On Fri, Nov 14, 2008 at 10:41:54AM -0700, Matthew Wilcox wrote: > > 2) File name is index# into MSIX table and contents is IRQ#. > > Yes, this was what I proposed, and Greg agreed to. ... > > So the driver needs a quick way to reverse map the current_CPU->MSIX_index. > > No, the driver needs to map current CPU -> best queue. The interrupt > number belongs to the queue and doesn't need to form part of the > driver's decision. Agreed. I associate each MSIX index entry with a queue (true for some devices) and mean it the same way you do with "belongs to the queue". I agree the IRQ number is irrelevant to the decision of "best queue". But the CPU targetted by that MSI is entirely relevant. It's the only way to minimize cacheline bouncing between the CPU starting a transaction and the (hopefully same) CPU handling the completion. thanks, grant -- 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
