Re: [PATCH] spi: dw: Add 32 bpw support to DW DMA Controller

[Joy Chakraborty <joychakr@xxxxxxxxxx>]

Hi Serge(y),

On Mon, Mar 20, 2023 at 7:34 PM Serge Semin <fancer.lancer@xxxxxxxxx> wrote:
>
> Hello Joy.
>
> On Mon, Mar 20, 2023 at 05:57:46AM +0000, Joy Chakraborty wrote:
> > If DW Controller is capable of 32 bits per word support then SW or DMA
> > controller has to write 32bit or 4byte data to the FIFO at a time.
> >
> > This Patch adds support for AxSize = 4 bytes configuration from dw dma
> > driver if n_bytes i.e. number of bytes per write to fifo is 4.
> >
> > Signed-off-by: Joy Chakraborty <joychakr@xxxxxxxxxx>
>
> > ---
> >  drivers/spi/spi-dw-dma.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/spi/spi-dw-dma.c b/drivers/spi/spi-dw-dma.c
> > index ababb910b391..7d06ecfdebe1 100644
> > --- a/drivers/spi/spi-dw-dma.c
> > +++ b/drivers/spi/spi-dw-dma.c
> > @@ -212,6 +212,8 @@ static enum dma_slave_buswidth dw_spi_dma_convert_width(u8 n_bytes)
> >               return DMA_SLAVE_BUSWIDTH_1_BYTE;
> >       else if (n_bytes == 2)
> >               return DMA_SLAVE_BUSWIDTH_2_BYTES;
>
> > +     else if (n_bytes == 4)
> > +             return DMA_SLAVE_BUSWIDTH_4_BYTES;
>
> In case of the DFS-width being of 32-bits size n_bytes can be 4 and
> theoretically _3_ (practically it's unluckily, but anyway). Here
> it is:
> ...
> if (dws->caps & DW_SPI_CAP_DFS32)
>         master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
> ...
> dws->n_bytes = DIV_ROUND_UP(transfer->bits_per_word, BITS_PER_BYTE);
> ...
>
> So what about converting the dw_spi_dma_convert_width() method to
> having the switch-case statement and adding the adjacent "case 3:
> case 4:" statement there?
>
> * We could add the individual case-3 branch with DMA_SLAVE_BUSWIDTH_3_BYTES
> * returned, but the DMA-engines with 3-bytes bus width capability are
> * so rare. So is the case of having n_bytes == 3. Thus I guess it
> * won't hurt to extend the bus up to four bytes even though there are
> * only three bytes required.
>
> Please also note. Currently the spi-dw-dma.o driver doesn't make sure
> that the requested buswidth is actually supported by the DMA-engine
> (see dma_slave_caps.{src,dst}_addr_widths fields semantics). It would
> be nice to have some sanity check in there, but until then note DMA
> may still fail even if you specify a correct buswidth.
>
> -Serge(y)

Will be creating a V2 Patch with the suggested changes.
For the case of n_bytes = 3 would be using DMA_SLAVE_BUSWIDTH_4_BYTES
since that is what
the FIFO driver is doing as well while writing to the FIFO register in
the core driver:
...
        if (dws->n_bytes == 1)
                txw = *(u8 *)(dws->tx);
        else if (dws->n_bytes == 2)
                txw = *(u16 *)(dws->tx);
        else
                txw = *(u32 *)(dws->tx);
...

For the case of sanity checking of dma capability, I am planning to
add it to can_dma function so that it
defaults to a non-dma mode if the DMA controller is not capable of
satisfying the bits/word requirement.

Thanks
Joy
>
> >
> >       return DMA_SLAVE_BUSWIDTH_UNDEFINED;
> >  }
> > --
> > 2.40.0.rc1.284.g88254d51c5-goog
> >


On Mon, Mar 20, 2023 at 7:34 PM Serge Semin <fancer.lancer@xxxxxxxxx> wrote:
>
> Hello Joy.
>
> On Mon, Mar 20, 2023 at 05:57:46AM +0000, Joy Chakraborty wrote:
> > If DW Controller is capable of 32 bits per word support then SW or DMA
> > controller has to write 32bit or 4byte data to the FIFO at a time.
> >
> > This Patch adds support for AxSize = 4 bytes configuration from dw dma
> > driver if n_bytes i.e. number of bytes per write to fifo is 4.
> >
> > Signed-off-by: Joy Chakraborty <joychakr@xxxxxxxxxx>
>
> > ---
> >  drivers/spi/spi-dw-dma.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/spi/spi-dw-dma.c b/drivers/spi/spi-dw-dma.c
> > index ababb910b391..7d06ecfdebe1 100644
> > --- a/drivers/spi/spi-dw-dma.c
> > +++ b/drivers/spi/spi-dw-dma.c
> > @@ -212,6 +212,8 @@ static enum dma_slave_buswidth dw_spi_dma_convert_width(u8 n_bytes)
> >               return DMA_SLAVE_BUSWIDTH_1_BYTE;
> >       else if (n_bytes == 2)
> >               return DMA_SLAVE_BUSWIDTH_2_BYTES;
>
> > +     else if (n_bytes == 4)
> > +             return DMA_SLAVE_BUSWIDTH_4_BYTES;
>
> In case of the DFS-width being of 32-bits size n_bytes can be 4 and
> theoretically _3_ (practically it's unluckily, but anyway). Here
> it is:
> ...
> if (dws->caps & DW_SPI_CAP_DFS32)
>         master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32);
> ...
> dws->n_bytes = DIV_ROUND_UP(transfer->bits_per_word, BITS_PER_BYTE);
> ...
>
> So what about converting the dw_spi_dma_convert_width() method to
> having the switch-case statement and adding the adjacent "case 3:
> case 4:" statement there?
>
> * We could add the individual case-3 branch with DMA_SLAVE_BUSWIDTH_3_BYTES
> * returned, but the DMA-engines with 3-bytes bus width capability are
> * so rare. So is the case of having n_bytes == 3. Thus I guess it
> * won't hurt to extend the bus up to four bytes even though there are
> * only three bytes required.
>
> Please also note. Currently the spi-dw-dma.o driver doesn't make sure
> that the requested buswidth is actually supported by the DMA-engine
> (see dma_slave_caps.{src,dst}_addr_widths fields semantics). It would
> be nice to have some sanity check in there, but until then note DMA
> may still fail even if you specify a correct buswidth.
>
> -Serge(y)
>
> >
> >       return DMA_SLAVE_BUSWIDTH_UNDEFINED;
> >  }
> > --
> > 2.40.0.rc1.284.g88254d51c5-goog
> >