Re: [PATCH RFC v4 1/1] scsi: ufs: Fix ufs power down/on specs violation

From: Bjorn Andersson
Date: Mon Jan 04 2021 - 13:59:47 EST


On Mon 28 Dec 19:48 CST 2020, Can Guo wrote:

> On 2020-12-29 09:18, Can Guo wrote:
> > On 2020-12-29 01:55, Bjorn Andersson wrote:
> > > On Tue 22 Dec 07:49 CST 2020, Ziqi Chen wrote:
> > >
> > > > As per specs, e.g, JESD220E chapter 7.2, while powering
> > > > off/on the ufs device, RST_N signal and REF_CLK signal
> > > > should be between VSS(Ground) and VCCQ/VCCQ2.
> > > >
> > > > To flexibly control device reset line, refactor the function
> > > > ufschd_vops_device_reset(sturct ufs_hba *hba) to ufshcd_
> > > > vops_device_reset(sturct ufs_hba *hba, bool asserted). The
> > > > new parameter "bool asserted" is used to separate device reset
> > > > line pulling down from pulling up.
> > > >
> > > > Cc: Kiwoong Kim <kwmad.kim@xxxxxxxxxxx>
> > > > Cc: Stanley Chu <stanley.chu@xxxxxxxxxxxx>
> > > > Signed-off-by: Ziqi Chen <ziqichen@xxxxxxxxxxxxxx>
> > > > ---
> > > > drivers/scsi/ufs/ufs-mediatek.c | 32
> > > > ++++++++++++++++----------------
> > > > drivers/scsi/ufs/ufs-qcom.c | 24 +++++++++++++++---------
> > > > drivers/scsi/ufs/ufshcd.c | 36
> > > > +++++++++++++++++++++++++-----------
> > > > drivers/scsi/ufs/ufshcd.h | 8 ++++----
> > > > 4 files changed, 60 insertions(+), 40 deletions(-)
> > > >
> > > > diff --git a/drivers/scsi/ufs/ufs-mediatek.c
> > > > b/drivers/scsi/ufs/ufs-mediatek.c
> > > > index 80618af..072f4db 100644
> > > > --- a/drivers/scsi/ufs/ufs-mediatek.c
> > > > +++ b/drivers/scsi/ufs/ufs-mediatek.c
> > > > @@ -841,27 +841,27 @@ static int
> > > > ufs_mtk_link_startup_notify(struct ufs_hba *hba,
> > > > return ret;
> > > > }
> > > >
> > > > -static int ufs_mtk_device_reset(struct ufs_hba *hba)
> > > > +static int ufs_mtk_device_reset(struct ufs_hba *hba, bool asserted)
> > > > {
> > > > struct arm_smccc_res res;
> > > >
> > > > - ufs_mtk_device_reset_ctrl(0, res);
> > > > + if (asserted) {
> > > > + ufs_mtk_device_reset_ctrl(0, res);
> > > >
> > > > - /*
> > > > - * The reset signal is active low. UFS devices shall detect
> > > > - * more than or equal to 1us of positive or negative RST_n
> > > > - * pulse width.
> > > > - *
> > > > - * To be on safe side, keep the reset low for at least 10us.
> > > > - */
> > > > - usleep_range(10, 15);
> > > > -
> > > > - ufs_mtk_device_reset_ctrl(1, res);
> > > > -
> > > > - /* Some devices may need time to respond to rst_n */
> > > > - usleep_range(10000, 15000);
> > > > + /*
> > > > + * The reset signal is active low. UFS devices shall detect
> > > > + * more than or equal to 1us of positive or negative RST_n
> > > > + * pulse width.
> > > > + *
> > > > + * To be on safe side, keep the reset low for at least 10us.
> > > > + */
> > > > + usleep_range(10, 15);
> > >
> > > I see no point in allowing vendors to "tweak" the 1us->10us
> > > adjustment.
> > > The specification says 1us and we all agree that 10us gives us good
> > > enough slack. I.e. this is common code.
> >
> > Hi Bjron,
> >
> > We tried, but Samsung fellows wanted 5us. We couldn't get a agreement
> > on this delay in short term, so we chose to leave it in vops.
> >
> > >
> > > > + } else {
> > > > + ufs_mtk_device_reset_ctrl(1, res);
> > > >
> > > > - dev_info(hba->dev, "device reset done\n");
> > > > + /* Some devices may need time to respond to rst_n */
> > > > + usleep_range(10000, 15000);
> > >
> > > The comment in both the Qualcomm and Mediatek drivers claim that
> > > this is
> > > sleep relates to the UFS device (not host), so why should it be
> > > different?
> > >
> > > What happens if I take the device that Mediatek see a need for a 10ms
> > > delay and hook that up to a Qualcomm host? This really should go in
> > > the
> > > common code.
> > >
> >
> > Agree, but Qualcomm host didn't have any problems with 10us yet, so if
> > we put
> > the 10ms delay to common code, Qualcomm host would suffer longer delay
> > when
> > device reset happens - both bootup and resume(xpm_lvl = 5/6) latency
> > would
> > be increased.
> >
> > Regards,
> > Can Guo.
> >
>
> Besides, currently this device reset vops is only registered by ufs-qcom.c
> and ufs-mediatek.c, meaning any delays that we put in the common code are
> not
> necessary for those who do not have this vops registered, i.e ufs-exynos.c,
> ufs-hisi.c.
>

Surely we can detect this in the common code and only sleep if the vops
is implemented - and successfully deasserted the reset.

Regards,
Bjorn