Re: [PATCH RFC] hw/arm/virt: Avoid unexpected warning from Linux guest on host with Fujitsu CPUs

[Zhenyu Zhang]

On Fri, Jun 7, 2024 at 6:17 PM Peter Maydell <peter.maydell@linaro.org> wrote:
>
> On Thu, 6 Jun 2024 at 22:18, Robin Murphy <robin.murphy@arm.com> wrote:
> >
> > On 2024-06-06 6:13 pm, Jonathan Cameron wrote:
> > > On Thu, 6 Jun 2024 12:56:59 +0100
> > > Peter Maydell <peter.maydell@linaro.org> wrote:
> > >
> > >> On Thu, 6 Jun 2024 at 11:48, Zhenyu Zhang <zhenyzha@redhat.com> wrote:
> > >>> In Linux, a check is applied to every device which is exposed through 
> > >>> device-tree
> > >>> node. The warning message is raised when the device isn't DMA coherent 
> > >>> and the
> > >>> cache line size is larger than ARCH_DMA_MINALIGN (128 bytes). The cache 
> > >>> line is
> > >>> sorted from CTR_EL0[CWG], which corresponds to 256 bytes on the guest 
> > >>> CPUs.
> > >>> The DMA coherent capability is claimed through 'dma-coherent' in their
> > >>> device-tree nodes.
> > >>
> > >> For QEMU emulated all our DMA is always coherent, so where we
> > >> have DMA-capable devices we should definitely tell the kernel
> > >> that that DMA is coherent.
> >
> > The trick for that is to put the "dma-coherent" property right in the
> > root of the DT so it plausibly communicates "the whole platform is
> > coherent", and is then inherited by all devices, even those which
> > shouldn't technically need it.
>
> Ah, cool -- that's a pretty small change and it makes sense, and
> avoids us having potential bugs in future where we forget to
> mark a really-does-do-DMA device as dma-coherent. I like that
> a lot better than adding incorrect dma-coherent tags to lots
> of device nodes that aren't for DMA-capable devices.
Hi Robin, Peter, and Jonathan,

Thanks for your suggestions, it helps me a lot!
I will submit a v2 patch that puts the "dma-coherent" property right
in the root of the DT so it plausibly communicates.
Yes, these discussions make sense.
I will modify and test it again on Fujitsu host.

Thanks again
Zhenyu

>
> thanks
> -- PMM
>