imx8mp.c is bound , so drivers/clk/imx/clk-imx8mp.c does look up in the clock-controller@30380000 node clock-names and clocks . And from there on, it is always on phandle to &osc_24m , right ?

author reviewer
#4 lore raw
In reply to Marek Vasut

Hi Marek,

Quoted 1 line · Marek Vasut

The "clock-output-names" is NEVER used for any clock look up.

In Linux, "clock-output-names" is used to register the name of the fixed-rate clock.

From Linux drivers/clk/clk-fixed-rate.c:

```
static struct clk_hw *_of_fixed_clk_setup(struct device_node *node)
{
....

of_property_read_string(node, "clock-output-names", &clk_name);

hw = clk_hw_register_fixed_rate_with_accuracy(NULL, clk_name, NULL,
    0, rate, accuracy);
```

In U-Boot, we should register a fixed-rate clock with the same name as in Linux.

This patch aims to make the U-Boot fixed-rate clock name conform to Linux standards.

Quoted 3 lines

I am fine with the first half of the description, but the second half is wrong and this actually does not fix anything. Simply remove clock-output-names from the osc_24m and the code will be broken again.

In Linux, if "clock-output-names" is removed from the osc_24m node, the console becomes corrupted
on an imx8mp-evk.

Quoted 6 lines

The clk_resolve_parent_clk() is meant to perform the aforementioned resolution(), that is what "[PATCH v2 00/24] clk: Add clk_resolve_parent_clk() and fix up iMX clock drivers" actually fixed. It seems there is something still missing and clk_resolve_parent_clk() does not do the resolution properly for some clock, but that is what needs to be understood and fixed, not worked around this way.

Please try to reproduce the problem on your end.

On an imx8mp-based board, run 'usb start' in U-Boot, and this command will fail.

Thanks

#5 lore raw
Quoted 43 lines · Fabio Estevam

Hi Marek,

Quoted 1 line · Marek Vasut

The "clock-output-names" is NEVER used for any clock look up.

In Linux, "clock-output-names" is used to register the name of the fixed-rate clock.

From Linux drivers/clk/clk-fixed-rate.c:

```
static struct clk_hw *_of_fixed_clk_setup(struct device_node *node)
{
....

of_property_read_string(node, "clock-output-names", &clk_name);

hw = clk_hw_register_fixed_rate_with_accuracy(NULL, clk_name, NULL,
    0, rate, accuracy);
```

In U-Boot, we should register a fixed-rate clock with the same name as in Linux.

This patch aims to make the U-Boot fixed-rate clock name conform to Linux standards.

Quoted 3 lines

I am fine with the first half of the description, but the second half is wrong and this actually does not fix anything. Simply remove clock-output-names from the osc_24m and the code will be broken again.

In Linux, if "clock-output-names" is removed from the osc_24m node, the console becomes corrupted
on an imx8mp-evk.

Quoted 6 lines

The clk_resolve_parent_clk() is meant to perform the aforementioned resolution(), that is what "[PATCH v2 00/24] clk: Add clk_resolve_parent_clk() and fix up iMX clock drivers" actually fixed. It seems there is something still missing and clk_resolve_parent_clk() does not do the resolution properly for some clock, but that is what needs to be understood and fixed, not worked around this way.

Quoted 3 lines

Please try to reproduce the problem on your end.

On an imx8mp-based board, run 'usb start' in U-Boot, and this command will fail.

From what I can tell, on the imx8mp, the usb_phy_ref parent->dev->name value s "clock-osc-24m" which never matches "osc_24" and I put some debug code into clk_mux_fetch_parent_index.

I think clk_mux_fetch_parent_index needs some additional logic to go up the chain to the root node and look for 'clock-names' to match in addition to what it already does. I am just not sure of the best method to go up the tree to find the root node that defines it.

adam

Quoted 2 lines

Thanks

#6 lore raw
Quoted 48 lines · Adam Ford
Quoted 43 lines · Fabio Estevam

Hi Marek,

Quoted 1 line · Marek Vasut

The "clock-output-names" is NEVER used for any clock look up.

In Linux, "clock-output-names" is used to register the name of the fixed-rate clock.

From Linux drivers/clk/clk-fixed-rate.c:

```
static struct clk_hw *_of_fixed_clk_setup(struct device_node *node)
{
....

of_property_read_string(node, "clock-output-names", &clk_name);

hw = clk_hw_register_fixed_rate_with_accuracy(NULL, clk_name, NULL,
     0, rate, accuracy);
```

In U-Boot, we should register a fixed-rate clock with the same name as in Linux.

