[PATCH 0/3] arm64: qcom: sm8650: misc enhancements

[PATCH 0/3] arm64: qcom: sm8650: misc enhancements

[Neil Armstrong]

Misc enhancements for the SM8650 platform:
- update the cpus capacity-dmips-mhz
- add the CPU cache sizes
- correct the soundwire ports

Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
---
Neil Armstrong (3):
      arm64: dts: qcom: sm8650: update the cpus capacity-dmips-mhz
      arm64: dts: qcom: sm8650: add CPU cache size properties
      arm64: dts: qcom: sm8650: fix soundwire ports properties

 arch/arm64/boot/dts/qcom/sm8650.dtsi | 110 ++++++++++++++++++++++++++---------
 1 file changed, 83 insertions(+), 27 deletions(-)
---
base-commit: 63804fed149a6750ffd28610c5c1c98cce6bd377
change-id: 20260128-topic-sm8650-upstream-cpu-props-0754ccef3e01

Best regards,
-- 
Neil Armstrong <neil.armstrong@linaro.org>

[PATCH 1/3] arm64: dts: qcom: sm8650: update the cpus capacity-dmips-mhz

[Neil Armstrong]

After some more advanced benchmarks with Integer, Floaring Point,
Encryption, Compression, NEON, ... on the A520, A720 and X4 cpus,
the median gain with the same frequency range is:
- 281% of A720 over A520
- 126% of X4 over A720

When adjusted with the frequency delta, we get better values
describing the difference in capacity, showing the weakness of
the A520 designed for very small tasks while the A720 and X4
are much more powerful.

Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
---
 arch/arm64/boot/dts/qcom/sm8650.dtsi | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/arch/arm64/boot/dts/qcom/sm8650.dtsi b/arch/arm64/boot/dts/qcom/sm8650.dtsi
index f8e1950a74ac..8671c25dd68f 100644
--- a/arch/arm64/boot/dts/qcom/sm8650.dtsi
+++ b/arch/arm64/boot/dts/qcom/sm8650.dtsi
@@ -153,7 +153,7 @@ cpu2: cpu@200 {
 
 			enable-method = "psci";
 			next-level-cache = <&l2_200>;
-			capacity-dmips-mhz = <1792>;
+			capacity-dmips-mhz = <2909>;
 			dynamic-power-coefficient = <238>;
 
 			qcom,freq-domain = <&cpufreq_hw 3>;
@@ -189,7 +189,7 @@ cpu3: cpu@300 {
 
 			enable-method = "psci";
 			next-level-cache = <&l2_300>;
-			capacity-dmips-mhz = <1792>;
+			capacity-dmips-mhz = <2909>;
 			dynamic-power-coefficient = <238>;
 
 			qcom,freq-domain = <&cpufreq_hw 3>;
@@ -225,7 +225,7 @@ cpu4: cpu@400 {
 
 			enable-method = "psci";
 			next-level-cache = <&l2_400>;
-			capacity-dmips-mhz = <1792>;
+			capacity-dmips-mhz = <2909>;
 			dynamic-power-coefficient = <238>;
 
 			qcom,freq-domain = <&cpufreq_hw 3>;
@@ -261,7 +261,7 @@ cpu5: cpu@500 {
 
 			enable-method = "psci";
 			next-level-cache = <&l2_500>;
-			capacity-dmips-mhz = <1792>;
+			capacity-dmips-mhz = <2909>;
 			dynamic-power-coefficient = <238>;
 
 			qcom,freq-domain = <&cpufreq_hw 1>;
@@ -297,7 +297,7 @@ cpu6: cpu@600 {
 
 			enable-method = "psci";
 			next-level-cache = <&l2_600>;
-			capacity-dmips-mhz = <1792>;
+			capacity-dmips-mhz = <2909>;
 			dynamic-power-coefficient = <238>;
 
 			qcom,freq-domain = <&cpufreq_hw 1>;
@@ -333,7 +333,7 @@ cpu7: cpu@700 {
 
 			enable-method = "psci";
 			next-level-cache = <&l2_700>;
-			capacity-dmips-mhz = <1894>;
+			capacity-dmips-mhz = <3591>;
 			dynamic-power-coefficient = <588>;
 
 			qcom,freq-domain = <&cpufreq_hw 2>;

