On Tue, 3 Mar 2026 at 17:22, Bruce Richardson <[email protected]> wrote: > > As explained in patch 2, change the link status tracking to be via receipt > of an adminq message rather than directly querying the link status on > interrupt. As part of this change, we can reduce the cost of using the > adminq by removing the need to allocate and free memory but instead just > use the stack to store any received adminq messages. > > V4: resubmit to recheck in CI, now that next-net-intel tree is clean > > V3: added patch 1 to reduce cost of querying the adminq. > > v2: rebase on next-net-intel > > Bruce Richardson (2): > net/ice: remove malloc call for admin queue msg > net/ice: get link status updates via adminq message > > drivers/net/intel/ice/ice_ethdev.c | 62 ++++++++++-------------------- > 1 file changed, 21 insertions(+), 41 deletions(-)
Did a quick test with 2 E810 25G ports connected to another E810, with and without LSC interrupt in testpmd, and in OVS. Those ports link status gets updated quickly in all scenarios. I did not test on a long duration, nor with other types of interrupts in the mix. For the series, Tested-by: David Marchand <[email protected]> -- David Marchand