This patch aims to make the U-Boot fixed-rate clock name conform to Linux standards.

Quoted 3 lines

I am fine with the first half of the description, but the second half is wrong and this actually does not fix anything. Simply remove clock-output-names from the osc_24m and the code will be broken again.

In Linux, if "clock-output-names" is removed from the osc_24m node, the console becomes corrupted
on an imx8mp-evk.

Quoted 6 lines

The clk_resolve_parent_clk() is meant to perform the aforementioned resolution(), that is what "[PATCH v2 00/24] clk: Add clk_resolve_parent_clk() and fix up iMX clock drivers" actually fixed. It seems there is something still missing and clk_resolve_parent_clk() does not do the resolution properly for some clock, but that is what needs to be understood and fixed, not worked around this way.

Quoted 3 lines

Please try to reproduce the problem on your end.

On an imx8mp-based board, run 'usb start' in U-Boot, and this command will fail.

Quoted 8 lines

From what I can tell, on the imx8mp, the usb_phy_ref parent->dev->name
value s "clock-osc-24m" which never matches "osc_24" and I put some debug code into clk_mux_fetch_parent_index.

I think clk_mux_fetch_parent_index needs some additional logic to go up the chain to the root node and look for 'clock-names' to match in addition to what it already does. I am just not sure of the best method to go up the tree to find the root node that defines it.

[...]

Try something like this (also attached), maybe this needs to be made generic ?

drivers/clk/clk-composite.c +2 -0
index 9e3b5191767..207224b1320 100644
@@ -155,6 +155,8 @@ struct clk *clk_register_composite(struct udevice *dev, const char *name,
 		goto err;
 	}
 
+	composite->dev = dev;
+
 	if (composite->mux)
 		composite->mux->dev = clk->dev;
 	if (composite->rate)
drivers/clk/imx/clk-composite-8m.c +29 -1
index 14c5b92939c..e1a3c0af308 100644
@@ -116,14 +116,42 @@ static const struct clk_ops imx8m_clk_composite_divider_ops = {
 	.set_rate = imx8m_clk_composite_divider_set_rate,
 };
 
+static int imx8m_clk_mux_fetch_parent_index(struct udevice *cdev, struct clk *clk, struct clk *parent)
+{
+	struct clk_mux *mux = to_clk_mux(clk);
+	struct clk cclk;
+	int ret;
+	int i;
+
+	if (!parent)
+		return -EINVAL;
+
+	for (i = 0; i < mux->num_parents; i++) {
+		ret = clk_get_by_name(cdev, mux->parent_names[i], &cclk);
+		if (!ret && ofnode_equal(dev_ofnode(parent->dev), dev_ofnode(cclk.dev)))
+			return i;
+
+		if (!strcmp(parent->dev->name, mux->parent_names[i]))
+			return i;
+		if (!strcmp(parent->dev->name,
+			    clk_resolve_parent_clk(clk->dev,
+						   mux->parent_names[i])))
+			return i;
+	}
+
+	return -EINVAL;
+}
+
+
 static int imx8m_clk_mux_set_parent(struct clk *clk, struct clk *parent)
 {
 	struct clk_mux *mux = to_clk_mux(clk);
+	struct clk_composite *composite = (struct clk_composite *)clk->data;
 	int index;
 	u32 val;
 	u32 reg;
 
-	index = clk_mux_fetch_parent_index(clk, parent);
+	index = imx8m_clk_mux_fetch_parent_index(composite->dev, clk, parent);
 	if (index < 0) {
 		log_err("Could not fetch index\n");
 		return index;
include/linux/clk-provider.h +2 -0
index 5ea2171492e..267757939e0 100644
@@ -219,6 +219,8 @@ struct clk_composite {
 	const struct clk_ops	*mux_ops;
 	const struct clk_ops	*rate_ops;
 	const struct clk_ops	*gate_ops;
+
+	struct udevice *dev;
 };
 
 #define to_clk_composite(_clk) container_of(_clk, struct 

clk_composite, clk)

drivers/clk/clk-composite.c +2 -0
index 9e3b5191767..207224b1320 100644
@@ -155,6 +155,8 @@ struct clk *clk_register_composite(struct udevice *dev, const char *name,
 		goto err;
 	}
 
+	composite->dev = dev;
+
 	if (composite->mux)
 		composite->mux->dev = clk->dev;
 	if (composite->rate)
drivers/clk/imx/clk-composite-8m.c +29 -1
index 14c5b92939c..e1a3c0af308 100644
@@ -116,14 +116,42 @@ static const struct clk_ops imx8m_clk_composite_divider_ops = {
 	.set_rate = imx8m_clk_composite_divider_set_rate,
 };
 
+static int imx8m_clk_mux_fetch_parent_index(struct udevice *cdev, struct clk *clk, struct clk *parent)
+{
+	struct clk_mux *mux = to_clk_mux(clk);
+	struct clk cclk;
+	int ret;
+	int i;
+
+	if (!parent)
+		return -EINVAL;
+
+	for (i = 0; i < mux->num_parents; i++) {
+		ret = clk_get_by_name(cdev, mux->parent_names[i], &cclk);
+		if (!ret && ofnode_equal(dev_ofnode(parent->dev), dev_ofnode(cclk.dev)))
+			return i;
+
+		if (!strcmp(parent->dev->name, mux->parent_names[i]))
+			return i;
+		if (!strcmp(parent->dev->name,
+			    clk_resolve_parent_clk(clk->dev,
+						   mux->parent_names[i])))
+			return i;
+	}
+
+	return -EINVAL;
+}
+
+
 static int imx8m_clk_mux_set_parent(struct clk *clk, struct clk *parent)
 {
 	struct clk_mux *mux = to_clk_mux(clk);
+	struct clk_composite *composite = (struct clk_composite *)clk->data;
 	int index;
 	u32 val;
 	u32 reg;
 
-	index = clk_mux_fetch_parent_index(clk, parent);
+	index = imx8m_clk_mux_fetch_parent_index(composite->dev, clk, parent);
 	if (index < 0) {
 		log_err("Could not fetch index\n");
 		return index;
include/linux/clk-provider.h +2 -0
index 5ea2171492e..267757939e0 100644
@@ -219,6 +219,8 @@ struct clk_composite {
 	const struct clk_ops	*mux_ops;
 	const struct clk_ops	*rate_ops;
 	const struct clk_ops	*gate_ops;
+
+	struct udevice *dev;
 };
 
 #define to_clk_composite(_clk) container_of(_clk, struct clk_composite, clk)
#8 lore raw
Quoted 8 lines · Fabio Estevam
Quoted 2 lines · Marek Vasut

Try something like this (also attached), maybe this needs to be made generic ?

It works for me:

Can you send that as a fix ?

And also, fix this patch up and drop the bogus second half of the commit message and do a V3.

Thanks

Tested-by
#9 lore raw
Quoted 6 lines · Marek Vasut
Quoted 4 lines · Fabio Estevam
Quoted 2 lines · Marek Vasut

Try something like this (also attached), maybe this needs to be made generic ?

Thanks for doing this.

Quoted 9 lines
Quoted 4 lines

It works for me:

Can you send that as a fix ?

And also, fix this patch up and drop the bogus second half of the commit message and do a V3.

Feel free to add:

adam

Quoted 2 lines

Thanks

author reviewer
#10 lore raw
In reply to Marek Vasut

Hi Marek,

Quoted 4 lines · Marek Vasut

Can you send that as a fix ?

And also, fix this patch up and drop the bogus second half of the commit message and do a V3.

Sorry for the delay.

I found another issue on an imx8mm-evk board.

U-Boot SPL 2025.04-01379-g10f48365112b-dirty (Apr 24 2025 - 22:56:59 -0300) No pmic

WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
SEC0:  RNG instantiated

Trying to boot from MMC1

(Pause here for about 3 seconds)

NOTICE:  Do not release JR0 to NS as it can be used by HAB
NOTICE:  BL31: v2.12.0(release):v2.12.0
NOTICE:  BL31: Built : 15:28:24, Apr 24 2025

U-Boot 2025.04-01379-g10f48365112b-dirty (Apr 24 2025 - 22:56:59 -0300)

CPU:   NXP i.MX8MMQ Rev1.0 A53 at 1200 MHz
CPU:   Consumer temperature grade (0C to 95C) at 37C
Model: FSL i.MX8MM EVK board
DRAM:  2 GiB
Core:  188 devices, 24 uclasses, devicetree: separate
WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
MMC:   FSL_SDHC: 1, FSL_SDHC: 2
Loading Environment from MMC... Reading from MMC(1)... OK
In:    serial@30890000
Out:   serial@30890000
Err:   serial@30890000
SEC0:  RNG instantiated
Net:   eth0: ethernet@30be0000
Hit any key to stop autoboot:  0

u-boot=>

The issue above is the long pause after "Trying to boot from MMC1".

