On 28-03-24, 08:21, Alexander Sverdlin wrote:
> Hello Vinod!
>
> On Thu, 2024-03-28 at 10:01 +0530, Vinod Koul wrote:
> > On 26-03-24, 12:18, Nikita Shubin via B4 Relay wrote:
> > > From: Nikita Shubin <nikita.shubin@xxxxxxxxxxx>
> > >
> > > Convert Cirrus EP93xx DMA to device tree usage:
> >
> > Subsytem is dmaengine: pls fix that
> >
> > >
> > > - add OF ID match table with data
> > > - add of_probe for device tree
> > > - add xlate for m2m/m2p
> > > - drop subsys_initcall code
> > > - drop platform probe
> > > - drop platform structs usage
> > >
> > > > From now on it only supports device tree probing.
> > >
> > > Co-developed-by: Alexander Sverdlin <alexander.sverdlin@xxxxxxxxx>
> > > Signed-off-by: Alexander Sverdlin <alexander.sverdlin@xxxxxxxxx>
> > > Signed-off-by: Nikita Shubin <nikita.shubin@xxxxxxxxxxx>
> > > ---
> > > drivers/dma/ep93xx_dma.c | 239 ++++++++++++++++++++++++-------
> > > include/linux/platform_data/dma-ep93xx.h | 6 +
> > > 2 files changed, 191 insertions(+), 54 deletions(-)
> > >
> > > diff --git a/drivers/dma/ep93xx_dma.c b/drivers/dma/ep93xx_dma.c
> > > index d6c60635e90d..17c8e2badee2 100644
> > > --- a/drivers/dma/ep93xx_dma.c
> > > +++ b/drivers/dma/ep93xx_dma.c
>
> [...]
>
> > >
> > > @@ -104,6 +106,11 @@
> > > #define DMA_MAX_CHAN_BYTES 0xffff
> > > #define DMA_MAX_CHAN_DESCRIPTORS 32
> > >
> > > +enum ep93xx_dma_type {
> > > + M2P_DMA,
> >
> > Is this missing P2M?
>
> Not really. It's not the most obvious one, but anyway a way to enumerate
> two types of DMA engines:
>
> "7.1.1 DMA Features List
> DMA specific features are:
> • Ten fully independent, programmable DMA controller internal M2P/P2M
> channels (5 Tx and 5 Rx).
> • Two dedicated channels for Memory-to-Memory (M2M) and
> Memory-to-External Peripheral Transfers (external M2P/P2M)."
>
> Now the confusing part is that this "M2M" engine is actually used
> to transfer to and from *some* devices, like SPI and IDE.
> So both engines are capable of M2P and P2M, maybe Cirrus has named
> two engines in a sub-optimal way decades ago and this is now a bit
> historical naming.
Okay, thanks for clarifying (might be worthwhile to mention this in
comments)
--
~Vinod
