Re: [RFC PATCH v1 3/4] PCI/ASPM: Remove struct pcie_link_state.clkpm_enabled

[Bjorn Helgaas]

On Wed, Sep 29, 2021 at 02:43:59AM +0200, Saheed O. Bolarinwa wrote:
> From: "Bolarinwa O. Saheed" <refactormyself@xxxxxxxxx>
> 
> The clkpm_enabled member of the struct pcie_link_state stores the
> current Clock PM state for the device. However, when the state changes
> it is persisted and can be retrieve by calling pcie_get_clkpm_state()
> introduced in patch [1/3] in this series.
> 
> This patch:
>    - removes clkpm_enabled from the struct pcie_link_state
>    - removes all instance where clkpm_enable is set
>    - replaces references to clkpm_enabled with a call to
>      pcie_get_clkpm_state()

Similar commit log comments as previous patch.

> Signed-off-by: Bolarinwa O. Saheed <refactormyself@xxxxxxxxx>
> ---
>  drivers/pci/pcie/aspm.c | 7 ++-----
>  1 file changed, 2 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> index 9e65da9a22dd..368828cd427d 100644
> --- a/drivers/pci/pcie/aspm.c
> +++ b/drivers/pci/pcie/aspm.c
> @@ -61,7 +61,6 @@ struct pcie_link_state {
>  	u32 aspm_disable:7;		/* Disabled ASPM state */
>  
>  	/* Clock PM state */
> -	u32 clkpm_enabled:1;		/* Current Clock PM state */
>  	u32 clkpm_disable:1;		/* Clock PM disabled */
>  
>  	/* Exit latencies */
> @@ -190,7 +189,6 @@ static void pcie_set_clkpm_nocheck(struct pcie_link_state *link, int enable)
>  		pcie_capability_clear_and_set_word(child, PCI_EXP_LNKCTL,
>  						   PCI_EXP_LNKCTL_CLKREQ_EN,
>  						   val);
> -	link->clkpm_enabled = !!enable;
>  }
>  
>  static void pcie_set_clkpm(struct pcie_link_state *link, int enable)
> @@ -203,14 +201,13 @@ static void pcie_set_clkpm(struct pcie_link_state *link, int enable)
>  	if (!capable || link->clkpm_disable)
>  		enable = 0;
>  	/* Need nothing if the specified equals to current state */
> -	if (link->clkpm_enabled == enable)
> +	if (pcie_get_clkpm_state(link->pdev) == enable)

Instead of pcie_get_clkpm_state(), I think I would add this:

  static int pcie_clkpm_enabled(struct pci_dev *pdev)
  {
    struct pci_dev *child;
    struct pci_bus *linkbus = pdev->subordinate;
    u16 ctl;

    /* CLKREQ_EN is only applicable for Upstream Ports */
    list_for_each_entry(child, &linkbus->devices, bus_list) {
      pcie_capability_read_word(PCI_EXP_LNKCTL, &ctl);
      if (!(ctl & PCI_EXP_LNKCTL_CLKREQ_EN))
        return 0;
    }
    return 1;
  }

And I would rename pcie_is_clkpm_capable() from the previous patch to
match, i.e., pcie_clkpm_capable().  I suggested "bool" for it, but maybe
it should stay "int" to match this.  They both could be "bool", but
that seems a little messy because "enable" comes from
policy_to_clkpm_state(), so it would involve quite a few more changes.

>  		return;
>  	pcie_set_clkpm_nocheck(link, enable);
>  }
>  
>  static void pcie_clkpm_cap_init(struct pcie_link_state *link, int blacklist)
>  {
> -	link->clkpm_enabled = pcie_get_clkpm_state(link->pdev);
>  	link->clkpm_disable = blacklist ? 1 : 0;
>  }
>  
> @@ -1287,7 +1284,7 @@ static ssize_t clkpm_show(struct device *dev,
>  	struct pci_dev *pdev = to_pci_dev(dev);
>  	struct pcie_link_state *link = pcie_aspm_get_link(pdev);
>  
> -	return sysfs_emit(buf, "%d\n", link->clkpm_enabled);
> +	return sysfs_emit(buf, "%d\n", pcie_get_clkpm_state(link->pdev));
>  }
>  
>  static ssize_t clkpm_store(struct device *dev,
> -- 
> 2.20.1
>