Re: [PATCH net-next 1/2] tcp: Measure TIME-WAIT reuse delay with millisecond precision

[Jakub Sitnicki <jakub@cloudflare.com>]

On Wed, Dec 04, 2024 at 08:22 PM +01, Eric Dumazet wrote:
> On Wed, Dec 4, 2024 at 7:53 PM Jakub Sitnicki <jakub@cloudflare.com> wrote:
>
>> A low effort alternative would be to introduce a new field to hold a
>> millisecond timestamp for measuring the TW reuse delay. However, this would
>> cause the struct tcp_timewait_socket size to go over 256 bytes and overflow
>> into another cache line.
>
> s/tcp_timewait_socket/tcp_timewait_sock/
>
> Can you elaborate on this ?
>
> Due to SLUB management, note that timewait_sockets are not cache
> aligned, and use 264 bytes already:
>
> # grep tw_sock_TCP /proc/slabinfo
> tw_sock_TCPv6       3596   3596    264   62    4 : tunables    0    0
>   0 : slabdata     58     58      0
> tw_sock_TCP            0      0    264   62    4 : tunables    0    0
>   0 : slabdata      0      0      0
>
> In any case, there is one 4 byte hole in struct inet_timewait_sock
> after tw_priority

You're right. <facepalm> Thanks for keeping me honest here.

I must have checked pahole on the host (Ubuntu LTS kernel) instead of
inside the dev VM, which shows:

# grep tw_sock_TCP /proc/slabinfo
tw_sock_TCPv6          0      0    288   28    2 : tunables    0    0    0 : slabdata      0      0      0
tw_sock_TCP            0      0    288   28    2 : tunables    0    0    0 : slabdata      0      0      0
# pahole -C tcp_timewait_sock
struct tcp_timewait_sock {
        struct inet_timewait_sock  tw_sk;                /*     0   256 */
        /* --- cacheline 4 boundary (256 bytes) --- */
        u32                        tw_rcv_wnd;           /*   256     4 */
        u32                        tw_ts_offset;         /*   260     4 */
        u32                        tw_ts_recent;         /*   264     4 */
        u32                        tw_last_oow_ack_time; /*   268     4 */
        u32                        tw_ts_recent_stamp;   /*   272     4 */
        u32                        tw_tx_delay;          /*   276     4 */

        /* size: 280, cachelines: 5, members: 7 */
        /* last cacheline: 24 bytes */
};

#

Let me pivot to the simplest approach then and make use of that 4-byte
hole in inet_timewait_sock. (Which I didn't consider either, so thank
you for the idea.) This would save me from having to touch the PAWS
code.