| Message ID | 20211222062201.36302-1-michal.swiatkowski@linux.intel.com |
|---|---|
| Headers | show |
| Series | refactor irq allocation in ice | expand |
On 12/22/21 01:21, Michal Swiatkowski wrote: > The ice driver uses the old PCI irq reseveration API. Change the ice > driver to use the current API. > > Implement a fallback mechanism where, if the driver can't reserve the > maximum number of interrupts, it will limit the number of queues or > disable capabilities. > > First two patches add ability to turn on and off eswitch offload. This > is needed when driver can't reserve maximum number of interrupts. In > this case driver turns off eswitch offload as driver can work > without it. Additionally, the eswitch can be supported only if SRIOV is > available, so set eswitch capabilities only if SRIOV is supported. > > Michal Swiatkowski (3): > ice: add check for eswitch support > ice: change mode only if eswitch is supported > ice: use new alloc irqs API > > drivers/net/ethernet/intel/ice/Makefile | 3 +- > drivers/net/ethernet/intel/ice/ice.h | 4 +- > drivers/net/ethernet/intel/ice/ice_arfs.c | 3 +- > drivers/net/ethernet/intel/ice/ice_eswitch.c | 46 +++- > drivers/net/ethernet/intel/ice/ice_eswitch.h | 12 + > drivers/net/ethernet/intel/ice/ice_irq.c | 213 ++++++++++++++++++ > drivers/net/ethernet/intel/ice/ice_irq.h | 12 + > drivers/net/ethernet/intel/ice/ice_lib.c | 5 +- > drivers/net/ethernet/intel/ice/ice_main.c | 220 +------------------ > drivers/net/ethernet/intel/ice/ice_xsk.c | 3 +- > 10 files changed, 300 insertions(+), 221 deletions(-) > create mode 100644 drivers/net/ethernet/intel/ice/ice_irq.c > create mode 100644 drivers/net/ethernet/intel/ice/ice_irq.h > Reviewed-by: Jonathan Toppins <jtoppins@redhat.com>
On 12/21/2021 10:21 PM, Michal Swiatkowski wrote: > The ice driver uses the old PCI irq reseveration API. Change the ice > driver to use the current API. > > Implement a fallback mechanism where, if the driver can't reserve the > maximum number of interrupts, it will limit the number of queues or > disable capabilities. A very similar implementation was already rejected by netdev [1]. I believe there's kernel work being done for dynamic MSIX allocations which is the route we should look to take on this (when it's completed). Thanks, Tony [1] https://lore.kernel.org/netdev/20210113234226.3638426-1-anthony.l.nguyen@intel.com/