Re: [ECOS] Question on Posix Select function

[Andrew Lunn <andrew@lunn.ch>]

On Mon, Jun 19, 2006 at 11:17:42AM +0200, e.coullien@faiveley.com wrote:
> 
> 

> We are working on a package to complete the eCos Intrument package
> for communicating with a PC software through Ethernet. The goal is
> to record in real time the Instrument informations from the board to
> the PC and to watch a graph to show tasks and other instrument
> traces.

Interesting. How do you solve the obvious race condition that using
the network to send the events generate events?

Ideally you should use the RedBoot network stack for this. 

> We began to graph a module with a SELECT function but we have some
> difficulties to understand what we see on the Graph.  In a task,
> before sending, we call the select function. In fact the task which
> calls the SELECT function is switching on even if the driver is not
> ready to send.  So the graph shows that this task is active very
> often instead to sleep until the driver is ready to send.  Do
> someone know how to explain this phenomenon and how does the Posix
> select function really work ?

There is pritty good comments in the code. Have you read them:

// A flag is used to block a thread until data from the device is
// available. This prevents all threads from waking up at the same
// time and polling for changes.  Each thread is allocated a flag bit
// via the SELECT_WAIT_FLAG_GET() macro when the thread registers for
// selection via cyg_selrecord (). The flag is stored in the driver's
// select info block. Only those threads specified via the flags in
// the select info are woken up by cyg_selwakeup ().  If there are
// more than 32 threads in the system, then there is a chance that
// cyg_selwakeup () may wake up more than one thread. Each thread then
// polls for changes.

So you probably have the same bit set as some other thread using the network.

  Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss