Re: [PATCH RESEND 0/2] media: rockchip: rga: Add rk3568 support

[Robin Murphy]

On 2023-05-22 11:29, Michael Tretter wrote:
> On Sun, 21 May 2023 21:32:51 +0200, Diederik de Haas wrote:
> > On Sunday, 21 May 2023 12:44:58 CEST Heiko Stuebner wrote:
> > > On Fri, 20 Jan 2023 10:14:21 +0100, Michael Tretter wrote:
> > > > The RGA2 on the Rockchip rk3568 is the same core as the RGA2 on the
> > > > Rockchip rk3288.
> > > >
> > > > This series adds the necessary device tree binding and node in the device
> > > > tree to enable the RGA2 on the Rockchip rk3568.
> > > >
> > > > I tested the driver with the GStreamer v4l2convert element on a Rock3
> > > > Model A board.
> > > >
> > > > [...]
> > >
> > > Applied, thanks!
> > >
> > > [1/2] media: dt-bindings: media: rockchip-rga: add rockchip,rk3568-rga
> > >        commit: 9b12ceb5a80d1fb45d293265de100e33b5843943
> > > [2/2] arm64: dts: rockchip: Add RGA2 support to rk356x
> > >        commit: 0c3391f8bb06b744df521651534cd99e3d77e0a8
> >
> > https://lore.kernel.org/all/TY3P286MB26115F60D273E840D36A610598CA9@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> >
> > indicated that there was a problem with device >= 4GB (RAM?):
> > > Since we have the over-4GB problem now, should we mark this problem as a
> > > TODO or something?
> >
> > I thought that was the reason that these patches weren't picked up before?
>
> That's what I thought, too.
>
> > I have no insight into this problem, so I can't comment on the technical
> > aspects, but I had made a note for myself 'locally' about it.
>
> Using the RGA2 with the driver in its current form on devices with more than 4
> GB system memory may lead to memory corruption as buffer addresses are
> silently truncated to 32 bits.

That's because the driver is completely broken and is not using the DMA 
API anywhere near properly. The fact that it's been getting away with it 
so far can be mostly put down to good luck and nobody using 
CONFIG_DMA_API_DEBUG.

> I'm not sure if that's actually a blocker for merging these patches.

If anything, hopefully getting more SoC support merged might provide 
more motivation for someone to fix the existing code :)

Robin.