On 7/14/23 4:13 PM, Ivan Babrou wrote:
As requested by Jakub Kicinski and David Ahern here:
* https://lore.kernel.org/netdev/20230713201427.2c50fc7b@kernel.org/ (local)
I made some aggregations for the stacks we see leading into
skb:kfree_skb endpoint. There's a lot of data that is not easily
digestible, so I lightly massaged the data and added flamegraphs in
addition to raw stack counts. Here's the gist link:
* https://gist.github.com/bobrik/0e57671c732d9b13ac49fed85a2b2290
I see a lot of packet_rcv as the tip before kfree_skb. How many packet
sockets do you have running on that box? Can you accumulate the total
packet_rcv -> kfree_skb_reasons into 1 count -- regardless of remaining
stacktrace?
Let me know if any other format works better for you. I have perf
script output stashed just in case.
I was expecting more like perf report which should consolidate the
similar stack traces, but the flamegraph worked.
As a reminder (also mentioned in the gist), we're on v6.1, which is
the latest LTS.
I can't explain the reasons for all the network paths we have, but our
kernel / network people are CC'd if you have any questions.