Re: [PATCH v8 13/18] media: ti: j721e-csi2rx: add multistream support

[Jai Luthra]

Quoting Tomi Valkeinen (2025-12-09 17:22:08)
> Hi,
> 
> On 09/12/2025 12:36, Jai Luthra wrote:
> > Hi Rishikesh,
> > 
> > Quoting Rishikesh Donadkar (2025-12-09 15:38:33)
> >>
> >> On 01/12/25 18:33, Tomi Valkeinen wrote:
> >>> Hi,
> >>
> >>
> >> Hi Tomi,
> >>
> >> Thank you for the review !
> >>
> >>>
> >>> On 12/11/2025 13:54, Rishikesh Donadkar wrote:
> >>>> From: Jai Luthra <j-luthra@xxxxxx>
> >>>>
> >>>> Each CSI2 stream can be multiplexed into 4 independent streams, each
> >>> Well, that's not true, at least generally speaking (there can be more
> >>> than 4). Is that specific to TI hardware?
> >>
> >>
> >> Yes, The commit message talks about how TI CSI does the multiplexing of 
> >> CSI stream from the sensor into 4 streams as show in the Figure 12-388 
> >> in AM62A TRM[1]. I will modify the commit message to mention that this 
> >> is TI CSI specific.
> > 
> > Figure 12-388 shows the internal pixel stream coming from Cadence to
> > different hardware blocks like TI's Shim (DMA) and VP0 (ISP) and VP1. I
> > don't see that being related to CSI2 VC/DT support, which is handled by the
> > Shim using different DMA contexts and channels.
> > 
> > In the TRM, under 12.6.1.1.1 CSI_RX_IF Features, I see:
> >  * Compliant to MIPI CSI v1.3
> >  * Supports up to 16 virtual channels per input (partial MIPI CSI v2.0 feature)
> > 
> > So 16 VCs is supported by TI's CSI pipeline, despite it being MIPI CSI2
> > v1.3 compliant otherwise. I think I might have been confused with DS90UB960
> > while writing this commit message originally, which strictly supports a
> > maximum of 4 VCs.
> > 
> > Secondly, even with just CSI2 v1.0 compliant source, this could
> > theoretically handle 8 "streams" of data with 4 different VCs x 2 data
> > types each. So please update the paragraph in next revision.
> Where does the "8" come from? Do we have 8 context registers, to which
> we program the VC + DT filter?

Indeed, 8 was just an example. The hardware can support upto 32
combinations, as that is the total number of DMA contexts available.

Thanks,
    Jai

> 
> Also, it could as well be 8 streams, all with VC 0, but each different
> DT (probably unlikely scenario =). But I just want to highlight that VC
> is not the "stream". The "stream" is the VC+DT tuple.
> 
>  Tomi
>