Re: Mac mini sound woes

[Takashi Iwai <tiwai@suse.de>]

At Tue, 29 Mar 2005 11:22:07 +0200,
Marcin Dalecki wrote:
> 
> 
> On 2005-03-29, at 10:18, Benjamin Herrenschmidt wrote:
> >
> > Well, we are claiming _and_ obviously proposing a solution ;)
> 
> I beg to differ.
> 
> >> 1. Where do you have true "real-time" under linux? Kernel or user 
> >> space?
> >
> > That's bullshit.
> 
> Wait a moment...
> 
> > you don't need "true" real time for the mixing/volume
> > processing in most cases.
> 
> Yeah! Give me a break: *Most cases*. Playing sound and video is
> paramount for requiring asserted timing. Isn't that a property
> RT is defined by?

No, still you don't need "true" real-time OS.
(More exactly, do we have a "true" RT OS? :)

> > I've been doing sound drivers on various
> > platforms who don't have anything that look like true realtime neither
> > and beleive, it works. Besides, if doing it linux shows latency
> > problems, let's just fix them.
> 
> Perhaps as an exercise you could fix the jerky mouse movements on
> Linux - too? I would be very glad to see the mouse, which has truly 
> modest
> RT requirements, to start to behave the way it's supposed to do.
> And yes I expect it to still move smoothly when doing "make -j100 
> world".

On the contrary, doing the soft-mixing/volume in kernel is the source
of latency when schedule isn't done properly without preemption.


> >> 2. Where would you put the firmware for an DSP? Far away or as near to
> >> hardware as possible?
> >
> > Yes. This point is moot. The firmware is somewhere in your filesystem
> > and obtained with the request_firmware() interface, that has nothing to
> > do in the kernel. If it's really small, it might be ok to stuff it in
> > the kernel. But anyway, this point is totally unrelated to the 
> > statement
> > you are replying to.
> 
> No. You didn't get it. I'm taking the view that mixing sound is simply
> a task you would typically love to make a DSP firmware do.
> However providing a DSP for sound processing at 44kHZ on the same
> PCB as an 1GHZ CPU is a ridiculous waste of resources. Thus most 
> hardware
> vendors out there decided to use the main CPU instead. Thus the 
> "firmware"
> is simply running on the main CPU now. Now where should it go? I'm 
> convinced
> that its better to put it near the hardware in the whole stack.

I don't understand this logic...

>  You 
> think
> it's best to put it far away and to invent artificial synchronization
> problems between different applications putting data down to the
> same hardware device.
> 
> >> 3. How do you synchronize devices on non real time system?
> >
> > I'm not sure I understand what you mean here. I suppose it's about
> > propagation of clock sources, which is traditionally done in the slave
> > way; that is the producer (whatever it is, mixer, app, ...) is "sucked"
> > by the lowest level at a given rate, the sample count beeing the
> > timestamp, variable sample size having other means (and less precise of
> > course) to synchronize.
> 
> No I'm simply taking the view that most of the time it's not only a 
> single
> application which will feed the sound output. And quite frequently you 
> have
> to synchronize even with video output.

Hmm, how is this related to the topic whether a job is done in user or
kernel space...?


> >> 4. Why the hell do we have whole network protocols inside the kernel?
> >> Couldn't those
> >> be perfectly handled in user space? Or maybe there are good reasons?
> >
> > Network protocol do very few computation on the data in the packets
> > (well, except for IPsec for security reasons mostly) but this is a gain
> > totally unrelated. Like comparing apples and pears.
> 
> No it's not that far away. The same constraints which did lead most 
> people
> to move TCP in to the kernel basically apply to sound output.
> It's just a data stream those days after all.

It depends on the efficiency, too.  And, if you think of efficiency,
user-space has a big gain that it can use SIMD operations.


> >> 5. Should a driver just basically map the hardware to the user space 
> >> or
> >> shouldn't
> >> it perhaps provide abstraction from the actual hardware implementing 
> >> it?
> >
> > This is in no way incompatible with having the mixing and volume 
> > control
> > in userspace. It's actually quite a good idea to have a userland 
> > library
> > that isolates you from the low level "raw" kernel intefaces of the
> > driver, and in the case of sound, provides you with the means to setup
> > codec chains, mixing components, etc...
> 
> It is not. At least every other OS out there with significant care for
> sound did came to a different conclusion.

ALSA provides the "driver" feature in user-space because it's more
flexible, more efficient and safer than doing in kernel.  It's
transparent from apps perspective.  It really doesn't matter whether
it's in kernel or user space.

I think your misunderstanding is that you beliieve user-space can't do
RT.  It's wrong.  See JACK (jackit.sf.net), for example.


Takashi