Hi Marek,
On 17.08.21 14:50, Marek Behún wrote:
+#elif defined(CONFIG_ARMADA_3700)
+/* SAR values for Armada 3700 */
+#define MVEBU_TEST_PIN_LATCH_N MVEBU_REGISTER(0x13808)
+#define MVEBU_XTAL_MODE_MASK BIT(9)
+#define CONFIG_SYS_REF_CLK ((readl(MVEBU_TEST_PIN_LATCH_N) &
MVEBU_XTAL_MODE_MASK) ? \
+ 40000000 : 25000000)
Stefan,
we discussed this with Pali yesterday. For now, I am okay with it, but
I consider this bad design.
I have to admit, that I'm also not 100% happy. But I didn't want to
burden Pali with more cleanup here.
This either should be done via the clk API or, if we insist on keeping
it simple, via a static inline function, with caching via a static
variable, so that the register needs to be read only once (at least for
each compilation unit).
Regarding the clk API kernel has driver for armada-37xx-xtal. But I am
thinking about expanding the clk-fixed-mmio driver so that it would be
more generic and could be used for this.
What do you think?
I would very much welcome such a change / enhancement. This legacy type
input clock handling in MVEBU (and some other platforms) definitely
needs the get changed to "correct CLK API" handling.
Thanks,
Stefan