RE: [RFC PATCH v3 2/2] ufs: change the way to complete fDeviceInit

["Kiwoong Kim" <kwmad.kim@samsung.com>]

> > Currently, UFS driver checks if fDeviceInit is cleared at several
> > times, not period. This patch is to wait its completion with the
> > period, not retrying.
> > Many device vendors usually provides the specification on it with just
> > period, not a combination of a number of retrying and period. So it
> > could be proper to regard to the information coming from device
> > vendors.
> >
> > I first added one device specific value regarding the information.
> >
> > Signed-off-by: Kiwoong Kim <kwmad.kim@samsung.com>
> I still think that this patch alone is fine, and you don't need its
> predecessor.
> The spec requires polling, so this is a form of a more-effective-polling:
> so be it.

If what you're mentioning 'effective' means being able to just wait for
some long time upon completion of being cleared, it's not proper in real
products because fDeviceInit latency usually has the biggest overhead of
steps run during boot and some companies often try to manage its latency
as KPI. The method like a combination of retrying and small delay make them
harder to make it.
Or if I understand what you meant, please let me know.

> 
> 
> > ---
> >  drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++------------
> >  1 file changed, 24 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
> > index b26f182..6c08ed2 100644
> > --- a/drivers/scsi/ufs/ufshcd.c
> > +++ b/drivers/scsi/ufs/ufshcd.c
> > @@ -208,6 +208,7 @@ static struct ufs_dev_fix ufs_fixups[] = {  };
> >
> >  static const struct ufs_dev_value ufs_dev_values[] = {
> > +       {UFS_VENDOR_TOSHIBA, UFS_ANY_MODEL, DEV_VAL_FDEVICEINIT, 2000,
> > false},
> >         {0, 0, 0, 0, false},
> >  };
> >
> > @@ -4162,9 +4163,12 @@ EXPORT_SYMBOL_GPL(ufshcd_config_pwr_mode);
> >   */
> >  static int ufshcd_complete_dev_init(struct ufs_hba *hba)  {
> > -       int i;
> > +       u32 dev_init_compl_in_ms = 1000;
> > +       unsigned long timeout;
> >         int err;
> >         bool flag_res = true;
> > +       bool is_dev_val;
> > +       u32 val;
> >
> >         err = ufshcd_query_flag_retry(hba, UPIU_QUERY_OPCODE_SET_FLAG,
> >                 QUERY_FLAG_IDN_FDEVICEINIT, 0, NULL); @@ -4175,20
> > +4179,28 @@ static int ufshcd_complete_dev_init(struct ufs_hba *hba)
> >                 goto out;
> >         }
> >
> > -       /* poll for max. 1000 iterations for fDeviceInit flag to clear */
> > -       for (i = 0; i < 1000 && !err && flag_res; i++)
> > -               err = ufshcd_query_flag_retry(hba,
> > UPIU_QUERY_OPCODE_READ_FLAG,
> > -                       QUERY_FLAG_IDN_FDEVICEINIT, 0, &flag_res);
> > +       /* Poll fDeviceInit flag to be cleared */
> > +       is_dev_val = ufs_get_dev_specific_value(hba,
> > + DEV_VAL_FDEVICEINIT,
> > &val);
> > +       dev_init_compl_in_ms = (is_dev_val) ? val : 500;
> If you want dev_init_compl_in_ms to take its default 1,000, you should:
> dev_init_compl_in_ms = (!is_dev_val) ? : val;
Got it.

> 
> > +       timeout = jiffies + msecs_to_jiffies(dev_init_compl_in_ms);
> > +       do {
> > +               err = ufshcd_query_flag(hba, UPIU_QUERY_OPCODE_READ_FLAG,
> > +                                       QUERY_FLAG_IDN_FDEVICEINIT, 0, &flag_res);
> > +               if (!flag_res)
> > +                       break;
> > +               usleep_range(5, 10);
> Per Grant's comment:
> usleep_range(5000, 10000); 
Got it.

Thanks.
Kiwoong Kim