When veb-stats are enabled ethtool -S hits the WARN_ONCE in i40e_add_one_ethtool_stat(). To reproduce: ethtool --set-priv-flags <int> veb-stats on ethtool -S <int> unexpected stat size for veb.tc_%u_tx_packets WARNING: CPU: 61 PID: 10113 at drivers/net/ethernet/intel/i40e/i40e_ethtool.c:110 i40e_add_one_ethtool_stat+0x140/0x1a0 [i40e] <cut> [ 649.377973] Call Trace: [ 649.377980] [c000003b6d937810] [c00800000debab54] i40e_add_one_ethtool_stat+0x13c/0x1a0 [i40e] (unreliable) [ 649.377996] [c000003b6d937890] [c00800000def3b4c] __i40e_add_ethtool_stats+0x84/0x6698 [i40e] [ 649.378010] [c000003b6d937920] [c00800000debe904] i40e_get_ethtool_stats+0x2bc/0x4e0 [i40e] [ 649.378023] [c000003b6d937a20] [c000000000c99854] dev_ethtool+0x6c4/0x2be0 [ 649.378033] [c000003b6d937b30] [c000000000c1340c] dev_ioctl+0x5ac/0xa50 [ 649.378042] [c000003b6d937bc0] [c000000000b8ca7c] sock_do_ioctl+0xac/0x1b0 [ 649.378051] [c000003b6d937c40] [c000000000b8d564] sock_ioctl+0x234/0x4a0 [ 649.778238] [c000003b6d937d10] [c000000000559c40] do_vfs_ioctl+0xe0/0xbd0 [ 649.778247] [c000003b6d937de0] [c00000000055a904] sys_ioctl+0xc4/0x160 [ 649.778257] [c000003b6d937e30] [c00000000000b408] system_call+0x5c/0x70
This looks like a bug in the way veb_tc_stats are setup because the structure is a series of arrays instead of an array of struct stats. I think the solution to fixing this is very similar to how the PFC stats are handled: copy the stats from the array into a suitable single-struct setup for filling in.