RE: [PATCH v2 5/9] arm64: dts: ls1046a: make dma-coherent global to the SoC

From: Leo Li
Date: Wed Sep 28 2022 - 19:16:07 EST




> -----Original Message-----
> From: Sean Anderson <sean.anderson@xxxxxxxx>
> Sent: Friday, September 23, 2022 11:35 AM
> To: Leo Li <leoyang.li@xxxxxxx>; shawnguo@xxxxxxxxxx;
> devicetree@xxxxxxxxxxxxxxx
> Cc: robh+dt@xxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; Laurentiu Tudor <laurentiu.tudor@xxxxxxx>
> Subject: Re: [PATCH v2 5/9] arm64: dts: ls1046a: make dma-coherent global
> to the SoC
>
>
>
> On 9/23/22 12:26 PM, Leo Li wrote:
> >
> >
> >> -----Original Message-----
> >> From: Sean Anderson <sean.anderson@xxxxxxxx>
> >> Sent: Friday, September 23, 2022 11:11 AM
> >> To: Leo Li <leoyang.li@xxxxxxx>; shawnguo@xxxxxxxxxx;
> >> devicetree@xxxxxxxxxxxxxxx
> >> Cc: robh+dt@xxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-
> >> kernel@xxxxxxxxxxxxxxx; Laurentiu Tudor <laurentiu.tudor@xxxxxxx>
> >> Subject: Re: [PATCH v2 5/9] arm64: dts: ls1046a: make dma-coherent
> >> global to the SoC
> >>
> >>
> >> Hi All,
> >>
> >> On 9/15/22 7:34 PM, Li Yang wrote:
> >> > These SoCs are really completely dma coherent in their entirety so
> >> > add the dma-coherent property at the soc level in the device tree
> >> > and drop the instances where it's specifically added to a few select
> devices.
> >> >
> >> > Signed-off-by: Laurentiu Tudor <laurentiu.tudor@xxxxxxx>
> >> > Signed-off-by: Li Yang <leoyang.li@xxxxxxx>
> >> > ---
> >> > arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi | 5 +----
> >> > 1 file changed, 1 insertion(+), 4 deletions(-)
> >> >
> >> > diff --git a/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi
> >> > b/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi
> >> > index 27033c558e3e..e406499a26b4 100644
> >> > --- a/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi
> >> > +++ b/arch/arm64/boot/dts/freescale/fsl-ls1046a.dtsi
> >> > @@ -273,6 +273,7 @@ soc: soc {
> >> > #size-cells = <2>;
> >> > ranges;
> >> > dma-ranges = <0x0 0x0 0x0 0x0 0x10000 0x00000000>;
> >> > + dma-coherent;
> >> >
> >> > ddr: memory-controller@1080000 {
> >> > compatible = "fsl,qoriq-memory-controller"; @@ -
> >> 355,7 +356,6 @@
> >> > crypto: crypto@1700000 {
> >> > ranges = <0x0 0x00 0x1700000 0x100000>;
> >> > reg = <0x00 0x1700000 0x0 0x100000>;
> >> > interrupts = <GIC_SPI 75 IRQ_TYPE_LEVEL_HIGH>;
> >> > - dma-coherent;
> >> >
> >> > sec_jr0: jr@10000 {
> >> > compatible = "fsl,sec-v5.4-job-ring", @@ -
> >> 794,7 +794,6 @@ pcie1:
> >> > pcie@3400000 {
> >> > #address-cells = <3>;
> >> > #size-cells = <2>;
> >> > device_type = "pci";
> >> > - dma-coherent;
> >> > num-viewport = <8>;
> >> > bus-range = <0x0 0xff>;
> >> > ranges = <0x81000000 0x0 0x00000000 0x40
> >> 0x00010000 0x0 0x00010000 /* downstream I/O */
> >> > @@ -834,7 +833,6 @@ pcie2: pcie@3500000 {
> >> > #address-cells = <3>;
> >> > #size-cells = <2>;
> >> > device_type = "pci";
> >> > - dma-coherent;
> >> > num-viewport = <8>;
> >> > bus-range = <0x0 0xff>;
> >> > ranges = <0x81000000 0x0 0x00000000 0x48
> >> 0x00010000 0x0 0x00010000 /* downstream I/O */
> >> > @@ -874,7 +872,6 @@ pcie3: pcie@3600000 {
> >> > #address-cells = <3>;
> >> > #size-cells = <2>;
> >> > device_type = "pci";
> >> > - dma-coherent;
> >> > num-viewport = <8>;
> >> > bus-range = <0x0 0xff>;
> >> > ranges = <0x81000000 0x0 0x00000000 0x50
> >> 0x00010000 0x0 0x00010000 /* downstream I/O */
> >> >
> >>
> >> I'd like to summarize the conclusions of [1] below. This patch breaks
> >> I2C0, which is the only user of eDMA at the moment. eDMA is
> >> noncoherent because snooping is not enabled for it. I have submitted
> >> a patch [2] to U- Boot to enable snooping for eDMA. For now, this
> >> patch must add dma- noncoherent to the i2c0 node.
> >
> > I have sent a V3 yesterday to set dma-noncoherent on edma node. But are
> you saying that the dma-noncoherent need to be added to the i2c node to
> make it work?
>
> I believe dma coherency is a property of the consumer, not the provider. See
> e.g.
> really_probe/platform_dma_configure/of_dma_configure/of_dma_is_cohe
> rent.

You are probably right. The provider dma driver only maps the MMIO register while the consumer i2c driver maps the data buffer in memory which needs correct coherency setting.

>
> > For the u-boot patch, I will check with the hardware team to see if it is safe
> to set the reserved bit for edma snooping.
>
> Thanks. I'm curious as to whether this omission is intentional or not.

I have asked around but no one seems to know why LS1046a is different. So from your experiment, changing the reserved bit does make a impact on the final result? And setting it together with i2c nodes having dma-coherent enabled make it work. Right? If that's the case, probably it is just a documentation issue that we should fix.

- Leo