If the clock-frequency property is not set in the cpu node but in the
parent /cpus node, the following warning is emitted:
/cpus/cpu@X missing clock-frequency property
The devicetree specification in section 3.8 however specifies that
"properties that have identical values across cpu nodes may be placed in
the /cpus node instead. A client program must first examine a specific cpu
node, but if an expected property is not found then it should look at the
parent /cpus node."
Signed-off-by: Stefan Wiehler <[email protected]>
---
arch/arm/kernel/topology.c | 21 ++++++++++++++++++---
1 file changed, 18 insertions(+), 3 deletions(-)
diff --git a/arch/arm/kernel/topology.c b/arch/arm/kernel/topology.c
index ef0058de432b..32fc1c8d9d11 100644
--- a/arch/arm/kernel/topology.c
+++ b/arch/arm/kernel/topology.c
@@ -85,15 +85,24 @@ static bool cap_from_dt = true;
static void __init parse_dt_topology(void)
{
const struct cpu_efficiency *cpu_eff;
- struct device_node *cn = NULL;
+ struct device_node *pcn = NULL, *cn = NULL;
unsigned long min_capacity = ULONG_MAX;
unsigned long max_capacity = 0;
unsigned long capacity = 0;
int cpu = 0;
+ const __be32 *common_rate;
+ int common_rate_len;
__cpu_capacity = kcalloc(nr_cpu_ids, sizeof(*__cpu_capacity),
GFP_NOWAIT);
+ pcn = of_find_node_by_path("/cpus");
+ if (!pcn) {
+ pr_err("missing CPU root device node\n");
+ return;
+ }
+ common_rate = of_get_property(pcn, "clock-frequency", &common_rate_len);
+
for_each_possible_cpu(cpu) {
const __be32 *rate;
int len;
@@ -121,8 +130,12 @@ static void __init parse_dt_topology(void)
rate = of_get_property(cn, "clock-frequency", &len);
if (!rate || len != 4) {
- pr_err("%pOF missing clock-frequency property\n", cn);
- continue;
+ if (common_rate && common_rate_len == 4) {
+ rate = common_rate;
+ } else {
+ pr_err("%pOF missing clock-frequency property\n", cn);
+ continue;
+ }
}
capacity = ((be32_to_cpup(rate)) >> 20) * cpu_eff->efficiency;
@@ -154,6 +167,8 @@ static void __init parse_dt_topology(void)
if (cap_from_dt)
topology_normalize_cpu_scale();
+
+ of_node_put(pcn);
}
/*
--
2.42.0
On 01/02/2024 13:36, Stefan Wiehler wrote:
> If the clock-frequency property is not set in the cpu node but in the
> parent /cpus node, the following warning is emitted:
>
> /cpus/cpu@X missing clock-frequency property
>
> The devicetree specification in section 3.8 however specifies that
> "properties that have identical values across cpu nodes may be placed in
> the /cpus node instead. A client program must first examine a specific cpu
> node, but if an expected property is not found then it should look at the
> parent /cpus node."
>
> Signed-off-by: Stefan Wiehler <[email protected]>
> ---
> arch/arm/kernel/topology.c | 21 ++++++++++++++++++---
> 1 file changed, 18 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm/kernel/topology.c b/arch/arm/kernel/topology.c
> index ef0058de432b..32fc1c8d9d11 100644
> --- a/arch/arm/kernel/topology.c
> +++ b/arch/arm/kernel/topology.c
> @@ -85,15 +85,24 @@ static bool cap_from_dt = true;
> static void __init parse_dt_topology(void)
> {
> const struct cpu_efficiency *cpu_eff;
> - struct device_node *cn = NULL;
> + struct device_node *pcn = NULL, *cn = NULL;
> unsigned long min_capacity = ULONG_MAX;
> unsigned long max_capacity = 0;
> unsigned long capacity = 0;
> int cpu = 0;
> + const __be32 *common_rate;
> + int common_rate_len;
>
> __cpu_capacity = kcalloc(nr_cpu_ids, sizeof(*__cpu_capacity),
> GFP_NOWAIT);
>
> + pcn = of_find_node_by_path("/cpus");
> + if (!pcn) {
> + pr_err("missing CPU root device node\n");
> + return;
> + }
> + common_rate = of_get_property(pcn, "clock-frequency", &common_rate_len);
Aren't you adding new property? Is it already documented in the
bindings? After a quick look I think this is not documented.
