On Fri, 4 Mar 2011, Micha Nelissen wrote: > Jesse Barnes wrote: > > On Sun, 13 Feb 2011 21:25:21 +0100 > > Micha Nelissen <micha@xxxxxxxxxxxxxx> wrote: > > > > > Patch is based on earlier patch from Matthew Wilcox. > > > > I like this before when Matthew was working on it too > > If I understood Thomas' concerns well, it sounds like this feature is > inherently incompatible with big SMP systems. Therefore it has no chance of > ever being accepted into mainline in whatever form? First of all MSI is replaced by MSI-X which does not have the shortcomings of MSI vs. the vector allocation. Though if we can come up with a scheme which does not create nested loops over possibly hundreds of CPUs with interrupts disabled and global locks held, I have no objections. One possible solution would be to reserve a block of vectors on all cpus at boot time via a command line option for MSI block allocations. That would simply use a bitmap protected by a mutex to lookup a free vector space. That would avoid the whole loop issue and work for most of the systems which need to deal with such MSI multi vector devices. Warning: Just an idea :) Thanks, tglx -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
