ARM: dts: exynos: Add timer node for exynos5420

This patch adds timer node for exynos542x/5800.
Generic Timer Architecture is needed by KVM.

Signed-off-by Adel Belhouane <[email protected]>

---
 arch/arm/boot/dts/exynos5420.dtsi | 10 ++++++++++
 1 file changed, 10 insertions(+)

arch/arm/boot/dts/exynos5420.dtsi +3 -0

index 48a0a55314f5..4663006d15ab 100644
@@ -1140,6 +1140,16 @@
 		power-domains = <&disp_pd>;
 		#iommu-cells = <0>;
 	};
+
+	timer {
+		compatible = "arm,cortex-a15-timer",

Hi Adel,

Quoted 1 line · Adel Belhouane

This patch adds timer node for exynos542x/5800.

Did you test this arch_timer after deleting the MCT Device Tree node from dts file?

As I knew, the generic arch time uses same H/W with MCT device in Exynos SoC. So, If we remove the MCT device-tree node from dts, the arch timer is not working. Just, I tested it on Exynos3250 SoC.

On ODROID XU3, after having all the Exynos5422 cores detected as started in HYP mode with a patch[1], KVM still won't work because it needs the Generic Timer Architecture. It's defined in other exynos5 models (5250, 5440) DTs, but not the 542x/5800, so I just added it.

Actually this timer was defined in ODROID Hardkernel's repository [2], so I took its clock-frequency value too (which isn't provided by firmware/U-Boot).

KVM now works on ODROID XU3 with kernel v4.5 plus these changes.

regards,
Adel Belhouane.

[1] from Brian Kim:
https://gist.github.com/bkrepo/6f4eea33be133ad306e3

[2]
https://github.com/hardkernel/linux/blob/odroidxu3-3.10.y/arch/arm/boot/dts/exynos5422-odroidxu3.dts

Please look at this thread:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/359806.html

Unfortunately I do not have any more details about this.

Quoted 4 lines · Krzysztof Kozlowski

Please look at this thread:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/359806.html

Unfortunately I do not have any more details about this.

So I understand MCT and Arch Timer use the same hardware timer and there were issues when the MCT interface reset it at boot and during resume and the Arch Timer consumers didn't expect it. In this thread there were links [1] to a patch to have the MCT not writing the timer anymore, thus allowing the two interfaces to co-exist.

This patch was applied in commit 1d80415db64b. So no problem is to be expected when using both interfaces at the same time.

I can see exynos5250 got its Arch Timer definition for KVM:

- From the description of commit 2b7da98870bc
ARM: dts: Add architected timer nodes for exynos5250 [...]
This is required by KVM.

So is there something else blocking adding this definition for 542x/5800 considering its effect is enabling KVM ?

Doug Anderson:
According to Samsung, there are issues where using cp15 could sometimes return the wrong value, especially if you happen to read it while on an A7 instead of an A15.

Kukjin Kim:
One more thing, if you change it for your own private platform, it's > up to you but if not, i.e., for mass product, please don't change it. Because it's not guaranteed even you haven't seen its failure...

Quoted 4 lines · Adel Belhouane

Quoted 3 lines · Krzysztof Kozlowski

Please look at this thread:
http://lists.infradead.org/pipermail/linux-arm-kernel/2015-July/359806.html

Quoted 8 lines

Quoted 1 line

Unfortunately I do not have any more details about this.

This patch was applied in commit 1d80415db64b. So no problem is to be expected when using both interfaces at the same time.

So is there something else blocking adding this definition for 542x/5800 considering its effect is enabling KVM ?

Yeah, this is blocking:

Quoted 4 lines

Doug Anderson:
According to Samsung, there are issues where using cp15 could sometimes return the wrong value, especially if you happen to read it while on an A7 instead of an A15.

Quoted 4 lines

Kukjin Kim:
One more thing, if you change it for your own private platform, it's > up to you but if not, i.e., for mass product, please don't change it. Because it's not guaranteed even you haven't seen its failure...

Best regards,
Krzysztof