RE: [PATCH v5 1/8] scsi: ufs: enable WriteBooster on some pre-3.1 UFS devices

[Stanley Chu <stanley.chu@mediatek.com>]

Hi Avri,

On Mon, 2020-05-04 at 10:37 +0000, Avri Altman wrote:
> > 
> >  static void ufshcd_wb_probe(struct ufs_hba *hba, u8 *desc_buf)
> >  {
> > +       if (!ufshcd_is_wb_allowed(hba))
> > +               return;
> > +
> > +       if (hba->desc_size.dev_desc <=
> > DEVICE_DESC_PARAM_EXT_UFS_FEATURE_SUP)
> Should be 
> DEVICE_DESC_PARAM_EXT_UFS_FEATURE_SUP + 4 

I think this description length check is redundant because the device
quirk shall be added only after WriteBooster supportat is confirmed in
attached UFS device. So I will remove this in next version.

> 
> > +               goto wb_disabled;
> > +
> >         hba->dev_info.d_ext_ufs_feature_sup =
> >                 get_unaligned_be32(desc_buf +
> >                                    DEVICE_DESC_PARAM_EXT_UFS_FEATURE_SUP);
> 
> 
> > 
> >  static int ufs_get_device_desc(struct ufs_hba *hba)
> > @@ -6862,10 +6890,6 @@ static int ufs_get_device_desc(struct ufs_hba
> > *hba)
> > 
> >         model_index = desc_buf[DEVICE_DESC_PARAM_PRDCT_NAME];
> > 
> > -       /* Enable WB only for UFS-3.1 */
> > -       if (dev_info->wspecversion >= 0x310)
> > -               ufshcd_wb_probe(hba, desc_buf);
> > -
> >         err = ufshcd_read_string_desc(hba, model_index,
> >                                       &dev_info->model, SD_ASCII_STD);
> >         if (err < 0) {
> > @@ -6874,6 +6898,16 @@ static int ufs_get_device_desc(struct ufs_hba
> > *hba)
> >                 goto out;
> >         }
> > 
> > +       ufs_fixup_device_setup(hba);
> I don't think you should "hide" ufs_fixup_device_setup inside ufs_get_device_desc.

The reason is as below,

ufshcd_wb_probe() needs the contents of Device Descriptor for
initialization. To avoid double reading the Device Descriptor, I keep
ufshcd_wb_probe() inside ufs_get_device_desc() to use its "desc_buf".

And ufshcd_wb_probe() needs well-configured device quirk for entrance
check, thus ufs_fixup_device_setup() shall be moved before
ufshcd_wb_probe().

This change does not affect the behavior and functionality of
ufs_fixup_device_setup() since it is still executed once only during
ufshcd_init() flow and not be executed again in the future.

Thanks,
Stanley Chu

> 
> Thanks,
> Avri
> 
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-mediatek