[PATCH V5 09/11] scsi: ufs: qcom : Refactor phy_power_on/off calls

[Manivannan Sadhasivam]

On Thu, May 22, 2025 at 03:10:48AM +0530, Nitin Rawat wrote:
> 
> 
> On 5/21/2025 7:27 PM, Manivannan Sadhasivam wrote:
> > On Thu, May 15, 2025 at 09:57:20PM +0530, Nitin Rawat wrote:
> > > Commit 3f6d1767b1a0 ("phy: ufs-qcom: Refactor all init steps into
> > > phy_poweron") removes the phy_power_on/off from ufs_qcom_setup_clocks
> > 
> > s/removes/moved
> 
> Sure, will address in next patchset.
> 
> > 
> > > to suspend/resume func.
> > > 
> > > To have a better power saving, remove the phy_power_on/off calls from
> > > resume/suspend path and put them back to ufs_qcom_setup_clocks, so that
> > > PHY regulators & clks can be turned on/off along with UFS's clocks.
> > > 
> > > Since phy phy_power_on is separated out from phy calibrate, make
> > > separate calls to phy_power_on and phy_calibrate calls from ufs qcom
> > > driver.
> > > 
> > 
> > This patch is not calling phy_calibrate().
> 
> updated commit text to remove phy_Calibrate from commit text
> 
> > 
> > > Co-developed-by: Can Guo <quic_cang at quicinc.com>
> > > Signed-off-by: Can Guo <quic_cang at quicinc.com>
> > > Signed-off-by: Nitin Rawat <quic_nitirawa at quicinc.com>
> > > ---
> > >   drivers/ufs/host/ufs-qcom.c | 61 ++++++++++++++++++-------------------
> > >   1 file changed, 29 insertions(+), 32 deletions(-)
> > > 
> > 
> > [...]
> > 
> > >   static int ufs_qcom_setup_clocks(struct ufs_hba *hba, bool on,
> > >   				 enum ufs_notify_change_status status)
> > >   {
> > >   	struct ufs_qcom_host *host = ufshcd_get_variant(hba);
> > > +	struct phy *phy = host->generic_phy;
> > > +	int err;
> > >   	/*
> > >   	 * In case ufs_qcom_init() is not yet done, simply ignore.
> > > @@ -1157,10 +1141,22 @@ static int ufs_qcom_setup_clocks(struct ufs_hba *hba, bool on,
> > >   				/* disable device ref_clk */
> > >   				ufs_qcom_dev_ref_clk_ctrl(host, false);
> > >   			}
> > > +
> > > +			err = phy_power_off(phy);
> > > +			if (err) {
> > > +				dev_err(hba->dev, "phy power off failed, ret=%d\n", err);
> > > +				return err;
> > > +			}
> > >   		}
> > >   		break;
> > >   	case POST_CHANGE:
> > >   		if (on) {
> > > +			err = phy_power_on(phy);
> > > +			if (err) {
> > > +				dev_err(hba->dev, "phy power on failed, ret = %d\n", err);
> > > +				return err;
> > > +			}
> > > +
> > >   			/* enable the device ref clock for HS mode*/
> > >   			if (ufshcd_is_hs_mode(&hba->pwr_info))
> > >   				ufs_qcom_dev_ref_clk_ctrl(host, true);
> > > @@ -1343,9 +1339,10 @@ static int ufs_qcom_init(struct ufs_hba *hba)
> > >   static void ufs_qcom_exit(struct ufs_hba *hba)
> > >   {
> > >   	struct ufs_qcom_host *host = ufshcd_get_variant(hba);
> > > +	struct phy *phy = host->generic_phy;
> > >   	ufs_qcom_disable_lane_clks(host);
> > > -	phy_power_off(host->generic_phy);
> > > +	phy_power_off(phy);
> > 
> > This change is not at all needed.
> 
> In the current code, PHY resources (clocks and rails) remain active even
> when the link is in a hibernate state and all controller clocks are off.
> This results in a significant power penalty and can prevent CX power
> collapse.
> 
> These rails and clocks are only turned off when a system suspend is
> triggered, and even then, only at SPM level 5 where the link is turned off.
> This approach is not power-efficient.
> 
> As part of this series, the code that enables/disables the PHY's regulators
> and clocks is now confined to the phy_power_on and phy_power_off calls, with
> the rest moved to the calibration phase.
> 
> Therefore, we are invoking the phy_power_off and phy_power_on calls from
> ufs_qcom_setup_clocks, ensuring that the PHY's regulators and clocks can be
> turned on/off along with the UFS clocks, thereby achieving power savings.
> 
> Therefore, this patch is the cornerstone of this series.
> 

I did not question the patch, but just the change that you did in the
ufs_qcom_exit() function. You added a local variable for 'host->generic_phy',
which is not related to this patch.

- Mani

-- 
மணிவண்ணன் சதாசிவம்