Re: [PATCH 2/8 v2] Documentation: bindings: k3dma: Add binding for dma-avail-chan

From: Vinod Koul
Date: Mon Jan 07 2019 - 01:27:08 EST


On 05-01-19, 19:38, Manivannan Sadhasivam wrote:
> Hi Vinod,
>
> On Sat, Jan 05, 2019 at 07:16:10PM +0530, Vinod Koul wrote:
> > On 05-01-19, 10:23, Manivannan Sadhasivam wrote:
> > > On Fri, Jan 04, 2019 at 08:39:34PM -0800, John Stultz wrote:
> > > > On Fri, Jan 4, 2019 at 8:00 PM Manivannan Sadhasivam
> > > > <manivannan.sadhasivam@xxxxxxxxxx> wrote:
> > > > >
> > > > > Hi John,
> > > > >
> > > > > On Fri, Jan 04, 2019 at 12:56:22PM -0800, John Stultz wrote:
> > > > > > Some dma channels can be reserved for secure mode or other
> > > > > > hardware on the SoC, so provide a binding for a bitmask
> > > > > > listing the available channels for the kernel to use.
> > > > > >
> > > > > > Cc: Vinod Koul <vkoul@xxxxxxxxxx>
> > > > > > Cc: Rob Herring <robh+dt@xxxxxxxxxx>
> > > > > > Cc: Mark Rutland <mark.rutland@xxxxxxx>
> > > > > > Cc: Tanglei Han <hantanglei@xxxxxxxxxx>
> > > > > > Cc: Zhuangluan Su <suzhuangluan@xxxxxxxxxxxxx>
> > > > > > Cc: Ryan Grachek <ryan@xxxxxxxxx>
> > > > > > Cc: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> > > > > > Cc: dmaengine@xxxxxxxxxxxxxxx
> > > > > > Cc: devicetree@xxxxxxxxxxxxxxx
> > > > > > Signed-off-by: John Stultz <john.stultz@xxxxxxxxxx>
> > > > > > ---
> > > > > > Documentation/devicetree/bindings/dma/k3dma.txt | 3 +++
> > > > > > 1 file changed, 3 insertions(+)
> > > > > >
> > > > > > diff --git a/Documentation/devicetree/bindings/dma/k3dma.txt b/Documentation/devicetree/bindings/dma/k3dma.txt
> > > > > > index 10a2f15..1c466c1 100644
> > > > > > --- a/Documentation/devicetree/bindings/dma/k3dma.txt
> > > > > > +++ b/Documentation/devicetree/bindings/dma/k3dma.txt
> > > > > > @@ -14,6 +14,9 @@ Required properties:
> > > > > > have specific request line
> > > > > > - clocks: clock required
> > > > > >
> > > > > > +Optional properties:
> > > > > > +- dma-avail-chan: Bitmask of available physical channels
> > > > > > +
> > > > >
> > > > > This property looks too generic. Since this is specific to HiSi SoCs,
> > > > > this could be "hisi-dma-avail-chan"?
> > > >
> > > > I'm fine to change it, but I'm not sure I fully understand the
> > > > rational. Can you help me understand?
> > > > Are device node-binding names supposed to have global scope? I assumed
> > > > the node property names are basically scoped to the entry?
> > >
> > > IIUC properties documented in subsystem binding (dma.txt in this case)
> > > will have global scope. Those which are not documented in this binding
> > > are specific to vendor IPs and should be prefixed with the vendor
> > > prefix (hisi in this case).
> > >
> > > > Further, having some dma channels be reserved doesn't seem to be too
> > > > unique a concept, so I'm not sure what we gain long term by prefixing
> > > > it?
> > > >
> > >
> > > Right, but this brings up the point of having this functionality in
> > > generic DMA engine so that the DMA controller drivers need not handle.
> > > So either we should move this available channel check to DMA Engine
> > > and document the property in dma.txt so that other IPs can also use it
> > > or keep the functionality in K3 driver and use HiSi prefix for the
> > > property.
> > >
> > > But I'd like to hear Vinod/Rob's opinion on this!
> >
> > So there are two parts, first is if this new property of using 'some'
> > channels of controller is generic enough, the answer is unfortunately
> > yes, so we should move this to dma.txt as a generic property
> >
> > But I don't agree the dmaengine core should handle it, we may add
> > helpers, but controllers registers N channels and they would do so, core
> > should not do filtering
> >
>
> Okay. But won't it create ambiguity? What if a new driver developer
> assmes that he can use this property to filter the channels for his own
> DMA controller? Since we are _explicitly_ stating that these channels
> should be filtered, why the dmaengine core can't handle it?
>
> If the property is generic, then it makes sense to keep the
> functionality also generic.

Core doesnt have a view of channels to be filtered, it looks at N
channels getting registered and works on those.

Till now folks do not create channels for 'filtered' ones and register
the ones kernel can use..

--
~Vinod