If I manually revert commit b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*"), then it boots normally, without the long pause.

Applying your patch helped resolve the USB issue, but not the long pause issue.

Thanks

#11 lore raw
Quoted 44 lines · Fabio Estevam

Hi Marek,

Quoted 4 lines · Marek Vasut

Can you send that as a fix ?

And also, fix this patch up and drop the bogus second half of the commit message and do a V3.

Sorry for the delay.

I found another issue on an imx8mm-evk board.

U-Boot SPL 2025.04-01379-g10f48365112b-dirty (Apr 24 2025 - 22:56:59 -0300) No pmic

WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
SEC0:  RNG instantiated

Trying to boot from MMC1

(Pause here for about 3 seconds)

NOTICE:  Do not release JR0 to NS as it can be used by HAB
NOTICE:  BL31: v2.12.0(release):v2.12.0
NOTICE:  BL31: Built : 15:28:24, Apr 24 2025

U-Boot 2025.04-01379-g10f48365112b-dirty (Apr 24 2025 - 22:56:59 -0300)

CPU:   NXP i.MX8MMQ Rev1.0 A53 at 1200 MHz
CPU:   Consumer temperature grade (0C to 95C) at 37C
Model: FSL i.MX8MM EVK board
DRAM:  2 GiB
Core:  188 devices, 24 uclasses, devicetree: separate
WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
MMC:   FSL_SDHC: 1, FSL_SDHC: 2
Loading Environment from MMC... Reading from MMC(1)... OK
In:    serial@30890000
Out:   serial@30890000
Err:   serial@30890000
SEC0:  RNG instantiated
Net:   eth0: ethernet@30be0000
Hit any key to stop autoboot:  0

u-boot=>

Quoted 7 lines

The issue above is the long pause after "Trying to boot from MMC1".

If I manually revert commit b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*"), then it boots normally, without the long pause.

Applying your patch helped resolve the USB issue, but not the long pause issue.

Probably some timeout in the esdhc driver, maybe some clock are not enabled ? I don't see it on the MX8MP.

Did you already apply
Re: [PATCH] clk: imx: Finish converting clock-osc-24 back to osc_24

?

author reviewer
#12 lore raw
Quoted 2 lines · Marek Vasut

Probably some timeout in the esdhc driver, maybe some clock are not enabled ? I don't see it on the MX8MP.

Same here. I didn't see this error on an imx8mp-evk.

Quoted 3 lines

Did you already apply
Re: [PATCH] clk: imx: Finish converting clock-osc-24 back to osc_24

Yes, I applied this one locally, but it didn't make a difference.

I haven't had a chance to debug this issue yet.

#13 lore raw
Quoted 7 lines · Fabio Estevam
Quoted 2 lines · Marek Vasut

Probably some timeout in the esdhc driver, maybe some clock are not enabled ? I don't see it on the MX8MP.

Same here. I didn't see this error on an imx8mp-evk.

If memory serves, I thought SPL used the Boot ROM on Plus and Nano to load U-Boot where the Mini has an older boot ROM and required more driver initialization.

I could be wrong. I'll try to test my Mini this weekend.

adam

Quoted 8 lines
Quoted 4 lines

Did you already apply
Re: [PATCH] clk: imx: Finish converting clock-osc-24 back to osc_24

Yes, I applied this one locally, but it didn't make a difference.

I haven't had a chance to debug this issue yet.

#14 lore raw
Quoted 8 lines · Adam Ford
Quoted 6 lines · Fabio Estevam
Quoted 2 lines · Marek Vasut

Probably some timeout in the esdhc driver, maybe some clock are not enabled ? I don't see it on the MX8MP.

Quoted 7 lines
Quoted 1 line

Same here. I didn't see this error on an imx8mp-evk.

If memory serves, I thought SPL used the Boot ROM on Plus and Nano to load U-Boot where the Mini has an older boot ROM and required more driver initialization.

I could be wrong. I'll try to test my Mini this weekend.

I pulled my mini out. With Marek's clk.diff file applied on top of U-Boot next, it boots quickly without delay, and I don't get a PMIC error. I don't have an EVK at my home lab to test.

