Am Freitag, 7. Juni 2013, 13:46:32 schrieb Linus Walleij:

On Thu, Jun 6, 2013 at 9:08 PM, Heiko St?bner [off-list ref] wrote:

quoted

There exist platforms, namely at least all Rockchip Cortex-A9 based ones,
that don't use the paradigm of reading-changing-writing the register
contents, but instead only write the changes to the register with a mask
that indicates the changed bits.

This patch adds flags and code to support the case where the lower 16 bit
of hold the information and the upper 16 bit are used as mask to
indicate the written changes.

As hardware-specific flags should not be added to the common clk flags,
the flags are added to gate, mux and divider clocks individually.

Signed-off-by: Heiko Stuebner <heiko@sntech.de>

(...)

quoted

+       if ((clk_gate_flags & CLK_GATE_MASK_UPPER_HALF) && bit_idx > 15)
{ +               pr_err("%s: bit_idx %d invalid\n", __func__, bit_idx);
+               return ERR_PTR(-EINVAL);
+       }

Now this looks *EXTREMELY* familiar to a patch just sent by Haojian
for HiSilicon.

"[PATCH v2 3/6] clk: divider: add CLK_DIVIDER_HIWORD_MASK flag"
http://marc.info/?l=linux-arm-kernel&m=137035873916777&w=2

What kind of coincidence is this? Are Rockchip and HiSilicon using
the same silicon IP or are they of a common origin? (It is a small
world after all.)

Now this is really interesting :-) and the handling really seems to follow the 
same pattern.

I think you two guys need to read each others patch sets closely
here. I'd like Haojian to look at Heiko's patches and Heiko to look
at Haojian's patches, just to make sure you're not actually writing
two drivers for the same hardware in the end.

I'll take a look


Heiko