On Wed, 08 Jul 2015 19:10:32 +0200, Pierre-Louis Bossart wrote:
This is only enabled when the NO_REWIND hardware flag is used, so that the low-level driver/hardware to opportunistically pre-fetch data.
FIXME: should we rely on .ack for this? Signed-off-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com
Hmm, OK, so the forward is allowed but with workarounds... But then why rewind won't work in a similar way? DSP might be able to cancel some of inflight data.
Nope, this is explicitly not supported, so unfortunately if we want to optimize for power and let hardware fetch data when it's most appropriate rewinds need to be disabled.
In other words, I see no reason to strict notify callback only for no_rewinds. This is an optional ops in anyway.
It's fine to remove the check. I added this based on internal review comments but I agree with your point.
OK, then let's treat the NO_REWIND and new ops individually.
Also, I find the name "notify" a bit too ambiguous. In this case, it's notifying the applptr change. So, a name related with the function would be more understandable.
The first open I had was to know if we could use .ack for this? if a different callback is needed, we can use 'appl_ptr_update' instead of 'notify'
As there are no many users of ack callback, I don't mind to reuse it. But then we need to extend the function to receive a new argument indicating the type of ack, right?
thanks,
Takashi