U-Boot SPL 2025.04-rc5-00835-g61128a5b704d-dirty (Apr 24 2025 - 21:53:22 -0500)
WDT: Not starting watchdog@30280000
Trying to boot from MMC1

NOTICE:  Do not release JR0 to NS as it can be used by HAB
NOTICE:  BL31: v2.10.14(release):lts-v2.10.14-380-gd7da75fef
NOTICE:  BL31: Built : 20:12:47, Apr  3 2025

U-Boot 2025.04-rc5-00835-g61128a5b704d-dirty (Apr 24 2025 - 21:53:22 -0500)

CPU:   NXP i.MX8MMQ Rev1.0 A53 at 1200 MHz
CPU:   Industrial temperature grade (-40C to 105C) at 30C
Model: Beacon EmbeddedWorks i.MX8M Mini Development Kit
DRAM:  2 GiB
Core:  225 devices, 26 uclasses, devicetree: separate
WDT:   Not starting watchdog@30280000
MMC:   FSL_SDHC: 0, FSL_SDHC: 1, FSL_SDHC: 2
Loading Environment from MMC... Reading from MMC(2)... OK
In:    serial@30890000
Out:   serial@30890000
Err:   serial@30890000
SEC0:  RNG instantiated
Net:   eth0: ethernet@30be0000
Hit any key to stop autoboot:  0

u-boot=>

Quoted 10 lines

adam

Quoted 8 lines
Quoted 4 lines

Did you already apply
Re: [PATCH] clk: imx: Finish converting clock-osc-24 back to osc_24

Yes, I applied this one locally, but it didn't make a difference.

I haven't had a chance to debug this issue yet.

author reviewer
#15 lore raw
In reply to Fabio Estevam
Quoted 1 line · Fabio Estevam

I haven't had a chance to debug this issue yet.

I applied this debug patch:

drivers/mmc/fsl_esdhc_imx.c +1 -0
@@ -1539,6 +1539,7 @@ static int fsl_esdhc_probe(struct udevice *dev)
        }
 
        priv->sdhc_clk = clk_get_rate(&priv->per_clk);
+       printf("****** sdhc clk is %d\n", priv->sdhc_clk);
 #else
        init_clk_usdhc(dev_seq(dev));
 

With the top-of-tree U-Boot, SPL calculates the wrong sdhc_clk on the imx8mm-evk:

U-Boot SPL 2025.04-01380-ga9820e12db18-dirty (Apr 25 2025 - 09:40:45 -0300) No pmic

WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
SEC0:  RNG instantiated

Trying to boot from MMC1
****** sdhc clk is 204522251

If I revert b4734c9c333b ("clk: imx: Convert clock-osc-* back to osc_*"), the clock is calculated correctly:

U-Boot 2025.04-01380-ga9820e12db18-dirty (Apr 25 2025 - 09:40:45 -0300)

CPU:   NXP i.MX8MMQ Rev1.0 A53 at 1200 MHz
CPU:   Consumer temperature grade (0C to 95C) at 42C
Model: FSL i.MX8MM EVK board
DRAM:  2 GiB
Core:  188 devices, 24 uclasses, devicetree: separate
WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
MMC:   fsl_esdhc mmc@30b50000: no vqmmc-supply

****** sdhc clk is 200000000

If I do:

dts/upstream/src/arm64/freescale/imx8mm-evk.dtsi +0 -2
@@ -614,8 +614,6 @@
 };
 
 &usdhc2 {
-       assigned-clocks = <&clk IMX8MM_CLK_USDHC2>;
-       assigned-clock-rates = <200000000>;
        pinctrl-names = "default", "state_100mhz", "state_200mhz";
        pinctrl-0 = <&pinctrl_usdhc2>, <&pinctrl_usdhc2_gpio>;
        pinctrl-1 = <&pinctrl_usdhc2_100mhz>, <&pinctrl_usdhc2_gpio>;

Then the pause is gone, but the printed clock is incorrect:

U-Boot SPL 2025.04-01380-ga9820e12db18-dirty (Apr 25 2025 - 09:52:28 -0300) No pmic

WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
SEC0:  RNG instantiated

Trying to boot from MMC1
****** sdhc clk is 1431655607

Adam, imx8mm-beacon does not have the
assigned-clocks/assigned-clock-rates properties.

