Hello Fukaumi,
On 2024-10-23 01:30, FUKAUMI Naoki wrote:
On 10/23/24 05:13, Dragan Simic wrote:
On 2024-10-14 07:16, Dragan Simic wrote:
On 2024-10-14 06:38, FUKAUMI Naoki wrote:
On 10/13/24 02:04, Dragan Simic wrote:
Add new SoC dtsi file for the RK3566T variant of the Rockchip
RK3566 SoC.
The difference between the RK3566T variant and the "full-fat"
RK3566 variant
is in fewer supported CPU and GPU OPPs on the RK3566T, and in the
absence of
a functional NPU, which we currently don't have to worry about.
Examples of the boards based on the RK3566T include the Pine64
Quartz64 Zero
SBC, [2] the Radxa ROCK 3C and the Radxa ZERO 3E/3W SBCs.
Unfortunately,
Radxa doesn't mention the use of RK3566T officially, but its
official SBC
specifications do state that the maximum frequency for the
Cortex-A55 cores
on those SBCs is lower than the "full-fat" RK3566's 1.8 GHz, which
makes
spotting the presence of the RK3566T SoC variant rather easy.
[3][4][5] An
additional, helpful cue is that Radxa handles the CPU and GPU OPPs
for the
RK3566T variant separately in its downstream kernel. [6]
The CPU and GPU OPPs supported on the RK3566T SoC variant are taken
from the
vendor kernel source, [1] which uses the values of the
"opp-supported-hw" OPP
properties to determine which ones are supported on a particular
SoC variant.
The actual values of the "opp-supported-hw" properties make it
rather easy
to see what OPPs are supported on the RK3566T SoC variant, but
that, rather
unfortunately, clashes with the maximum frequencies advertised
officially
for the Cortex-A55 CPU cores on the above-mentioned SBCs.
[2][3][4][5] The
vendor kernel source indicates that the maximum frequency for the
CPU cores
is 1.4 GHz, while the SBC specifications state that to be 1.6 GHz.
Unless
that discrepancy is resolved somehow, let's take the safe approach
and use
the lower maximum frequency for the CPU cores.
Update the dts files of the currently supported RK3566T-based
boards to use
the new SoC dtsi for the RK3566T variant. This actually takes the
CPU cores
and the GPUs found on these boards out of their earlier overclocks,
but it
also means that the officially advertised specifications
[2][3][4][5] of the
highest supported frequencies for the Cortex-A55 CPU cores on these
boards
may actually be wrong, as already explained above.
The correctness of the introduced changes was validated by
decompiling and
comparing all affected board dtb files before and after these
changes.
[1]
https://raw.githubusercontent.com/rockchip-linux/kernel/f8b9431ee38ed561650be7092ab93f564598daa9/arch/arm64/boot/dts/rockchip/rk3568.dtsi
[2] https://wiki.pine64.org/wiki/Quartz64
[3]
https://dl.radxa.com/rock3/docs/hw/3c/radxa_rock3c_product_brief.pdf
[4]
https://dl.radxa.com/zero3/docs/hw/3e/radxa_zero_3e_product_brief.pdf
[5]
https://dl.radxa.com/zero3/docs/hw/3w/radxa_zero_3w_product_brief.pdf
[6]
https://github.com/radxa/kernel/commit/2dfd51da472e7ebb5ef0d3db78f902454af826b8
Cc: TL Lim <tllim@xxxxxxxxxx>
Cc: Marek Kraus <gamiee@xxxxxxxxxx>
Cc: Tom Cubie <tom@xxxxxxxxx>
Cc: FUKAUMI Naoki <naoki@xxxxxxxxx>
Helped-by: Nicolas Frattaroli <frattaroli.nicolas@xxxxxxxxx>
Helped-by: Jonas Karlman <jonas@xxxxxxxxx>
Signed-off-by: Dragan Simic <dsimic@xxxxxxxxxxx>
---
.../dts/rockchip/rk3566-radxa-zero-3.dtsi | 2 +-
.../boot/dts/rockchip/rk3566-rock-3c.dts | 2 +-
arch/arm64/boot/dts/rockchip/rk3566t.dtsi | 90
+++++++++++++++++++
3 files changed, 92 insertions(+), 2 deletions(-)
create mode 100644 arch/arm64/boot/dts/rockchip/rk3566t.dtsi
diff --git a/arch/arm64/boot/dts/rockchip/rk3566-radxa-zero-3.dtsi
b/arch/arm64/boot/dts/rockchip/rk3566-radxa-zero-3.dtsi
index de390d92c35e..1ee5d96a46a1 100644
--- a/arch/arm64/boot/dts/rockchip/rk3566-radxa-zero-3.dtsi
+++ b/arch/arm64/boot/dts/rockchip/rk3566-radxa-zero-3.dtsi
@@ -3,7 +3,7 @@
#include <dt-bindings/gpio/gpio.h>
#include <dt-bindings/leds/common.h>
#include <dt-bindings/soc/rockchip,vop2.h>
-#include "rk3566.dtsi"
+#include "rk3566t.dtsi"
could you drop this change for now?
This patch is also going to be used for the upcoming board dts
for the Pine64 Quartz64 Zero, so there's no need for dropping it.
