Hello Krzysztof,

On 07/10/2015 01:30 AM, Krzysztof Kozlowski wrote:

On 10.07.2015 00:43, Bartlomiej Zolnierkiewicz wrote:

quoted

With the addition of the new Samsung specific cpu-clock type, the
arm clock can be represented as a cpu-clock type. Add the CPU clock
configuration data and instantiate the CPU clock type for Exynos4x12.

Based on the earlier work by Thomas Abraham.

Cc: Tomasz Figa <redacted>
Cc: Michael Turquette <mturquette@baylibre.com>
Cc: Javier Martinez Canillas <javier@dowhile0.org>
Cc: Thomas Abraham <redacted>
Signed-off-by: Bartlomiej Zolnierkiewicz <redacted>
---
 drivers/clk/samsung/clk-exynos4.c | 50 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 50 insertions(+)

diff --git a/drivers/clk/samsung/clk-exynos4.c b/drivers/clk/samsung/clk-exynos4.c
index cae2c048..3071260 100644
--- a/drivers/clk/samsung/clk-exynos4.c
+++ b/drivers/clk/samsung/clk-exynos4.c

@@ -1396,6 +1396,45 @@ static const struct exynos_cpuclk_cfg_data e4210_armclk_d[] __initconst = {
 	{  0 },
 };
 
+static const struct exynos_cpuclk_cfg_data e4212_armclk_d[] __initconst = {
+	{ 1500000, E4210_CPU_DIV0(2, 1, 6, 0, 7, 3), E4210_CPU_DIV1(2, 6), },
+	{ 1400000, E4210_CPU_DIV0(2, 1, 6, 0, 7, 3), E4210_CPU_DIV1(2, 6), },
+	{ 1300000, E4210_CPU_DIV0(2, 1, 5, 0, 7, 3), E4210_CPU_DIV1(2, 5), },
+	{ 1200000, E4210_CPU_DIV0(2, 1, 5, 0, 7, 3), E4210_CPU_DIV1(2, 5), },
+	{ 1100000, E4210_CPU_DIV0(2, 1, 4, 0, 6, 3), E4210_CPU_DIV1(2, 4), },
+	{ 1000000, E4210_CPU_DIV0(1, 1, 4, 0, 5, 2), E4210_CPU_DIV1(2, 4), },
+	{  900000, E4210_CPU_DIV0(1, 1, 3, 0, 5, 2), E4210_CPU_DIV1(2, 3), },
+	{  800000, E4210_CPU_DIV0(1, 1, 3, 0, 5, 2), E4210_CPU_DIV1(2, 3), },
+	{  700000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4210_CPU_DIV1(2, 3), },
+	{  600000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4210_CPU_DIV1(2, 3), },
+	{  500000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4210_CPU_DIV1(2, 3), },
+	{  400000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4210_CPU_DIV1(2, 3), },
+	{  300000, E4210_CPU_DIV0(1, 1, 2, 0, 4, 2), E4210_CPU_DIV1(2, 3), },
+	{  200000, E4210_CPU_DIV0(1, 1, 1, 0, 3, 1), E4210_CPU_DIV1(2, 3), },
+	{  0 },
+};
+
+#define E4412_CPU_DIV1(cores, hpm, copy)				\
+		(((cores) << 8) | ((hpm) << 4) | ((copy) << 0))
+
+static const struct exynos_cpuclk_cfg_data e4412_armclk_d[] __initconst = {
+	{ 1500000, E4210_CPU_DIV0(2, 1, 6, 0, 7, 3), E4412_CPU_DIV1(7, 0, 6), },
+	{ 1400000, E4210_CPU_DIV0(2, 1, 6, 0, 7, 3), E4412_CPU_DIV1(6, 0, 6), },
+	{ 1300000, E4210_CPU_DIV0(2, 1, 5, 0, 7, 3), E4412_CPU_DIV1(6, 0, 5), },
+	{ 1200000, E4210_CPU_DIV0(2, 1, 5, 0, 7, 3), E4412_CPU_DIV1(5, 0, 5), },
+	{ 1100000, E4210_CPU_DIV0(2, 1, 4, 0, 6, 3), E4412_CPU_DIV1(5, 0, 4), },
+	{ 1000000, E4210_CPU_DIV0(1, 1, 4, 0, 5, 2), E4412_CPU_DIV1(4, 0, 4), },
+	{  900000, E4210_CPU_DIV0(1, 1, 3, 0, 5, 2), E4412_CPU_DIV1(4, 0, 3), },
+	{  800000, E4210_CPU_DIV0(1, 1, 3, 0, 5, 2), E4412_CPU_DIV1(3, 0, 3), },
+	{  700000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4412_CPU_DIV1(3, 0, 3), },
+	{  600000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4412_CPU_DIV1(2, 0, 3), },
+	{  500000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4412_CPU_DIV1(2, 0, 3), },
+	{  400000, E4210_CPU_DIV0(1, 1, 3, 0, 4, 2), E4412_CPU_DIV1(1, 0, 3), },
+	{  300000, E4210_CPU_DIV0(1, 1, 2, 0, 4, 2), E4412_CPU_DIV1(1, 0, 3), },
+	{  200000, E4210_CPU_DIV0(1, 1, 1, 0, 3, 1), E4412_CPU_DIV1(0, 0, 3), },
+	{  0 },
+};

Numbers look fine!

quoted

+
 /* register exynos4 clocks */
 static void __init exynos4_clk_init(struct device_node *np,
 				    enum exynos4_soc soc)

@@ -1489,6 +1528,17 @@ static void __init exynos4_clk_init(struct device_node *np,
 		samsung_clk_register_fixed_factor(ctx,
 			exynos4x12_fixed_factor_clks,
 			ARRAY_SIZE(exynos4x12_fixed_factor_clks));
+		if (of_machine_is_compatible("samsung,exynos4412")) {

The driver uses here enum exynos4_soc to differentiate between SoC
(unless I missed some changes). This of_machine_is_compatible() makes
sense but introduces inconsistency. I would prefer sticking to one
convention: always enum or switch everything (before this patch) to
of_compatible.

When reviewing this patch I also ran into the same thing because as you
said, it's not consistent. But digging a little bit I found that is not
that easy since the two are not checking exactly the same.

The enum is to differentiate between "samsung,exynos4412-clock" and
"samsung,exynos4210-clock" while the of_machine_is_compatible() is for
"samsung,exynos4412" and "samsung,exynos4212".

The problem is that both exynos4412 and exynos4212 use the same
"samsung,exynos4412-clock" compatible for their clock controller nodes.
But there are differences so it would had been better to also have a
"samsung,exynos4212-clock" to avoid the of_machine_is_compatible() but
that is not possible anymore without breaking DT backward compatibility.

On the other hand, if of_machine_is_compatible() is used for everything,
then there is no point anymore to have both "samsung,exynos4412-clock"
and "samsung,exynos4210-clock". A single "samsung,exynos4-clock" plus
checking the SoC would had been enough.

That's why I thought that Bart's approach was sensible although is true
that the of_compatible() check can be moved to exynos4412_clk_init()
and the enum be extended so at least exynos4_clk_init() is consistent.

Best regards,
Krzysztof

Best regards,

-- 
Javier Martinez Canillas
Open Source Group
Samsung Research America