Maybe this is why you haven't observed the problem?

author reviewer
#16 lore raw
Quoted 9 lines · Fabio Estevam

With the top-of-tree U-Boot, SPL calculates the wrong sdhc_clk on the imx8mm-evk:

U-Boot SPL 2025.04-01380-ga9820e12db18-dirty (Apr 25 2025 - 09:40:45 -0300) No pmic

WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
SEC0:  RNG instantiated

Trying to boot from MMC1
****** sdhc clk is 204522251

I have just sent a series that fixes this problem.

It contains some clock patches from Marek that I missed applying earlier, as well as an IMX8MM SPL OSC24 M clock fix.

Marek, care to submit your clk.diff patch formally so that the i.MX8MP USB functionality could be restored?

Thanks

#17 lore raw
Quoted 12 lines · Fabio Estevam
Quoted 9 lines · Fabio Estevam

With the top-of-tree U-Boot, SPL calculates the wrong sdhc_clk on the imx8mm-evk:

U-Boot SPL 2025.04-01380-ga9820e12db18-dirty (Apr 25 2025 - 09:40:45 -0300) No pmic

WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
SEC0:  RNG instantiated

Trying to boot from MMC1
****** sdhc clk is 204522251

Quoted 7 lines

I have just sent a series that fixes this problem.

It contains some clock patches from Marek that I missed applying earlier, as well as an IMX8MM SPL OSC24 M clock fix.

Marek, care to submit your clk.diff patch formally so that the i.MX8MP USB functionality could be restored?

If it is OK with you, send it as part of the fixup series, else I will send it sometimes next week.

#18 lore raw
In reply to Fabio Estevam
Quoted 19 lines · Fabio Estevam
Quoted 1 line · Fabio Estevam

I haven't had a chance to debug this issue yet.

I applied this debug patch:

drivers/mmc/fsl_esdhc_imx.c +1 -0
@@ -1539,6 +1539,7 @@ static int fsl_esdhc_probe(struct udevice *dev)
         }
 
         priv->sdhc_clk = clk_get_rate(&priv->per_clk);
+       printf("****** sdhc clk is %d\n", priv->sdhc_clk);
  #else
         init_clk_usdhc(dev_seq(dev));
 

With the top-of-tree U-Boot, SPL calculates the wrong sdhc_clk on the imx8mm-evk:

Quoted 6 lines

U-Boot SPL 2025.04-01380-ga9820e12db18-dirty (Apr 25 2025 - 09:40:45 -0300) No pmic

WDT:   Started watchdog@30280000 with servicing every 1000ms (60s timeout)
SEC0:  RNG instantiated

Trying to boot from MMC1
****** sdhc clk is 204522251

Look at cmd/clk.c and import soc_clk_dump() into this driver, and then call it here, so you would get the current topology of the clock tree. What do you see there in the output ? Please include it here.

#19 lore raw
In reply to Fabio Estevam
Quoted 27 lines · Fabio Estevam

Hi Marek,

Quoted 1 line · Marek Vasut

The "clock-output-names" is NEVER used for any clock look up.

In Linux, "clock-output-names" is used to register the name of the fixed-rate clock.

From Linux drivers/clk/clk-fixed-rate.c:

```
static struct clk_hw *_of_fixed_clk_setup(struct device_node *node)
{
....

of_property_read_string(node, "clock-output-names", &clk_name);

hw = clk_hw_register_fixed_rate_with_accuracy(NULL, clk_name, NULL,
     0, rate, accuracy);
```

In U-Boot, we should register a fixed-rate clock with the same name as in Linux.

This patch aims to make the U-Boot fixed-rate clock name conform to Linux standards.

Quoted 7 lines
Quoted 3 lines

I am fine with the first half of the description, but the second half is wrong and this actually does not fix anything. Simply remove clock-output-names from the osc_24m and the code will be broken again.

In Linux, if "clock-output-names" is removed from the osc_24m node, the console becomes corrupted
on an imx8mp-evk.

This is likely unrelated bug, see reply from Adam.

Quoted 10 lines
Quoted 6 lines

The clk_resolve_parent_clk() is meant to perform the aforementioned resolution(), that is what "[PATCH v2 00/24] clk: Add clk_resolve_parent_clk() and fix up iMX clock drivers" actually fixed. It seems there is something still missing and clk_resolve_parent_clk() does not do the resolution properly for some clock, but that is what needs to be understood and fixed, not worked around this way.

Please try to reproduce the problem on your end.

On an imx8mp-based board, run 'usb start' in U-Boot, and this command will fail.

I'll see what I can do tomorrow-ish .