RE: [PATCH v4 6/8] scsi: ufs: add LU Dedicated buffer mode support for WriteBooster

[Stanley Chu]

Hi Avri,

On Sun, 2020-05-03 at 08:00 +0000, Avri Altman wrote:
> 
> > -       if (!(hba->dev_info.b_wb_buffer_type &&
> > -             hba->dev_info.d_wb_alloc_units))
> > -               goto wb_disabled;
> > +       if (hba->dev_info.b_wb_buffer_type == WB_BUF_MODE_SHARED) {
> > +               hba->dev_info.d_wb_alloc_units =
> > +               get_unaligned_be32(desc_buf +
> > +                                  DEVICE_DESC_PARAM_WB_SHARED_ALLOC_UNITS);
> > +               if (!hba->dev_info.d_wb_alloc_units)
> > +                       goto wb_disabled;
> > +       } else {
> > +               for (lun = 0; lun < hba->dev_info.max_lu_supported; lun++) {
> max_lu_supported is determined according to bMaxNumberLU in the geometry descriptor,
> which can be 32. WB buffer however, is only valid only for LU 0, ..., LU7.
> Better to add this new limit to ufs.h.

Thanks for noticing this. I will fix it in next version.

> 
> > +                       ret = ufshcd_read_unit_desc_param(hba,
> > +                                       lun,
> > +                                       UNIT_DESC_PARAM_WB_BUF_ALLOC_UNITS,
> > +                                       (u8 *)&d_lu_wb_buf_alloc,
> > +                                       sizeof(d_lu_wb_buf_alloc));
> > +                       if (ret)
> > +                               goto wb_disabled;
> I think you should just continue here, as it is ok for the query to fail.
> The spec says:
> The WriteBooster Buffer is available only for the logical units from 0 to 7 which are configured as
> "normal memory type" (bMemoryType = 00h) and "not Boot well known logical unit" (bBootLunID =
> 00h), otherwise the Query Request shall fail and the Query Response field shall be set to "General
> Failure".
> 
> Sorry for not noticing this earlier.

Always appreciate your review and always not being too late : )

The spec does not mention clearly that the Query Request shall fail for
Read or Write. Although I think it shall fail for Write only, i.e., fail
during configuration only, it is better to let fail being skipped and
check d_lu_wb_buf_alloc only anyway.

I will fix this as well in next version.

Thanks,
Stanley Chu

> 
> Thanks,
> Avri
> > +                       if (d_lu_wb_buf_alloc) {
> > +                               hba->dev_info.wb_dedicated_lu = lun;
> > +                               break;
> > +                       }
> > +               }
> > 
> > +               if (!d_lu_wb_buf_alloc)
> > +                       goto wb_disabled;
> > +       }
> >         return;
> > 
> >  wb_disabled:
> > diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
> > index 898883058e3a..f232a67fd9b3 100644
> > --- a/drivers/scsi/ufs/ufshcd.h
> > +++ b/drivers/scsi/ufs/ufshcd.h
> > @@ -861,6 +861,13 @@ static inline bool
> > ufshcd_keep_autobkops_enabled_except_suspend(
> >         return hba->caps &
> > UFSHCD_CAP_KEEP_AUTO_BKOPS_ENABLED_EXCEPT_SUSPEND;
> >  }
> > 
> > +static inline u8 ufshcd_wb_get_flag_index(struct ufs_hba *hba)
> > +{
> > +       if (hba->dev_info.b_wb_buffer_type ==
> > WB_BUF_MODE_LU_DEDICATED)
> > +               return hba->dev_info.wb_dedicated_lu;
> > +       return 0;
> > +}
> > +
> >  extern int ufshcd_runtime_suspend(struct ufs_hba *hba);
> >  extern int ufshcd_runtime_resume(struct ufs_hba *hba);
> >  extern int ufshcd_runtime_idle(struct ufs_hba *hba);
> > --
> > 2.18.0
> 
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@xxxxxxxxxxxxxxxxxxx
> http://lists.infradead.org/mailman/listinfo/linux-mediatek