Best regards,
Krzysztof
> Does this mean the /cpus property is like a default for when a CPU node
> doesn't specify the clock frequency, or does it mean that the /cpus
> property should only exist when all the values for each CPU are
> identical and thus the individual CPU node clock frequency should
> not be specified.
Good question, the device tree specification in Section 3.7 [1] says:
> The /cpus node may contain properties that are common across cpu
nodes. See Section 3.8 for details.
And in Section 3.8 [2]:
> Properties that have identical values across cpu nodes may be placed
> in the /cpus node instead. A client program must first examine a
> specific cpu node, but if an expected property is not found then it
> should look at the parent /cpus node. This results in a less verbose
> representation of properties which are identical across all CPUs.
So I think it is pretty clear that it should only be used for
common/identical values.
> Aren't you adding new property? Is it already documented in the
> bindings? After a quick look I think this is not documented.
You are right, clock-frequency is not mentioned neither in arm/cpus.yaml
nor in any other <arch>/cpus.yaml binding, but the DT spec has it as a
required property [3]. Should I add clock-frequency to all
<arch>/cpus.yaml bindings? Only the ARM one explicitly mentions
following the DT spec.
Kind regards,
Stefan
[1]
https://devicetree-specification.readthedocs.io/en/latest/chapter3-devicenodes.html#cpus-node-properties
[2]
https://devicetree-specification.readthedocs.io/en/latest/chapter3-devicenodes.html#cpus-cpu-node-properties
[3]
https://devicetree-specification.readthedocs.io/en/latest/chapter3-devicenodes.html#general-properties-of-cpus-cpu-nodes
On Thu, Feb 01, 2024 at 04:03:59PM +0100, Stefan Wiehler wrote:
> > Does this mean the /cpus property is like a default for when a CPU node
> > doesn't specify the clock frequency, or does it mean that the /cpus
> > property should only exist when all the values for each CPU are
> > identical and thus the individual CPU node clock frequency should
> > not be specified.
>
> Good question, the device tree specification in Section 3.7 [1] says:
>
> > The /cpus node may contain properties that are common across cpu
> nodes. See Section 3.8 for details.
>
> And in Section 3.8 [2]:
>
> > Properties that have identical values across cpu nodes may be placed
> > in the /cpus node instead. A client program must first examine a
> > specific cpu node, but if an expected property is not found then it
> > should look at the parent /cpus node. This results in a less verbose
> > representation of properties which are identical across all CPUs.
>
> So I think it is pretty clear that it should only be used for
> common/identical values.
Thanks for the clarification.
As this is DT specified behaviour, I question whether it should be
implemented in arch/arm/kernel/topology.c - what I'm meaning is
a helper such as:
const void *of_get_cpu_property(const struct device_node *node,
const char *name, int *lenp)
{
const void *res;
res = of_get_property(node, name, lenp);
if (!res) {
node = of_find_node_by_path("/cpus");
if (node)
res = of_get_property(node, name, lenp);
of_node_put(node);
}
return res;
}
?
--
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!
On 01/02/2024 16:03, Stefan Wiehler wrote:
>> Does this mean the /cpus property is like a default for when a CPU node
>> doesn't specify the clock frequency, or does it mean that the /cpus
>> property should only exist when all the values for each CPU are
>> identical and thus the individual CPU node clock frequency should
>> not be specified.
>
> Good question, the device tree specification in Section 3.7 [1] says:
>
> > The /cpus node may contain properties that are common across cpu
> nodes. See Section 3.8 for details.
>
> And in Section 3.8 [2]:
>
> > Properties that have identical values across cpu nodes may be placed
> > in the /cpus node instead. A client program must first examine a
> > specific cpu node, but if an expected property is not found then it
> > should look at the parent /cpus node. This results in a less verbose
> > representation of properties which are identical across all CPUs.
>
> So I think it is pretty clear that it should only be used for
> common/identical values.
>
>> Aren't you adding new property? Is it already documented in the
>> bindings? After a quick look I think this is not documented.
>
> You are right, clock-frequency is not mentioned neither in arm/cpus.yaml
> nor in any other <arch>/cpus.yaml binding, but the DT spec has it as a
> required property [3]. Should I add clock-frequency to all
> <arch>/cpus.yaml bindings? Only the ARM one explicitly mentions
> following the DT spec.
It should go to dtschema. dtschema cpu.yaml has it, so you need to
propose such to cpus.yaml, probably you could experiment with:
not:
- required:
- clock-frequency
- patternProperties:
cpu@....
- required:
- clock-frequency
Anyway, you cannot just keep adding some OF properties to the code
without documenting them.
Best regards,
Krzysztof