-- 
2.34.1

Re: [PATCH 1/3] arm64: dts: qcom: sm8650: update the cpus capacity-dmips-mhz

[Konrad Dybcio]

On 1/28/26 11:33 AM, Neil Armstrong wrote:
> After some more advanced benchmarks with Integer, Floaring Point,
> Encryption, Compression, NEON, ... on the A520, A720 and X4 cpus,
> the median gain with the same frequency range is:
> - 281% of A720 over A520
> - 126% of X4 over A720

While this is all magic numbers, I am definitely willing to believe
that x4 is 26% faster than a720 rather than the currently stated ~5.7%

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

[PATCH 2/3] arm64: dts: qcom: sm8650: add CPU cache size properties

[Neil Armstrong]

Add the L1 cache size and its line size (cache-size and
cache-line-size) with the corresponding L1-I cache and L1-D cache.

L1 cache is unified, but clidr_el1 register (get_cache_type) tells that
L1 cache is separated (CACHE_TYPE_SEPARATE), add i-cache-line-size and
d-cache-line-size and cache-line-size of L3 cache is specified.

All cache line sizes were confirmed by checking ccsidr_el1.

Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
---
 arch/arm64/boot/dts/qcom/sm8650.dtsi | 56 ++++++++++++++++++++++++++++++++++++
 1 file changed, 56 insertions(+)

diff --git a/arch/arm64/boot/dts/qcom/sm8650.dtsi b/arch/arm64/boot/dts/qcom/sm8650.dtsi
index 8671c25dd68f..f8563ec79dc6 100644
--- a/arch/arm64/boot/dts/qcom/sm8650.dtsi
+++ b/arch/arm64/boot/dts/qcom/sm8650.dtsi
@@ -75,6 +75,11 @@ cpu0: cpu@0 {
 			compatible = "arm,cortex-a520";
 			reg = <0 0>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 0>;
 
 			power-domains = <&cpu_pd0>;
@@ -103,11 +108,15 @@ l2_0: l2-cache {
 				cache-level = <2>;
 				cache-unified;
 				next-level-cache = <&l3_0>;
+				cache-size = <262144>;
+				cache-line-size = <64>;
 
 				l3_0: l3-cache {
 					compatible = "cache";
 					cache-level = <3>;
 					cache-unified;
+					cache-size = <12582912>;
+					cache-line-size = <64>;
 				};
 			};
 		};
@@ -117,6 +126,11 @@ cpu1: cpu@100 {
 			compatible = "arm,cortex-a520";
 			reg = <0 0x100>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 0>;
 
 			power-domains = <&cpu_pd1>;
@@ -146,6 +160,11 @@ cpu2: cpu@200 {
 			compatible = "arm,cortex-a720";
 			reg = <0 0x200>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 3>;
 
 			power-domains = <&cpu_pd2>;
@@ -174,6 +193,8 @@ l2_200: l2-cache {
 				cache-level = <2>;
 				cache-unified;
 				next-level-cache = <&l3_0>;
+				cache-size = <524288>;
+				cache-line-size = <64>;
 			};
 		};
 
@@ -182,6 +203,11 @@ cpu3: cpu@300 {
 			compatible = "arm,cortex-a720";
 			reg = <0 0x300>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 3>;
 
 			power-domains = <&cpu_pd3>;
@@ -210,6 +236,8 @@ l2_300: l2-cache {
 				cache-level = <2>;
 				cache-unified;
 				next-level-cache = <&l3_0>;
+				cache-size = <524288>;
+				cache-line-size = <64>;
 			};
 		};
 
@@ -218,6 +246,11 @@ cpu4: cpu@400 {
 			compatible = "arm,cortex-a720";
 			reg = <0 0x400>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 3>;
 
 			power-domains = <&cpu_pd4>;
@@ -246,6 +279,8 @@ l2_400: l2-cache {
 				cache-level = <2>;
 				cache-unified;
 				next-level-cache = <&l3_0>;
+				cache-size = <524288>;
+				cache-line-size = <64>;
 			};
 		};
 
@@ -254,6 +289,11 @@ cpu5: cpu@500 {
 			compatible = "arm,cortex-a720";
 			reg = <0 0x500>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 1>;
 
 			power-domains = <&cpu_pd5>;
@@ -282,6 +322,8 @@ l2_500: l2-cache {
 				cache-level = <2>;
 				cache-unified;
 				next-level-cache = <&l3_0>;
+				cache-size = <524288>;
+				cache-line-size = <64>;
 			};
 		};
 
