Re: suggest rvfree in mite_buf_alloc

On Mon, 11 Nov 2002, David Schleef wrote:

> You should note that rvmalloc is not exported by comedi.o.  This is
> on purpose.
> 
> > Could you explain what do you mean by "separate allocation and reserve
> > staged" ?
> 
> DMA and vmalloc() do not coexist well on all architectures, so it
> would be wise to have a model where vmalloc() can be easily changed
> with something else.

I support this. So as you are making a more general interface why not make
a buffer supplied by the (kernel) application one of the options?

I am using one big mbuff, with lots of data stractures and dedicated
window for DMA, which I access from both kernel and user space.

I understand it may be nonportable, but hey, you may warn people about it.

The comedi device is locked by the kernel module, but in this way user
space processes may still see the data.

You see, I need a model which allows random access to the data by multiple
readers. For example RT thread processes selected data, and user space
works as digital oscilloscope looking at this or some other time window.

Best regards,
--
Tomasz Motylewski

Received on 2002-11-17Z14:16:05