On Tue, Dec 09, 2025 at 12:54:15PM +0100, Marek Vasut wrote:
>On 12/8/25 2:51 PM, Patrice Chotard wrote:
>> In scmi_clk_probe(), in case of CLK_CCF is not enabled, parent private
>> data is not set, so in scmi_clk_gate(), an uninitialized priv struct is
>> retrieved.
>>
>> SCMI request is performed either using scmi_clk_state_in_v1 or
>> scmi_clk_state_in_v2 struct depending of the unpredictable value of
>> priv->version which leads to error during SCMI clock enable.
>>
>> Issue detected on STM32MP157C-DK2 board using the SCMI device tree
>> stm32mp157c-dk2-scmi.dts.
>>
>> Fixes: 0619cb32030b ("firmware: scmi: Add clock v3.2 CONFIG_SET support")
>>
>> Signed-off-by: Patrice Chotard <[email protected]>
>>
>> Cc: Alice Guo <[email protected]>
>> Cc: Marek Vasut <[email protected]>
>> Cc: Patrick Delaunay <[email protected]>
>> Cc: Peng Fan <[email protected]>
>> Cc: Sean Anderson <[email protected]>
>> Cc: Tom Rini <[email protected]>
>> Cc: Valentin Caron <[email protected]>
>> Cc: Vinh Nguyen <[email protected]>
>> Cc: [email protected]
>> ---
>> drivers/clk/clk_scmi.c | 7 ++++++-
>> 1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/clk/clk_scmi.c b/drivers/clk/clk_scmi.c
>> index f6132178205..e1c20f2c47c 100644
>> --- a/drivers/clk/clk_scmi.c
>> +++ b/drivers/clk/clk_scmi.c
>> @@ -137,7 +137,7 @@ static int scmi_clk_get_attribute(struct udevice *dev,
>> int clkid, char *name,
>> static int scmi_clk_gate(struct clk *clk, int enable)
>> {
>> - struct scmi_clock_priv *priv = dev_get_parent_priv(clk->dev);
>> + struct scmi_clock_priv *priv;
>> struct scmi_clk_state_in_v1 in_v1 = {
>> .clock_id = clk_get_id(clk),
>> .attributes = enable,
>> @@ -156,6 +156,11 @@ static int scmi_clk_gate(struct clk *clk, int enable)
>> in_v2, out);
>> int ret;
>> + if (!CONFIG_IS_ENABLED(CLK_CCF))
>> + priv = dev_get_priv(clk->dev);
>> + else
>> + priv = dev_get_parent_priv(clk->dev);
>> +
>Shouldn't this be doing similar resolution like existing code ?
>
>168 static int scmi_clk_get_ctrl_flags(struct clk *clk, u32 *ctrl_flags)
>169 {
>170 struct clk_scmi *clkscmi;
>171 struct udevice *dev;
>172 u32 attributes;
>173 struct clk *c;
>174 int ret;
>175
>176 ret = clk_get_by_id(clk->id, &c);
>177 if (ret)
>178 return ret;
>179
>180 dev = c->dev->parent; // <-----------------------------
No need. The current clk code is indeed confusing.
In drivers/clk/clk.c, clk_get_by_id is ready called, so the first param passed
to scmi_clk_gate is the clk that registered during clk driver probe.
Regards
Peng
>181
>182 clkscmi = container_of(c, struct clk_scmi, clk);
>183
>184 if (!clkscmi->attrs_resolved) {
>185 char name[SCMI_CLOCK_NAME_LENGTH_MAX];
>186 ret = scmi_clk_get_attibute(dev,
> ^^^^^^^^^^^^^^^^^^^^^^^^^
>...
> 90 static int scmi_clk_get_attibute(struct udevice *dev, int clkid, char
>*name,
> 91 u32 *attr)
> 92 {
> 93 struct scmi_clock_priv *priv = dev_get_priv(dev);
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
