diff mbox series

arm64: dts: juno: Remove GICv2m dma-range

Message ID 856c3f7192c6c3ce545ba67462f2ce9c86ed6b0c.1643046936.git.robin.murphy@arm.com (mailing list archive)
State New, archived
Headers show
Series arm64: dts: juno: Remove GICv2m dma-range | expand

Commit Message

Robin Murphy Jan. 24, 2022, 5:57 p.m. UTC
Although it is painstakingly honest to describe all 3 PCI windows in
"dma-ranges", it misses the the subtle distinction that the window for
the GICv2m range is normally programmed for Device memory attributes
rather than Normal Cacheable like the DRAM windows. Since MMU-401 only
offers stage 2 translation, this means that when the PCI SMMU is
enabled, accesses through that IPA range unexpectedly lose coherency if
mapped as cacheable at the SMMU, due to the attribute combining rules.
Since an extra 256KB is neither here nor there when we still have 10GB
worth of usable address space, rather than attempting to describe and
cope with this detail let's just remove the offending range. If the SMMU
is not used then it makes no difference anyway.

Fixes: 4ac4d146cb63 ("arm64: dts: juno: Describe PCI dma-ranges")
Reported-by: Anders Roxell <anders.roxell@linaro.org>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
---

Sorry it took a while to catch up on this...

 arch/arm64/boot/dts/arm/juno-base.dtsi | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

Comments

Liviu Dudau Jan. 25, 2022, 3:24 p.m. UTC | #1
On Mon, Jan 24, 2022 at 05:57:01PM +0000, Robin Murphy wrote:
> Although it is painstakingly honest to describe all 3 PCI windows in
> "dma-ranges", it misses the the subtle distinction that the window for
> the GICv2m range is normally programmed for Device memory attributes
> rather than Normal Cacheable like the DRAM windows. Since MMU-401 only
> offers stage 2 translation, this means that when the PCI SMMU is
> enabled, accesses through that IPA range unexpectedly lose coherency if
> mapped as cacheable at the SMMU, due to the attribute combining rules.
> Since an extra 256KB is neither here nor there when we still have 10GB
> worth of usable address space, rather than attempting to describe and
> cope with this detail let's just remove the offending range. If the SMMU
> is not used then it makes no difference anyway.
> 
> Fixes: 4ac4d146cb63 ("arm64: dts: juno: Describe PCI dma-ranges")
> Reported-by: Anders Roxell <anders.roxell@linaro.org>
> Signed-off-by: Robin Murphy <robin.murphy@arm.com>

Looks like a good idea.

Acked-by: Liviu Dudau <liviu.dudau@arm.com>

Sudeep, can you pick this one up through your tree?

Best regards,
Liviu

> ---
> 
> Sorry it took a while to catch up on this...
> 
>  arch/arm64/boot/dts/arm/juno-base.dtsi | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/boot/dts/arm/juno-base.dtsi b/arch/arm64/boot/dts/arm/juno-base.dtsi
> index 6288e104a089..a2635b14da30 100644
> --- a/arch/arm64/boot/dts/arm/juno-base.dtsi
> +++ b/arch/arm64/boot/dts/arm/juno-base.dtsi
> @@ -543,8 +543,7 @@ pcie_ctlr: pcie@40000000 {
>  			 <0x02000000 0x00 0x50000000 0x00 0x50000000 0x0 0x08000000>,
>  			 <0x42000000 0x40 0x00000000 0x40 0x00000000 0x1 0x00000000>;
>  		/* Standard AXI Translation entries as programmed by EDK2 */
> -		dma-ranges = <0x02000000 0x0 0x2c1c0000 0x0 0x2c1c0000 0x0 0x00040000>,
> -			     <0x02000000 0x0 0x80000000 0x0 0x80000000 0x0 0x80000000>,
> +		dma-ranges = <0x02000000 0x0 0x80000000 0x0 0x80000000 0x0 0x80000000>,
>  			     <0x43000000 0x8 0x00000000 0x8 0x00000000 0x2 0x00000000>;
>  		#interrupt-cells = <1>;
>  		interrupt-map-mask = <0 0 0 7>;
> -- 
> 2.28.0.dirty
>
Sudeep Holla Jan. 26, 2022, 10:26 a.m. UTC | #2
On Mon, 24 Jan 2022 17:57:01 +0000, Robin Murphy wrote:
> Although it is painstakingly honest to describe all 3 PCI windows in
> "dma-ranges", it misses the the subtle distinction that the window for
> the GICv2m range is normally programmed for Device memory attributes
> rather than Normal Cacheable like the DRAM windows. Since MMU-401 only
> offers stage 2 translation, this means that when the PCI SMMU is
> enabled, accesses through that IPA range unexpectedly lose coherency if
> mapped as cacheable at the SMMU, due to the attribute combining rules.
> Since an extra 256KB is neither here nor there when we still have 10GB
> worth of usable address space, rather than attempting to describe and
> cope with this detail let's just remove the offending range. If the SMMU
> is not used then it makes no difference anyway.
> 
> [...]

Applied to sudeep.holla/linux (for-next/juno), thanks!


[1/1] arm64: dts: juno: Remove GICv2m dma-range
      https://git.kernel.org/sudeep.holla/c/31eeb6b09f

--

Regards,
Sudeep
diff mbox series

Patch

diff --git a/arch/arm64/boot/dts/arm/juno-base.dtsi b/arch/arm64/boot/dts/arm/juno-base.dtsi
index 6288e104a089..a2635b14da30 100644
--- a/arch/arm64/boot/dts/arm/juno-base.dtsi
+++ b/arch/arm64/boot/dts/arm/juno-base.dtsi
@@ -543,8 +543,7 @@  pcie_ctlr: pcie@40000000 {
 			 <0x02000000 0x00 0x50000000 0x00 0x50000000 0x0 0x08000000>,
 			 <0x42000000 0x40 0x00000000 0x40 0x00000000 0x1 0x00000000>;
 		/* Standard AXI Translation entries as programmed by EDK2 */
-		dma-ranges = <0x02000000 0x0 0x2c1c0000 0x0 0x2c1c0000 0x0 0x00040000>,
-			     <0x02000000 0x0 0x80000000 0x0 0x80000000 0x0 0x80000000>,
+		dma-ranges = <0x02000000 0x0 0x80000000 0x0 0x80000000 0x0 0x80000000>,
 			     <0x43000000 0x8 0x00000000 0x8 0x00000000 0x2 0x00000000>;
 		#interrupt-cells = <1>;
 		interrupt-map-mask = <0 0 0 7>;