John, On Tue, May 13, 2008 at 1:19 PM, John Gilmore <[EMAIL PROTECTED]> wrote: > > The maximum number of multicast addresses per virtual device has been cut in > > half to ensure that the merged list can be accommodated by the hardware. > > If we allocated DRAM this way, no process could use more than 1/N of > the memory, where N is the number of processes. Surely this is > inappropriate. If eth0 is not in use, msh0 should be able to use all of > the hardware's multicast addresses. And vice verse, and all points in > between.
You have a point, but considering that: 1. the patch does fix a problem 2. N *is* 2 and will not grow beyond that 3. the size of the table is large (rtl8150 = 0, arlan, hermes, orinoco, zd1211 = 16, b44 = 32) and 4. the penalty for filling the table is minimal (see next point) I still recommend accepting the patch. I would certainly welcome a patch that implements the same table in a fully associative way. If you decide to work on that, keep in mind that with each address you also need to record the interface that it's bound to so you can eliminate it when that interface is brought down. > Signal an error to the caller when both devices combined > would exceed the hardware's capacity. No, turn on the ALLMULTI option that will configure the device to accept all multicast traffic. Cheers, Javier -- Javier Cardona cozybit Inc. _______________________________________________ Devel mailing list [email protected] http://lists.laptop.org/listinfo/devel
