Re: [PATCH v3 3/4] PCI: Add function for parsing 'slot-power-limit-milliwatt' DT property

[Bjorn Helgaas]

On Fri, Mar 25, 2022 at 10:38:26AM +0100, Pali Rohár wrote:
> Add function of_pci_get_slot_power_limit(), which parses the
> 'slot-power-limit-milliwatt' DT property, returning the value in
> milliwatts and in format ready for the PCIe Slot Capabilities Register.
> 
> Signed-off-by: Pali Rohár <pali@xxxxxxxxxx>
> Signed-off-by: Marek Behún <kabel@xxxxxxxxxx>
> Reviewed-by: Rob Herring <robh@xxxxxxxxxx>
> ---
> Changes in v3:
> * Set 600 W when DT slot-power-limit-milliwatt > 600 W
> Changes in v2:
> * Added support for PCIe 6.0 slot power limit encodings
> * Round down slot power limit value
> ---
>  drivers/pci/of.c  | 64 +++++++++++++++++++++++++++++++++++++++++++++++
>  drivers/pci/pci.h | 15 +++++++++++
>  2 files changed, 79 insertions(+)
> 
> diff --git a/drivers/pci/of.c b/drivers/pci/of.c
> index cb2e8351c2cc..5ebff26edd41 100644
> --- a/drivers/pci/of.c
> +++ b/drivers/pci/of.c
> @@ -633,3 +633,67 @@ int of_pci_get_max_link_speed(struct device_node *node)
>  	return max_link_speed;
>  }
>  EXPORT_SYMBOL_GPL(of_pci_get_max_link_speed);
> +
> +/**
> + * of_pci_get_slot_power_limit - Parses the "slot-power-limit-milliwatt"
> + *				 property.
> + *
> + * @node: device tree node with the slot power limit information
> + * @slot_power_limit_value: pointer where the value should be stored in PCIe
> + *			    Slot Capabilities Register format
> + * @slot_power_limit_scale: pointer where the scale should be stored in PCIe
> + *			    Slot Capabilities Register format
> + *
> + * Returns the slot power limit in milliwatts and if @slot_power_limit_value
> + * and @slot_power_limit_scale pointers are non-NULL, fills in the value and
> + * scale in format used by PCIe Slot Capabilities Register.
> + *
> + * If the property is not found or is invalid, returns 0.
> + */
> +u32 of_pci_get_slot_power_limit(struct device_node *node,
> +				u8 *slot_power_limit_value,
> +				u8 *slot_power_limit_scale)
> +{
> +	u32 slot_power_limit_mw;
> +	u8 value, scale;
> +
> +	if (of_property_read_u32(node, "slot-power-limit-milliwatt",
> +				 &slot_power_limit_mw))
> +		slot_power_limit_mw = 0;
> +
> +	/* Calculate Slot Power Limit Value and Slot Power Limit Scale */
> +	if (slot_power_limit_mw == 0) {
> +		value = 0x00;
> +		scale = 0;
> +	} else if (slot_power_limit_mw <= 255) {
> +		value = slot_power_limit_mw;
> +		scale = 3;
> +	} else if (slot_power_limit_mw <= 255*10) {
> +		value = slot_power_limit_mw / 10;
> +		scale = 2;
> +	} else if (slot_power_limit_mw <= 255*100) {
> +		value = slot_power_limit_mw / 100;
> +		scale = 1;
> +	} else if (slot_power_limit_mw <= 239*1000) {
> +		value = slot_power_limit_mw / 1000;
> +		scale = 0;
> +	} else if (slot_power_limit_mw <= 250*1000) {
> +		value = 0xF0;
> +		scale = 0;

I think the spec is poorly worded here.  PCIe r6.0, sec 7.5.3.9, says:

  F0h   > 239 W and <= 250 W Slot Power Limit

I don't think it's meaningful for the spec to include a range here.
The amount of power the slot can supply has a single maximum.  I
suspect the *intent* of F0h/00b is that a device in the slot may
consume up to 250W.

Your code above would mean that slot_power_limit_mw == 245,000 would
cause the slot to advertise F0h/00b (250W), which seems wrong.

I think we should do something like this instead:

  scale = 0;
  if (slot_power_limit_mw >= 600*1000) {
    value = 0xFE;
    slot_power_limit_mw = 600*1000;
  } else if (slot_power_limit_mw >= 575*1000) {
    value = 0xFD;
    slot_power_limit_mw = 575*1000;
  } ...

I raised an issue with the PCI SIG about this.

> +	} else if (slot_power_limit_mw <= 600*1000) {
> +		value = 0xF0 + (slot_power_limit_mw / 1000 - 250) / 25;
> +		scale = 0;
> +	} else {
> +		value = 0xFE;
> +		scale = 0;
> +	}
> +
> +	if (slot_power_limit_value)
> +		*slot_power_limit_value = value;
> +
> +	if (slot_power_limit_scale)
> +		*slot_power_limit_scale = scale;
> +
> +	return slot_power_limit_mw;

If the DT tells us 800W is available, we'll store (FEh/00b), which
means the slot can advertise to a downstream device that 600W is
available.  I think that's correct, since the current spec doesn't
provide a way to encode any value larger than 600W.

But the function still returns 800,000 mW, which means the next patch will
print:

  %s: Slot power limit 800.0W

even though it programs Slot Capabilities to advertise 600W.
That's why I suggested setting slot_power_limit_mw = 600*1000 above.

> +}
> +EXPORT_SYMBOL_GPL(of_pci_get_slot_power_limit);
> diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
> index 3d60cabde1a1..e10cdec6c56e 100644
> --- a/drivers/pci/pci.h
> +++ b/drivers/pci/pci.h
> @@ -627,6 +627,9 @@ struct device_node;
>  int of_pci_parse_bus_range(struct device_node *node, struct resource *res);
>  int of_get_pci_domain_nr(struct device_node *node);
>  int of_pci_get_max_link_speed(struct device_node *node);
> +u32 of_pci_get_slot_power_limit(struct device_node *node,
> +				u8 *slot_power_limit_value,
> +				u8 *slot_power_limit_scale);
>  void pci_set_of_node(struct pci_dev *dev);
>  void pci_release_of_node(struct pci_dev *dev);
>  void pci_set_bus_of_node(struct pci_bus *bus);
> @@ -653,6 +656,18 @@ of_pci_get_max_link_speed(struct device_node *node)
>  	return -EINVAL;
>  }
>  
> +static inline u32
> +of_pci_get_slot_power_limit(struct device_node *node,
> +			    u8 *slot_power_limit_value,
> +			    u8 *slot_power_limit_scale)
> +{
> +	if (slot_power_limit_value)
> +		*slot_power_limit_value = 0;
> +	if (slot_power_limit_scale)
> +		*slot_power_limit_scale = 0;
> +	return 0;
> +}
> +
>  static inline void pci_set_of_node(struct pci_dev *dev) { }
>  static inline void pci_release_of_node(struct pci_dev *dev) { }
>  static inline void pci_set_bus_of_node(struct pci_bus *bus) { }
> -- 
> 2.20.1
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel