Re: [PATCH net-next v6 1/3] dpll: zl3073x: scale poll interval proportionally to timeout

[Ivan Vecera]

Sashiko findings with comments:

> Since zl3073x_poll_zero_u8() is the single shared polling helper, 
does this
> also change the poll granularity for all the pre-existing callers [...]
> sleep_us becomes 500 us, 1000 us, 1000 us, 600 us and 1000 us 
respectively

Yes, the change affects all callers. Testing showed that the
existing callers typically completed in low hundreds of sleep cycles
with the fixed 10 us interval. With the scaled interval, the cycle
count dropped to single digits — a much better fit for the actual
hardware response times.

> Does the flash path also get slower? [...]
> over the tens of thousands of words in a firmware or utility image
> could add a noticeable amount of time to a devlink flash.

No. Firmware flash operations were tested with no observable
regression. The overall flash time remained the same because the
reduced number of sleep cycles compensates for the longer individual
sleep intervals.

> Would clamping the derived interval [...] reach the stated goal [...]
> without changing the existing 25-50 ms call sites?

Given the test results above, clamping is not necessary.

> Reading this, one would conclude the existing 25-50 ms call sites 
keep their
> old behaviour [...]  Could the changelog mention the effect on the 
existing
> [...] polls?

Fair point. Will update the changelog to mention that all callers are
affected.

> Is "approximately 50 iterations" accurate?

Will reword to avoid the specific iteration count claim.

Thanks,
Ivan