Re: [uml-devel] UMLWin32 - Network driver folks ?

[Nathan L. <lut...@li...>]

On Thu, Aug 22, 2002 at 02:31:22PM -0400, Chandan Kudige wrote:
> This basically involves having a polling loop similar to the=20
> write_sigio_workaround(), but instead of poll() it will use the win32
> equivalent (WaitForMultipleObjects). This is a logical first step in=20
> converting over to pure Win32 eventually.
>
> This polling loop should trigger the uml interrupt handler. We dont want=
=20
> to call the handler in the context of the polling thread since there is n=
o=20
> SMP protection on this code and can cause race conditions. My hack for th=
e=20
> console driver achieves it by triggering an event which causes the idle=
=20
> loop to call sigio_handler() the next time it is scheduled.

Actually, I took out the separate polling thread (that incremented
sigio_hack_trigger) and it still works fine.  The way it worked before=20
was:

Polling thread:
	while(1)
	{
		if( data is available on a read fd )
			++sigio_hack_trigger;
	}

Idle loop:
	while(1)
	{
		do scheduling stuff
		sleep a little
		if( sigio_hack_trigger > 0 )
		{
			--sigio_hack_trigger;
			sigio_handler()
		}
	}

I couldn't figure out how that is different from this:
	while(1)
	{
		do scheduling stuff
		sleep a little
		sigio_handler()
	}

I changed the code to the loop above and removed the polling thread, and=20
everything seems to work correctly.  The only difference is that=20
sigio_handler is called every time through the loop instead of only when=20
data is waiting, but since all it does is call poll() and exit when it=20
finds no data waiting, I don't see the difference.

Of course, calling sigio_handler() from idle loop means that no interrupts=
=20
are processed if a process is RUNNING...  -Nathan

--=20
Help reduce spam!  PGP-sign your email.