u-boot.git/drivers/clk/clk_scmi.c, branch master

clk: scmi: Fix protocol version fetch for non-CCF platforms

2026-04-09T02:14:19+00:00

The SCMI clock protocol version was only being fetched when CLK_CCF
was enabled. On non-CCF platforms, the probe function returned early
without fetching the version, leaving priv->version as 0.

This caused issues because code paths like scmi_clk_gate() and
scmi_clk_get_permissions() depend on priv->version to determine
which protocol message format to use, even in non-CCF mode.

Fix this by moving the scmi_generic_protocol_version() call before
the CLK_CCF check, ensuring the version is fetched for both CCF and
non-CCF platforms.

Tested on am62lx_evm.

Fixes: ae7e0330ce22 ("clk: scmi: add compatibility with clock protocol 2.0")
Signed-off-by: Kamlesh Gurudasani 
Signed-off-by: Peng Fan

clk: scmi: Remove duplicated scmi_generic_protocol_version() request

2026-01-05T02:04:48+00:00

scmi_generic_protocol_version() request is done twice in scmi_clk_probe().
Remove first call which is useless.

Fixes: ae7e0330ce22 ("clk: scmi: add compatibility with clock protocol 2.0")
Reviewed-by: Peng Fan 
Reviewed-by: Marek Vasut 
Signed-off-by: Patrice Chotard 
Signed-off-by: Peng Fan

clk: scmi: Fix priv initialization in scmi_clk_gate()

2026-01-05T02:04:48+00:00

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")
Reviewed-by: Peng Fan 
Reviewed-by: Marek Vasut 
Signed-off-by: Patrice Chotard 
Signed-off-by: Peng Fan

clk: scmi: Fix typo scmi_clk_get_attibute

2026-01-05T02:04:48+00:00

Fix typo attibute, rename scmi_clk_get_attibute() to
scmi_clk_get_attribute().

Reviewed-by: Peng Fan 
Reviewed-by: Marek Vasut 
Signed-off-by: Patrice Chotard 
Signed-off-by: Peng Fan

clk: scmi: Defer issue of SCMI_CLOCK_ATTRIBUTES

2025-11-10T12:57:49+00:00

Instead of resolving clock control flags using SCMI_CLOCK_ATTRIBUTES
during probe for each and every clock, resolve the clock control
flags using SCMI_CLOCK_ATTRIBUTES when the clock control flags are
first used. Because most clock are never used by U-Boot, this allows
reducing the amount of SCMI_CLOCK_ATTRIBUTES considerably, and this
improve probe time of the scmi clock driver and U-Boot start up time.

On Renesas X5H, with 1700+ SCMI clock, the boot time improved by 1.7s .

Reviewed-by: Peng Fan 
Signed-off-by: Marek Vasut 
Signed-off-by: Peng Fan

clk: scmi: Postpone clock name resolution

2025-11-10T12:57:48+00:00

The clock names are retrived via SCMI_CLOCK_ATTRIBUTES, called for each
clock ID. This may take a lot of time to complete and is not strictly
necessary. Register each clock as "scmi-%zu" instead, and let the first
call of SCMI_CLOCK_ATTRIBUTES fill in the actual clock name.

This has a side effect, which can be considered both an upside and also
a downside. Unused clock are never renamed and retain their placeholder
"scmi-%zu" name, which avoids empty clock names for nameless SCMI clock,
and avoids the name resolution and improves boot time. But for those
SCMI clock which do have name, that name is not listed until the clock
are used.

This is a preparatory patch for deferred issue of SCMI_CLOCK_ATTRIBUTES.

Reviewed-by: Peng Fan 
Signed-off-by: Marek Vasut 
Signed-off-by: Peng Fan

clk: scmi: Factor out clock control flags resolution

2025-11-10T12:57:48+00:00

Pull clock control flags resolution into dedicated function and
call it from each site that does access clock control flags. No
functional change.

This is a preparatory patch for deferred issue of SCMI_CLOCK_ATTRIBUTES.

Reviewed-by: Peng Fan 
Signed-off-by: Marek Vasut 
Signed-off-by: Peng Fan

clk: scmi: Bulk allocate all sub-driver instance data

2025-11-10T12:57:46+00:00

Allocate all sub-driver instance data at once. The amount of data that
have to be allocated is known up front, so is the size of the data, so
there is no need to call malloc() in a loop, mallocate all data at once.

The upside is, less heap fragmentation and fewer malloc() calls overall,
and a faster boot time.

The downside is, if some of the clock fail to register, then the clock
driver cannot free parts of the bulk allocated sub-driver instance data.
Such a failure can only occur if clk_register() were to fail, and if that
happens, the system has more significant problems. Worse, if a core clock
driver fails to probe, the system has even bigger problem.

Reviewed-by: Peng Fan 
Signed-off-by: Marek Vasut 
Signed-off-by: Peng Fan

clk: scmi: fix set_parent support when CCF is not being used

2025-11-10T12:57:44+00:00

When not using Common clock framework(CCF), calls to
scmi_clk_set_parent returns -ENOTSUPP, which should not be the case.
Fix that.

Fixes: 15fdfef6642c ("clk: scmi: check the clock state/parent/rate
control permissions)

Signed-off-by: Kamlesh Gurudasani 
Reviewed-by: Dhruva Gole 
Signed-off-by: Peng Fan

firmware: scmi: Add clock v3.2 CONFIG_SET support

2025-11-10T12:57:42+00:00

SCMI v3.2 introduces a new clock CONFIG_SET message format that can
optionally carry also OEM specific configuration values beside the usual
clock enable/disable requests. Add support to use such new format when
talking to a v3.2 compliant SCMI platform.

Support existing enable/disable operations across different clock protocol
versions: this patch still does not add protocol operations to support the
new OEM specific optional configuration capabilities.

No functional change for the SCMI drivers users of the related enable and
disable clock operations.

[Marek: Remodel after Linux e49e314a2cf7 ("firmware: arm_scmi: Add clock v3.2 CONFIG_SET support")
        Support both old < 2.1 and new >= 2.1 protocol versions.
	Update commit message based on Linux one]

Signed-off-by: Vinh Nguyen 
Signed-off-by: Marek Vasut 
Reviewed-by: Alice Guo 
Signed-off-by: Peng Fan