Re: [PATCH 7/8] OMAP: omap_device: warn about expected latencies only if non-zero

[Kevin Hilman <khilman@deeprootsystems.com>]

Paul Walmsley <paul@pwsan.com> writes:

> Hi,
>
> On Tue, 17 Nov 2009, Kevin Hilman wrote:
>
>> When checking measured [de]activate latencies against expected
>> latencies, only print warning if driver has provided a non-zero
>> latency.
>> 
>> This allows drivers to set their [de]activate latencies to zero when
>> they are not known, or are don't care, and the omap_device core will
>> not complain about unexpected latencies.
>
> I'm concerned that this will effectively mean that no driver maintainers 
> will bother to measure these latencies.   
>
> These are necessary for a reasonable implementation of
> omap_pm_set_max_dev_wakeup_lat().  They should not be difficult to
> measure: boot at the lowest-performance OPPs with the latencies
> uninitialized; then these warnings should indicate numbers that can
> be plugged back into the omap_device structure.  But perhaps I am
> misunderstanding the point you are making?

Part of the issue is being able to do a quick conversion, without
needing to measure latencies (which I'm guilty of.)  

The other part was that I was thinking (the RFC part) was that it
might be useful for performance reasons to not do do any measuring at
all.  Of course, this would preclude a useful max_dev_wakeup_lat() but
we could warn about that.

But, for now I agree. I think getting these latencies in place is more
important than having an optimization for avoiding measurement for now.

Kevin