Re: [PATCH v1 1/2] scsi: ufs: core: Remove UFS_DEVICE_QUIRK_DELAY_AFTER_LPM quirk

[Bao D. Nguyen]

On 10/2/2025 8:10 PM, Peter Wang (王信友) wrote:
> On Thu, 2025-10-02 at 11:48 -0700, Bao D. Nguyen wrote:
> > Hi Peter,
> >
> > The current Mediatek platform driver applies this quirk to all ufs
> > vendors which is consistent with what we would like to do in the
> > Qualcomm platform driver per the vendor's requests.
> >
> > I do see that, about 5 years ago, Mediatek merged a patch to keep the
> > device vcc always on, probably to workaround some HW issues. Since
>
> Hi Bao,
>
> Yes, some UFS devices may have issues when turning off VCC.
>
> > this
> > is a very old patch and the impact of this change on a broken
> > hardware
> > is minimal, I would like weight the benefit of cleaning up the ufs
> > core
> > driver by removing the unnecessary quirk
> > UFS_DEVICE_QUIRK_DELAY_AFTER_LPM vs the inconvenience of a 5ms
> > (potentially reduce to 2ms) delay impact it may cause on an old
> > broken
> > HW in the suspend/shutdown path.
> >
> > I believe removing the UFS_DEVICE_QUIRK_DELAY_AFTER_LPM quirk in the
> > ufs
> > core driver as well as all the platform drivers yields positive net
> > benefits in this case.
> >
> > Thanks, Bao
>
> I think you misunderstood my point.
> I am okay with removing this flag, but this patch will cause
> devices with VCC always on to unnecessarily wait for the
> delay, resulting in wasted time.

Are you referring to the always_on flag in the struct ufs_vreg? I 
believe currently the ufs_vreg's always_on flag isn't used in 
determining whether the delay is applied or not.

How about we add the check for the Vcc's always_on as shown below?
The Mediatek's workaround can avoid the extra delay by setting the 
always_on flag as it already did, without using the 
UFS_DEVICE_QUIRK_DELAY_AFTER_LPM.

if (vcc_off && hba->vreg_info.vcc && !hba->vreg_info.vcc->always_on)
	usleep_range(5000, 5100);

Thanks, Bao

> Thanks
> Peter