The Quartz64 Zero definitely uses the RK3566T.
We (Radxa) think we use RK3566.
Well, the available documentation for the Radxa ROCK 3C and ZERO
3E/3W boards doesn't say so; instead, everything points to the
RK3566T being used. The referenced commit in the Radxa downstream
kernel also indicates that RK3566T is used at least on some boards.
Also, some independent testing, by reading the efuses, has showed
that at least some ROCK 3C and ZERO 3E/3W boards actually have the
RK3566T, which means that we should handle them all as having the
RK3566T, to avoid overclocking the CPU cores and the GPU. I'll
get back to this below.
and vendor kernel[6] refers efuse value to determine it's -T or not.
can you do similar way?
Unfortunately not at the moment, because that would require major
changes to the way OPPs are handled in the upstream kernel. Maybe
we can do that at some point in the future, as part of my planned
work on supporting SoC binning.
With that in place, hopefully, we could handle any ROCK 3C and ZERO
3E/3W boards that actually have the "full-fat" RK3566 variant as
such, but until then, it's much safer to treat them all as having
the RK3566T, and avoid the possible overclocking.
Just checking, and having the subsequent discussion on IRC in mind,
are you fine with the above-proposed approach? Please let me know
if some further clarification is needed.
we have no objection. please do safer way.
Great, thanks! It's better to be on the safe side, until we get
the full support for SoC binning in the upstream kernel.
/ {
chosen {
diff --git a/arch/arm64/boot/dts/rockchip/rk3566-rock-3c.dts
b/arch/arm64/boot/dts/rockchip/rk3566-rock-3c.dts
index f2cc086e5001..9a8f4f774dbc 100644
--- a/arch/arm64/boot/dts/rockchip/rk3566-rock-3c.dts
+++ b/arch/arm64/boot/dts/rockchip/rk3566-rock-3c.dts
@@ -5,7 +5,7 @@
#include <dt-bindings/leds/common.h>
#include <dt-bindings/pinctrl/rockchip.h>
#include <dt-bindings/soc/rockchip,vop2.h>
-#include "rk3566.dtsi"
+#include "rk3566t.dtsi"
same here.
/ {
model = "Radxa ROCK 3C";
diff --git a/arch/arm64/boot/dts/rockchip/rk3566t.dtsi
b/arch/arm64/boot/dts/rockchip/rk3566t.dtsi
new file mode 100644
index 000000000000..cd89bd3b125b
--- /dev/null
+++ b/arch/arm64/boot/dts/rockchip/rk3566t.dtsi
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: (GPL-2.0+ OR MIT)
+
+#include "rk3566-base.dtsi"
+
+/ {
+ cpu0_opp_table: opp-table-0 {
+ compatible = "operating-points-v2";
+ opp-shared;
+
+ opp-408000000 {
+ opp-hz = /bits/ 64 <408000000>;
+ opp-microvolt = <850000 850000 1150000>;
+ clock-latency-ns = <40000>;
+ };
+
+ opp-600000000 {
+ opp-hz = /bits/ 64 <600000000>;
+ opp-microvolt = <850000 850000 1150000>;
+ clock-latency-ns = <40000>;
+ };
+
+ opp-816000000 {
+ opp-hz = /bits/ 64 <816000000>;
+ opp-microvolt = <850000 850000 1150000>;
+ clock-latency-ns = <40000>;
+ opp-suspend;
+ };
+
+ opp-1104000000 {
+ opp-hz = /bits/ 64 <1104000000>;
+ opp-microvolt = <900000 900000 1150000>;
+ clock-latency-ns = <40000>;
+ };
+
+ opp-1416000000 {
+ opp-hz = /bits/ 64 <1416000000>;
+ opp-microvolt = <1025000 1025000 1150000>;
+ clock-latency-ns = <40000>;
+ };
+ };
+
+ gpu_opp_table: opp-table-1 {
+ compatible = "operating-points-v2";
+
+ opp-200000000 {
+ opp-hz = /bits/ 64 <200000000>;
+ opp-microvolt = <850000 850000 1000000>;
+ };
+
+ opp-300000000 {
+ opp-hz = /bits/ 64 <300000000>;
+ opp-microvolt = <850000 850000 1000000>;
+ };
+
+ opp-400000000 {
+ opp-hz = /bits/ 64 <400000000>;
+ opp-microvolt = <850000 850000 1000000>;
+ };
+
+ opp-600000000 {
+ opp-hz = /bits/ 64 <600000000>;
+ opp-microvolt = <900000 900000 1000000>;
+ };
+
+ opp-700000000 {
+ opp-hz = /bits/ 64 <700000000>;
+ opp-microvolt = <950000 950000 1000000>;
+ };
+ };
+};
+
+&cpu0 {
+ operating-points-v2 = <&cpu0_opp_table>;
+};
+
+&cpu1 {
+ operating-points-v2 = <&cpu0_opp_table>;
+};
+
+&cpu2 {
+ operating-points-v2 = <&cpu0_opp_table>;
+};
+
+&cpu3 {
+ operating-points-v2 = <&cpu0_opp_table>;
+};
+
+&gpu {
+ operating-points-v2 = <&gpu_opp_table>;
+};
