In an interrupt callback function, the paramter intr_handle which is registered at rte_intr_callback_register call is passed back. It is supposed to feed rte_intr_enable, but it is not gareenteed to contain paramters as required, since it is a copy of rte_ethdev->intr_handle, if some field need to be updated (like max_intr) driver need to unregister and register the callback function again. So is it necessary to pass back a copy of intr_handle? why don't use rte_ethdev->intr_handle directly to feed rte_intr_enable? if look at the prototpye of rte_intr_callback_fn, void* cb_arg can be used to pass anything by the driver and the typical way is to pass a rte_ethdev(or any embedded bus dev) instance (actually, almost all driver implemented this already), and use this instance to get the right intr_handle.
One of the patch clean up this and the other update the test code to sync with the interface change. Qi Zhang (2): eal: improve interrupt handle test: update test code for interrupt callback change app/test/test_interrupts.c | 23 +++++++-------- drivers/net/bnxt/bnxt_irq.c | 3 +- drivers/net/e1000/em_ethdev.c | 8 ++---- drivers/net/e1000/igb_ethdev.c | 15 ++++------ drivers/net/enic/enic_main.c | 3 +- drivers/net/fm10k/fm10k_ethdev.c | 12 +++----- drivers/net/i40e/i40e_ethdev.c | 8 ++---- drivers/net/i40e/i40e_ethdev_vf.c | 5 ++-- drivers/net/ixgbe/ixgbe_ethdev.c | 14 ++++----- drivers/net/nfp/nfp_net.c | 6 ++-- drivers/net/qede/qede_ethdev.c | 4 +-- drivers/net/sfc/sfc_intr.c | 10 ++++--- drivers/net/virtio/virtio_ethdev.c | 5 ++-- lib/librte_eal/common/include/rte_interrupts.h | 3 +- lib/librte_eal/linuxapp/eal/eal_alarm.c | 5 ++-- lib/librte_eal/linuxapp/eal/eal_interrupts.c | 39 ++++---------------------- 16 files changed, 57 insertions(+), 106 deletions(-) -- 2.7.4
