@Colin, thank you for posting your findings in comment #6.
If I am reading the spreadsheet correctly, it seems like the majority of the time the CONFIG_NETWORK_PHY_TIMESTAMPING enabled kernel is on average using fewer instructions and branches for the server side (yellow cells H27-K36 and H65-K74) than the kernel without this setting (green cells B27-E36 and B65-E74). For 512 byte TCP buffers that's also the case for the client.
@Colin, thank you for posting your findings in comment #6.
If I am reading the spreadsheet correctly, it seems like the majority of the time the CONFIG_ NETWORK_ PHY_TIMESTAMPIN G enabled kernel is on average using fewer instructions and branches for the server side (yellow cells H27-K36 and H65-K74) than the kernel without this setting (green cells B27-E36 and B65-E74). For 512 byte TCP buffers that's also the case for the client.
Am I reading the spreadsheet correctly?