--- Mark Vojkovich <[EMAIL PROTECTED]> a �crit : > > > On Tue, 23 Sep 2003, [iso-8859-1] emmanuel ALLAUD > wrote: > > > > > > I'm assuming futexes are some interprocess > mutex > > > mechanism? > > > If so, I don't see how this helps. > > > > > > > Yes it is. Actually I am not familiar with them > > either, but if you had a way to be able to wake up > > your waiting process ASAP when eg DMA is finished, > > instead of waiting that the OS scheduler gives you > the > > CPU back (which can be very long), that would be > good. > > Problem is that you must in the first place be > able to > > have a way (here interrupt I guess) that tells you > > that the work is finished, which means something > in > > kernel (in case of DMA do you have an interrupt > driven > > way to know if it's completed?). > > The "nv" driver is completely in user-space and > requires > no kernel support. It should remain so due to > portability > concerns. Besides, I don't really want to know when > DMA is > finished. I want to know when there's enough free > room in > the DMA buffer so the CPU can continue putting data > into it > rather than polling for it. Yielding in my poll > loop has > been a good solution in other cases.
Yes you're right, and then I am not sure this can be done (I don't think it is really possible to have a report from DMA saying that the source buffer has "enough free space"). The problem with yielding is that you can have interactivity problem if the computer is loaded enough. At least that's what I have understood will happen with 2.6 linux kernels. The problem is to be able to tell the kernel : OK I know I have to wait so I give the CPU back, but I am an interactive process and I don't want my CPU back in 1 hour ;-) Perhaps there should be a "give my remaining slice time to someone else and then give me back the CPU" scheduler hint? Bye Manu ___________________________________________________________ Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en fran�ais ! Yahoo! Mail : http://fr.mail.yahoo.com _______________________________________________ Devel mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/devel
