[alsa-devel] [PATCH 0/3] Revised patch(es) for CLOCK_MONOTONIC_RAW timestamp

[Takashi Iwai]

At Wed, 16 Jul 2014 17:15:54 +0200,
Takashi Iwai wrote:
> 
> At Mon, 14 Jul 2014 18:19:26 +0200,
> Takashi Iwai wrote:
> > 
> > At Thu, 10 Jul 2014 15:04:15 +0200,
> > Takashi Iwai wrote:
> > > 
> > > Hi,
> > > 
> > > here is a series of revised kernel patches for adding
> > > CLOCK_MONOTONIC_RAW timestamp type.  In addition to Mark's original
> > > patch, the new field is added to sw_params.
> > > 
> > > The corresponding alsa-lib patches will follow after these.
> > 
> > FWIW, I merged the patches with acks to topic/monotonic branch of
> > sound git tree now.  It's merged to for-next branch.
> 
> And, now I'll revert the last part, the addition of sw_params again.
> 
> The addition to sw_params field would break the things when used with
> an old alsa-lib, because the old alsa-lib sets the field to zero and
> it overrides back to the gettimeofday from monotonic.
> 
> Unfortunately, there is no way in the kernel side whether the
> user-space calls in the old sw_params or not.  We should have a field
> or ioctl to set the protocol version the user-space talks, not only
> the kernel talks.
> 
> For alsa-lib, we'd just need two new functions that are decoupled from
> sw_params, e.g. snd_pcm_{get|set}_timestamp_type().

... thinking of this again, the less intrusive and future-proof change
is rather to add a new field indicating the protocol version from the
user-space side.  Then kernel evaluates the new fields only when a
valid proto number is set there.  This will help also any changes in
sw_params in future.  Patches will follow.


Takashi