@@ -290,6 +332,11 @@ cpu6: cpu@600 {
 			compatible = "arm,cortex-a720";
 			reg = <0 0x600>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 1>;
 
 			power-domains = <&cpu_pd6>;
@@ -318,6 +365,8 @@ l2_600: l2-cache {
 				cache-level = <2>;
 				cache-unified;
 				next-level-cache = <&l3_0>;
+				cache-size = <524288>;
+				cache-line-size = <64>;
 			};
 		};
 
@@ -326,6 +375,11 @@ cpu7: cpu@700 {
 			compatible = "arm,cortex-x4";
 			reg = <0 0x700>;
 
+			i-cache-size = <65536>;
+			i-cache-line-size = <64>;
+			d-cache-size = <65536>;
+			d-cache-line-size = <64>;
+
 			clocks = <&cpufreq_hw 2>;
 
 			power-domains = <&cpu_pd7>;
@@ -354,6 +408,8 @@ l2_700: l2-cache {
 				cache-level = <2>;
 				cache-unified;
 				next-level-cache = <&l3_0>;
+				cache-size = <2097152>;
+				cache-line-size = <64>;
 			};
 		};
 

-- 
2.34.1

Re: [PATCH 2/3] arm64: dts: qcom: sm8650: add CPU cache size properties

[Konrad Dybcio]

On 1/28/26 11:33 AM, Neil Armstrong wrote:
> Add the L1 cache size and its line size (cache-size and
> cache-line-size) with the corresponding L1-I cache and L1-D cache.
> 
> L1 cache is unified, but clidr_el1 register (get_cache_type) tells that
> L1 cache is separated (CACHE_TYPE_SEPARATE), add i-cache-line-size and
> d-cache-line-size and cache-line-size of L3 cache is specified.
> 
> All cache line sizes were confirmed by checking ccsidr_el1.
> 
> Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
> ---
>  arch/arm64/boot/dts/qcom/sm8650.dtsi | 56 ++++++++++++++++++++++++++++++++++++
>  1 file changed, 56 insertions(+)
> 
> diff --git a/arch/arm64/boot/dts/qcom/sm8650.dtsi b/arch/arm64/boot/dts/qcom/sm8650.dtsi
> index 8671c25dd68f..f8563ec79dc6 100644
> --- a/arch/arm64/boot/dts/qcom/sm8650.dtsi
> +++ b/arch/arm64/boot/dts/qcom/sm8650.dtsi
> @@ -75,6 +75,11 @@ cpu0: cpu@0 {
>  			compatible = "arm,cortex-a520";
>  			reg = <0 0>;
>  
> +			i-cache-size = <65536>;
> +			i-cache-line-size = <64>;
> +			d-cache-size = <65536>;
> +			d-cache-line-size = <64>;
> +
>  			clocks = <&cpufreq_hw 0>;
>  
>  			power-domains = <&cpu_pd0>;
> @@ -103,11 +108,15 @@ l2_0: l2-cache {
>  				cache-level = <2>;
>  				cache-unified;
>  				next-level-cache = <&l3_0>;
> +				cache-size = <262144>;

This should be 512 kiB, shared for c0/1, according to a block
diagram

Everything else lgtm, although I can't find a source for the
cache*line* sizes

Konrad

Re: [PATCH 2/3] arm64: dts: qcom: sm8650: add CPU cache size properties

[Neil Armstrong]

