Hello! > I also have an implementation of GSI routing on ARM, basically a rebase > of my old/first implementation of irqfd > (https://patches.linaro.org/32261/) based on irqchip gsi routing & qemu > part (https://lists.gnu.org/archive/html/qemu-devel/2014-07/msg01090.html).
I took a glance at it, and looks like it's already obsolete. We already have a convention of GSI number == SPI number. Kind of hardcoded default routing table which cannot be changed. It is used at least by GICv2m emulation. I think we should maintain backwards compatibility with it. I thought about something like: a) GSI < 8192 - correspond to SPIs and cannot be re-routed. b) GSI >= 8192 - correspond to MSI and need to be routed before use. During routing setup we could use either GSI with offset (starting from 8192), or raw number (starting from 0). In case of raw number we would have some complex structure of GSI field in KVM_CAP_IRQFD ioctl, similar to KVM_IRQ_LINE. Something like: bits: | 31 ... 24 | 23 ... 0 | field: | irq_type | irq_id | irq_type[0]: irq_id = SPI irq_type[3]: irq_id = GSI number routed to MSI Consequently, we have to implement only KVM_IRQ_ROUTING_MSI type and completely ignore KVM_IRQ_ROUTING_IRQCHIP. I hope i am clear enough... Kind regards, Pavel Fedin Expert Engineer Samsung Electronics Research center Russia -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
