Jakub Kicinski wrote:
Great! I run it on a couple of older machines.
mlx5:
TAP version 13 1..1 # timeout set to 0 # selftests: drivers/net/hw: csum.py # KTAP version 1 # 1..12 # ok 1 csum.ipv4_rx_tcp # SKIP Test requires IPv4 connectivity # ok 2 csum.ipv4_rx_tcp_invalid # SKIP Test requires IPv4 connectivity # ok 3 csum.ipv4_rx_udp # SKIP Test requires IPv4 connectivity # ok 4 csum.ipv4_rx_udp_invalid # SKIP Test requires IPv4 connectivity # ok 5 csum.ipv4_tx_udp_csum_offload # SKIP Test requires IPv4 connectivity # ok 6 csum.ipv4_tx_udp_zero_checksum # SKIP Test requires IPv4 connectivity # ok 7 csum.ipv6_rx_tcp # ok 8 csum.ipv6_rx_tcp_invalid # ok 9 csum.ipv6_rx_udp # ok 10 csum.ipv6_rx_udp_invalid # ok 11 csum.ipv6_tx_udp_csum_offload # ok 12 csum.ipv6_tx_udp_zero_checksum # # Totals: pass:6 fail:0 xfail:0 xpass:0 skip:6 error:0 ok 1 selftests: drivers/net/hw: csum.py
bnxt:
TAP version 13 1..1 # timeout set to 0 # selftests: drivers/net/hw: csum.py # KTAP version 1 # 1..12 # ok 1 csum.ipv4_rx_tcp # SKIP Test requires IPv4 connectivity # ok 2 csum.ipv4_rx_tcp_invalid # SKIP Test requires IPv4 connectivity # ok 3 csum.ipv4_rx_udp # SKIP Test requires IPv4 connectivity # ok 4 csum.ipv4_rx_udp_invalid # SKIP Test requires IPv4 connectivity # ok 5 csum.ipv4_tx_udp_csum_offload # SKIP Test requires IPv4 connectivity # ok 6 csum.ipv4_tx_udp_zero_checksum # SKIP Test requires IPv4 connectivity # ok 7 csum.ipv6_rx_tcp # ok 8 csum.ipv6_rx_tcp_invalid # ok 9 csum.ipv6_rx_udp # ok 10 csum.ipv6_rx_udp_invalid # ok 11 csum.ipv6_tx_udp_csum_offload # SKIP Test requires tx checksum offload on eth0 # ok 12 csum.ipv6_tx_udp_zero_checksum # SKIP Test requires tx checksum offload on eth0 # # Totals: pass:4 fail:0 xfail:0 xpass:0 skip:8 error:0 ok 1 selftests: drivers/net/hw: csum.py
Nice, thanks for testing!
On Wed, 1 May 2024 14:51:34 -0400 Willem de Bruijn wrote:
Run tools/testing/selftest/net/csum.c as part of drv-net. This binary covers multiple scenarios, based on arguments given, for both IPv4 and IPv6:
The use of csum.c is the only real concern I have. Could you move it to net/lib? I made net/lib into an automatically included target in commit b86761ff6374 ("selftests: net: add scaffolding for Netlink tests in Python").
It has a makefile like any selftest directory, so you should be able to do a simple move and minor path adjustments.
Without this if someone builds and deploys just the drivers/net{,/hw} targets the csum binary won't be there :( We could auto-include all of net but using the lib target felt a little cleaner.
Can do.
A few more may be in scope eventually: toeplitz, udpgso_bench, gro, so_txtime. Move them on a case-by-case basis?
Accept UDP correct checksum
Detect UDP invalid checksum
Accept TCP correct checksum
Detect TCP invalid checksum
Transmit UDP: basic checksum offload
Transmit UDP: zero checksum conversion
The test direction is reversed between receive and transmit tests, so that the NIC under test is always the local machine.
In total this adds up to 12 testcases, with more to follow. For conciseness, I replaced individual functions with a function factory. It saves a lot of boilerplate, but is a little harder to follow, so partially here as a point for discussion.
LGTM, FWIW, but let's hear if anyone feels it's too magical.
Warning that for now transmit errors are not detected, as for those the receiver runs remotely and failures with bkg are ignored.
Should I send a fix for that?
Please do. I did not grasp your suggestion well enough to take a stab. I may have already spotted the zero conversion test returning success, while explicit logging of the stderr output shows otherwise.