On 1/28/26 11:46, Konrad Dybcio wrote:
> On 1/28/26 11:33 AM, Neil Armstrong wrote:
>> Add the L1 cache size and its line size (cache-size and
>> cache-line-size) with the corresponding L1-I cache and L1-D cache.
>>
>> L1 cache is unified, but clidr_el1 register (get_cache_type) tells that
>> L1 cache is separated (CACHE_TYPE_SEPARATE), add i-cache-line-size and
>> d-cache-line-size and cache-line-size of L3 cache is specified.
>>
>> All cache line sizes were confirmed by checking ccsidr_el1.
>>
>> Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
>> ---
>>   arch/arm64/boot/dts/qcom/sm8650.dtsi | 56 ++++++++++++++++++++++++++++++++++++
>>   1 file changed, 56 insertions(+)
>>
>> diff --git a/arch/arm64/boot/dts/qcom/sm8650.dtsi b/arch/arm64/boot/dts/qcom/sm8650.dtsi
>> index 8671c25dd68f..f8563ec79dc6 100644
>> --- a/arch/arm64/boot/dts/qcom/sm8650.dtsi
>> +++ b/arch/arm64/boot/dts/qcom/sm8650.dtsi
>> @@ -75,6 +75,11 @@ cpu0: cpu@0 {
>>   			compatible = "arm,cortex-a520";
>>   			reg = <0 0>;
>>   
>> +			i-cache-size = <65536>;
>> +			i-cache-line-size = <64>;
>> +			d-cache-size = <65536>;
>> +			d-cache-line-size = <64>;
>> +
>>   			clocks = <&cpufreq_hw 0>;
>>   
>>   			power-domains = <&cpu_pd0>;
>> @@ -103,11 +108,15 @@ l2_0: l2-cache {
>>   				cache-level = <2>;
>>   				cache-unified;
>>   				next-level-cache = <&l3_0>;
>> +				cache-size = <262144>;
> 
> This should be 512 kiB, shared for c0/1, according to a block
> diagram

Thanks for confirming !

Neil

> 
> Everything else lgtm, although I can't find a source for the
> cache*line* sizes
> 
> Konrad

[PATCH 3/3] arm64: dts: qcom: sm8650: fix soundwire ports properties

[Neil Armstrong]

Since commit 9e53a66a2f2f ("soundwire: qcom: deprecate qcom,din/out-ports"),
the ports are checked against the actul hardware configuration, leading to:
qcom-soundwire 6ad0000.soundwire: din-ports (0) mismatch with controller (1)
qcom-soundwire 6d30000.soundwire: dout-ports (0) mismatch with controller (1)

Fix the ports count and properties of the corresponding soundwire
controllers.

Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
---
 arch/arm64/boot/dts/qcom/sm8650.dtsi | 42 ++++++++++++++++++------------------
 1 file changed, 21 insertions(+), 21 deletions(-)

diff --git a/arch/arm64/boot/dts/qcom/sm8650.dtsi b/arch/arm64/boot/dts/qcom/sm8650.dtsi
index f8563ec79dc6..3ddb5c3f097c 100644
--- a/arch/arm64/boot/dts/qcom/sm8650.dtsi
+++ b/arch/arm64/boot/dts/qcom/sm8650.dtsi
@@ -4728,18 +4728,18 @@ swr1: soundwire@6ad0000 {
 			pinctrl-0 = <&rx_swr_active>;
 			pinctrl-names = "default";
 
-			qcom,din-ports = <0>;
+			qcom,din-ports = <1>;
 			qcom,dout-ports = <11>;
 
