Re: SB16: 2.4.18 lockup on odd-numbered 16bit sound input

[Ingo Oeser <ingo.oeser@informatik.tu-chemnitz.de>]

On Sat, May 25, 2002 at 03:26:44PM +0200, Andreas Mohr wrote:
> I think the main question that this revolves around is what to do with
> odd-numbered/misaligned write() input in case of multi-byte audio formats.
> Should the kernel
> 1) discard odd remainders
> 2) tell the user program by returning incomplete counts that it discarded it
> 3) buffer unprocessable remainders for use in the next write()
> ?

2) Was the right answer for my own driver set, because it will
   not loose data and is a common situation the user space app is
   supposed to handle properly. 

The only arguable case is if we use a count of 0 after 
calculating (count & (align-1)), because we would basically
return EOF here instead of a short read (in  the read case).

But it can be worked around by disallowing seeks or contraining the
seeks ;-)

PS: I know, we are talking about (short) writes here, but the
   topic needs insight and I just handled these problems recently.

Regards

Ingo Oeser
-- 
Science is what we can tell a computer. Art is everything else. --- D.E.Knuth