On Mon, Jan 24, 2000 at 10:55:00AM -0500, Paul Koning wrote: > I was doing some code reading in V3.0Pre1 and started wondering about > the intended usage (as far as multiple accessors goes) of rt fifos. > The documentation is not particularly clear. > > Several possible answers come to mind: > > 1. Single reader, single writer per fifo > 2. Single reader, multiple writers per fifo > 3. Multiple readers, multiple writers per fifo > > Right now the code looks like major overkill for (1); it appears to > intend to support (3). However, as far as I can tell this works > correctly *only* if the set of readers and the set of writers each do > NOT contain a mix of Linux and RTLinux threads. The reason is that > the Linux fifo I/O functions interlock RTLinux when they pick up the > pointers, and then again when they update the pointers, but not in > between. So it looks like an RTLinux thread and a Linux thread could > both read the same data, or both write to the same region of an > rtfifo. > > I would tend to argue that the ability to mix things like that isn't > particularly interesting, but if it's not supposed to be done the > documentation should say that. Good point. > A related point: the rtf_get and rtf_put operations lock out > interrupts for the full duration of the transfer. You need that to > get mode (3). I expect there are plenty of fifos that don't need > this, they only need (1). You pay an interrupt latency penalty you > don't need in that case. How about another set of calls > (rtf_get_single?) or a mode setting to select (1) and avoid most or > all of the interrupt disabling? The question is whether the time spent during cli is significant and I think it is not, but perhaps I have not seen something. > Finally, I think I could use some additional rtfifo services, > specifically (a) a way to query how much is currently in the fifo and > how much space is left, so I can avoid doing an rtf_get or rtf_put > that will do a partial transfer, (b) a "flush the fifo" operation. > > Reason for (a) is: if you transfer structured data via a fifo (similar > to the "frank" sample application) then you don't want to do an > rtf_get and get less than sizeof (your_struct) coming back. One useful mode would be a "mailbox" mode where each input output will be in multiples of message-size. That would avoid the extra querying in the user of the fifo. IF you can add that efficiently, I'd be happy to take the patch. -- --------------------------------------------------------- Victor Yodaiken FSMLabs: www.fsmlabs.com www.rtlinux.com FSMLabs is a servicemark and a service of VJY Associates L.L.C, New Mexico. --- [rtl] --- To unsubscribe: echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED] ---- For more information on Real-Time Linux see: http://www.rtlinux.org/~rtlinux/