-			qcom,ports-sinterval =		/bits/ 16 <0x03 0x1f 0x1f 0x07 0x03 0xff 0xff 0x31 0xff 0xff 0xff>;
-			qcom,ports-offset1 =		/bits/ 8 <0x00 0x00 0x0b 0x09 0x01 0xff 0xff 0x00 0xff 0xff 0xff>;
-			qcom,ports-offset2 =		/bits/ 8 <0x00 0x00 0x0b 0x00 0x00 0xff 0xff 0x00 0xff 0xff 0xff>;
-			qcom,ports-hstart =		/bits/ 8 <0xff 0x03 0xff 0xff 0xff 0xff 0xff 0x00 0xff 0xff 0xff>;
-			qcom,ports-hstop =		/bits/ 8 <0xff 0x06 0xff 0xff 0xff 0xff 0xff 0x0f 0xff 0xff 0xff>;
-			qcom,ports-word-length =	/bits/ 8 <0x01 0x07 0x04 0xff 0xff 0xff 0xff 0x18 0xff 0xff 0xff>;
-			qcom,ports-block-pack-mode =	/bits/ 8 <0xff 0x00 0x01 0xff 0xff 0xff 0xff 0x01 0xff 0xff 0xff>;
-			qcom,ports-block-group-count =	/bits/ 8 <0xff 0xff 0xff 0x01 0x03 0xff 0xff 0x00 0xff 0xff 0xff>;
-			qcom,ports-lane-control =	/bits/ 8 <0x01 0x00 0x00 0x00 0x00 0xff 0xff 0x01 0xff 0xff 0xff>;
+			qcom,ports-sinterval =		/bits/ 16 <0x03 0x1f 0x1f 0x07 0x03 0xff 0xff 0x31 0xff 0xff 0xff 0xff>;
+			qcom,ports-offset1 =		/bits/ 8 <0x00 0x00 0x0b 0x09 0x01 0xff 0xff 0x00 0xff 0xff 0xff 0xff>;
+			qcom,ports-offset2 =		/bits/ 8 <0x00 0x00 0x0b 0x00 0x00 0xff 0xff 0x00 0xff 0xff 0xff 0xff>;
+			qcom,ports-hstart =		/bits/ 8 <0xff 0x03 0xff 0xff 0xff 0xff 0xff 0x00 0xff 0xff 0xff 0xff>;
+			qcom,ports-hstop =		/bits/ 8 <0xff 0x06 0xff 0xff 0xff 0xff 0xff 0x0f 0xff 0xff 0xff 0xff>;
+			qcom,ports-word-length =	/bits/ 8 <0x01 0x07 0x04 0xff 0xff 0xff 0xff 0x18 0xff 0xff 0xff 0xff>;
+			qcom,ports-block-pack-mode =	/bits/ 8 <0xff 0x00 0x01 0xff 0xff 0xff 0xff 0x01 0xff 0xff 0xff 0xff>;
+			qcom,ports-block-group-count =	/bits/ 8 <0xff 0xff 0xff 0x01 0x03 0xff 0xff 0x00 0xff 0xff 0xff 0xff>;
+			qcom,ports-lane-control =	/bits/ 8 <0x01 0x00 0x00 0x00 0x00 0xff 0xff 0x01 0xff 0xff 0xff 0xff>;
 
 			#address-cells = <2>;
 			#size-cells = <0>;
@@ -4825,17 +4825,17 @@ swr2: soundwire@6d30000 {
 			pinctrl-names = "default";
 
 			qcom,din-ports = <4>;
-			qcom,dout-ports = <0>;
-
-			qcom,ports-sinterval-low =	/bits/ 8 <0x01 0x01 0x03 0x03>;
-			qcom,ports-offset1 =		/bits/ 8 <0x00 0x00 0x01 0x01>;
-			qcom,ports-offset2 =		/bits/ 8 <0x00 0x00 0x00 0x00>;
-			qcom,ports-hstart =		/bits/ 8 <0xff 0xff 0xff 0xff>;
-			qcom,ports-hstop =		/bits/ 8 <0xff 0xff 0xff 0xff>;
-			qcom,ports-word-length =	/bits/ 8 <0xff 0xff 0xff 0xff>;
-			qcom,ports-block-pack-mode =	/bits/ 8 <0xff 0xff 0xff 0xff>;
-			qcom,ports-block-group-count =	/bits/ 8 <0xff 0xff 0xff 0xff>;
-			qcom,ports-lane-control =	/bits/ 8 <0x01 0x02 0x00 0x00>;
+			qcom,dout-ports = <1>;
+
+			qcom,ports-sinterval-low =	/bits/ 8 <0x00 0x01 0x01 0x03 0x03>;
+			qcom,ports-offset1 =		/bits/ 8 <0x00 0x00 0x00 0x01 0x01>;
+			qcom,ports-offset2 =		/bits/ 8 <0x00 0x00 0x00 0x00 0x00>;
+			qcom,ports-hstart =		/bits/ 8 <0xff 0xff 0xff 0xff 0xff>;
+			qcom,ports-hstop =		/bits/ 8 <0xff 0xff 0xff 0xff 0xff>;
+			qcom,ports-word-length =	/bits/ 8 <0xff 0xff 0xff 0xff 0xff>;
+			qcom,ports-block-pack-mode =	/bits/ 8 <0xff 0xff 0xff 0xff 0xff>;
+			qcom,ports-block-group-count =	/bits/ 8 <0xff 0xff 0xff 0xff 0xff>;
+			qcom,ports-lane-control =	/bits/ 8 <0xff 0x01 0x02 0x00 0x00>;
 
 			#address-cells = <2>;
 			#size-cells = <0>;

-- 
2.34.1