Re: [PATCH 2/5] soundwire: qcom: add support for mmio soundwire devices

[Jonathan Marek]

On 6/9/20 5:18 AM, Srinivas Kandagatla wrote:
> On 09/06/2020 05:34, Vinod Koul wrote:
> > Hi Jonathan,
> >
> > On 08-06-20, 16:43, Jonathan Marek wrote:
> > > Adds support for qcom soundwire devices with memory mapped IO registers.
> >
> > Please use 'SoundWire Master devices' instead :)
> >
> > > Signed-off-by: Jonathan Marek <jonathan@xxxxxxxx>
> > > ---
> > > Â drivers/soundwire/qcom.c | 25 +++++++++++++++++++++++--
> > > Â 1 file changed, 23 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/soundwire/qcom.c b/drivers/soundwire/qcom.c
> > > index f38d1fd3679f..628747df1c75 100644
> > > --- a/drivers/soundwire/qcom.c
> > > +++ b/drivers/soundwire/qcom.c
> > > @@ -90,6 +90,7 @@ struct qcom_swrm_ctrl {
> > > ÂÂÂÂÂ struct sdw_bus bus;
> > > ÂÂÂÂÂ struct device *dev;
> > > ÂÂÂÂÂ struct regmap *regmap;
> > > +ÂÂÂ void __iomem *mmio;
> > > ÂÂÂÂÂ struct completion *comp;
> > > ÂÂÂÂÂ struct work_struct slave_work;
> > > ÂÂÂÂÂ /* read/write lock */
> > > @@ -154,6 +155,20 @@ static int qcom_swrm_ahb_reg_write(struct 
> > > qcom_swrm_ctrl *ctrl,
> > > ÂÂÂÂÂ return SDW_CMD_OK;
> > > Â }
> > > +static int qcom_swrm_cpu_reg_read(struct qcom_swrm_ctrl *ctrl, int reg,
> > > +ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ u32 *val)
> > > +{
> > > +ÂÂÂ *val = readl(ctrl->mmio + reg);
> > > +ÂÂÂ return SDW_CMD_OK;
> > > +}
> > > +
> > > +static int qcom_swrm_cpu_reg_write(struct qcom_swrm_ctrl *ctrl, int 
> > > reg,
> > > +ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ int val)
> > > +{
> > > +ÂÂÂ writel(val, ctrl->mmio + reg);
> > > +ÂÂÂ return SDW_CMD_OK;
> > > +}
> >
> > this looks fine but regmap supports mmio also, so I am thinking we
> > should remove these and set regmap (mmio/slim)... Srini..?
>
> That is doable, but not going to add great value in this case, unless we 
> are having another layer of abstraction. So keeping it as readl/writel 
> seems okay to me.

Adding to this, the slim case doesn't use the regmap directly, it goes 
through AHB_BRIDGE registers. So using a single regmap path is not 
"doable" (at least not in a nice way IMO).

> --srini
>
>
> >