[ECOS] semi-blocking serial read

[ECOS] semi-blocking serial read

[Grant Edwards]

My serial.c driver code has diverged somewhat from the standard
serial.c.  Most of the things I've done have either evolved in
parallel in the official one (like a non-blocking mode) or are
things that other people don't want.

One unique thing that mine does have that is both general and
(IMO) pretty useful is semi-blocking reads.

A semi-blocking read will block until _some_ amount of data is
available and return after transferring whatever is available.

This is primarily useful for UARTs with FIFOs such that receive
data comes in "chunks".  A semi-blocking read waits for the
next "chunk".  This allows you to process incoming data with
the lowest possible latency without doing a non-blocking read
in a busy-wait loop.

The change is pretty trivial.  In serial_read() you do
something like this:

     1	        while (size < *len) {
     2	            if (cbuf->nb > 0) {
     3	#ifdef CYGPKG_IO_SERIAL_FLOW_CONTROL
     4	                if ( (cbuf->nb <= cbuf->low_water) && 
     5	                     (chan->flow_desc.flags & CYG_SERIAL_FLOW_IN_THROTTLED) )
     6	                    restart_rx( chan, false );
     7	#endif
     8	                *buf++ = cbuf->data[cbuf->get];
     9	                if (++cbuf->get == cbuf->len) cbuf->get = 0;
    10	                cbuf->nb--;
    11	                size++;
    12	            } else {
    13	#ifdef CYGOPT_IO_SERIAL_SUPPORT_NONBLOCKING
--- 14                  if (!cbuf->blocking) {
+++ 14	                if (cbuf->nonblocking || (cbuf->semiblocking && size))) {
    15	                    *len = size;        // characters actually read
    16	                    res = -EAGAIN;
    17	                    break;
    18	                }
    19	#endif // CYGOPT_IO_SERIAL_SUPPORT_NONBLOCKING


The cbuf struct needs to change a little to accomodate the two
flags "nonblocking" and "semiblocking", but I think you see
what I mean.

I don't use serial.c and probably won't get around to
submitting a patch for some time, but I thought I'd toss the
idea out in case anybody else could use something like it.

-- 
Grant Edwards
grante@visi.com

Re: [ECOS] semi-blocking serial read

[David Airlie]

I thought about doing this before.. and never got around to it .. god this
would save me a lot of extra crap I've added to my code .. loads of tslp
and stuff waiting for data...

I might go code up a patch for this in the next couple of weeks, when I
get the chance to sit down with my eCos devel system.. that is unless
someone beats me to it ...

Dave.


On Wed, 15 Aug 2001, Grant Edwards wrote:

> 
> My serial.c driver code has diverged somewhat from the standard
> serial.c.  Most of the things I've done have either evolved in
> parallel in the official one (like a non-blocking mode) or are
> things that other people don't want.
> 
> One unique thing that mine does have that is both general and
> (IMO) pretty useful is semi-blocking reads.
> 
> A semi-blocking read will block until _some_ amount of data is
> available and return after transferring whatever is available.
> 
> This is primarily useful for UARTs with FIFOs such that receive
> data comes in "chunks".  A semi-blocking read waits for the
> next "chunk".  This allows you to process incoming data with
> the lowest possible latency without doing a non-blocking read
> in a busy-wait loop.
> 
> The change is pretty trivial.  In serial_read() you do
> something like this:
> 
>      1	        while (size < *len) {
>      2	            if (cbuf->nb > 0) {
>      3	#ifdef CYGPKG_IO_SERIAL_FLOW_CONTROL
>      4	                if ( (cbuf->nb <= cbuf->low_water) && 
>      5	                     (chan->flow_desc.flags & CYG_SERIAL_FLOW_IN_THROTTLED) )
>      6	                    restart_rx( chan, false );
>      7	#endif
>      8	                *buf++ = cbuf->data[cbuf->get];
>      9	                if (++cbuf->get == cbuf->len) cbuf->get = 0;
>     10	                cbuf->nb--;
>     11	                size++;
>     12	            } else {
>     13	#ifdef CYGOPT_IO_SERIAL_SUPPORT_NONBLOCKING
> --- 14                  if (!cbuf->blocking) {
> +++ 14	                if (cbuf->nonblocking || (cbuf->semiblocking && size))) {
>     15	                    *len = size;        // characters actually read
>     16	                    res = -EAGAIN;
>     17	                    break;
>     18	                }
>     19	#endif // CYGOPT_IO_SERIAL_SUPPORT_NONBLOCKING
> 
> 
> The cbuf struct needs to change a little to accomodate the two
> flags "nonblocking" and "semiblocking", but I think you see
> what I mean.
> 
> I don't use serial.c and probably won't get around to
> submitting a patch for some time, but I thought I'd toss the
> idea out in case anybody else could use something like it.
> 
> 

-- 
      David Airlie, Software Engineer, Parthus Technologies plc.,
       Mary Rosse Centre, National Tech Park, Limerick, Ireland.
   t: +353-61-508116 / f: +353-61-508101 / David.Airlie@parthus.com

Re: [ECOS] semi-blocking serial read

[Jonathan Larmour]

David Airlie wrote:
> 
> I thought about doing this before.. and never got around to it .. god this
> would save me a lot of extra crap I've added to my code .. loads of tslp
> and stuff waiting for data...
> 
> I might go code up a patch for this in the next couple of weeks, when I
> get the chance to sit down with my eCos devel system.. that is unless
> someone beats me to it ...

If you do, it would be nice if it was overall controlled by a separate CDL
option, like the non-blocking stuff. After all, it's only useful on systems
that do receive stuff in chunks.

Jifl
-- 
Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine