Re: [PATCH v2 03/12] clk: divider: Introduce CLK_DIVIDER_EVEN_INTEGERS_NO_OFFSET flag

[Jie Luo <jie.luo@oss.qualcomm.com>]

On 8/11/2026 1:17 AM, Brian Masney wrote:
> Hi Luo,
> 
> On Thu, Aug 06, 2026 at 11:53:07PM -0700, Luo Jie wrote:
>> Some hardware dividers derive an even divisor directly from the raw
>> register value, i.e. divisor = 2 * val, with no +1 offset. This differs
>> from CLK_DIVIDER_EVEN_INTEGERS, which is one-based (divisor = 2 *
>> (val + 1)).
>>
>> Add CLK_DIVIDER_EVEN_INTEGERS_NO_OFFSET for this variant, updating
>> _get_maxdiv()/_get_div()/_get_val() accordingly. _is_valid_div() is
>> extended to also reject odd divisors, since _get_val()'s `div >> 1`
>> would otherwise silently decode them to the wrong even value (e.g. 3
>> -> 1 -> 2).
>>
>> To avoid regressing rate requests that previously computed an odd
>> divisor (which used to succeed with a silently wrong rate), round to
>> the nearest even divisor in _div_round_up(), _div_round_closest(), and
>> _next_div(), mirroring how CLK_DIVIDER_POWER_OF_TWO already rounds to
>> the nearest power of two.
>>
>> Also guard divider_ro_determine_rate() against a zero divisor, which
>> can occur from an unprogrammed/reset register, the same way
>> divider_recalc_rate() already does: WARN unless CLK_DIVIDER_ALLOW_ZERO
>> is set, and return -EINVAL.
>>
>> Signed-off-by: Luo Jie <jie.luo@oss.qualcomm.com>
> 
> One of the other patches in your series says:
> 
>     The NSS (network subsystem) clock is derived from the CMN PLL output
>     divided by 2 and then further divided by a configurable 6-bit divider.
> 
> Can you register a fixed-factor-clock parent with /2, then use
> CLK_DIVIDER_ONE_BASED? That should yield the same results and it may
> more accurately represent the hardware as well.
> 
> Brian
> 

Thanks for the suggestion. I will model this by registering a fixed-
factor clock with a /2 factor and using CLK_DIVIDER_ONE_BASED for the
subsequent 6-bit divider. This will preserve the current frequency
calculation while making the clock topology more accurately reflect the
hardware implementation.