On Mon, 23 Oct 2023 at 13:53, Albert Huang
<huangjie.albert@xxxxxxxxxxxxx> wrote:
>
> In the previous implementation, when multiple xsk sockets were
> associated with a single xsk_buff_pool, a situation could arise
> where the xsk_tx_list maintained data at the front for one xsk
> socket while starving the xsk sockets at the back of the list.
> This could result in issues such as the inability to transmit packets,
> increased latency, and jitter. To address this problem, we introduced
we introduced -> introduce
> a new variable called tx_budget_cache, which limits each xsk to transmit
> a maximum of MAX_PER_SOCKET_BUDGET tx descriptors. This allocation ensures
> equitable opportunities for subsequent xsk sockets to send tx descriptors.
> The value of MAX_PER_SOCKET_BUDGET is temporarily set to TX_BATCH_SIZE(32).
It is not temporary I hope ;-). Just say "The value of
MAX_PER_SOCKET_BUDGET is set to 32."
>
> Signed-off-by: Albert Huang <huangjie.albert@xxxxxxxxxxxxx>
> ---
> include/net/xdp_sock.h | 5 +++++
> net/xdp/xsk.c | 19 +++++++++++++++++++
> 2 files changed, 24 insertions(+)
>
> diff --git a/include/net/xdp_sock.h b/include/net/xdp_sock.h
> index 69b472604b86..08cbdf6fca85 100644
> --- a/include/net/xdp_sock.h
> +++ b/include/net/xdp_sock.h
> @@ -63,6 +63,11 @@ struct xdp_sock {
>
> struct xsk_queue *tx ____cacheline_aligned_in_smp;
> struct list_head tx_list;
> + /* Record the actual number of times xsk has transmitted a tx
> + * descriptor, with a maximum limit not exceeding MAX_PER_SOCKET_BUDGET
> + */
> + u32 tx_budget_cache;
Since this is not a cache, I would prefer a name like tx_budget_spent.
Reflects more what the purpose is of this variable. Do not forget to
change this in the commit message too.
> +
> /* Protects generic receive. */
> spinlock_t rx_lock;
>
> diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
> index f5e96e0d6e01..fd0d54b7c046 100644
> --- a/net/xdp/xsk.c
> +++ b/net/xdp/xsk.c
> @@ -33,6 +33,7 @@
> #include "xsk.h"
>
> #define TX_BATCH_SIZE 32
> +#define MAX_PER_SOCKET_BUDGET (TX_BATCH_SIZE)
>
> static DEFINE_PER_CPU(struct list_head, xskmap_flush_list);
>
> @@ -413,16 +414,25 @@ EXPORT_SYMBOL(xsk_tx_release);
>
> bool xsk_tx_peek_desc(struct xsk_buff_pool *pool, struct xdp_desc *desc)
> {
> + bool xsk_cache_full = false;
Same comment here, it is not a cache. How about just budget_exhausted?
No need for the xsk_ since this is a local variable.
> struct xdp_sock *xs;
>
> rcu_read_lock();
> +again:
> list_for_each_entry_rcu(xs, &pool->xsk_tx_list, tx_list) {
> + if (xs->tx_budget_cache >= MAX_PER_SOCKET_BUDGET) {
> + xsk_cache_full = true;
> + continue;
> + }
> +
> if (!xskq_cons_peek_desc(xs->tx, desc, pool)) {
> if (xskq_has_descs(xs->tx))
> xskq_cons_release(xs->tx);
> continue;
> }
>
> + xs->tx_budget_cache++;
> +
> /* This is the backpressure mechanism for the Tx path.
> * Reserve space in the completion queue and only proceed
> * if there is space in it. This avoids having to implement
> @@ -436,6 +446,14 @@ bool xsk_tx_peek_desc(struct xsk_buff_pool *pool, struct xdp_desc *desc)
> return true;
> }
>
> + if (xsk_cache_full) {
> + list_for_each_entry_rcu(xs, &pool->xsk_tx_list, tx_list) {
> + xs->tx_budget_cache = 0;
> + }
No need for the {} here since there is only a single line in the loop.
Please remove them.
> + xsk_cache_full = false;
> + goto again;
> + }
> +
> out:
> rcu_read_unlock();
> return false;
> @@ -1230,6 +1248,7 @@ static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
> xs->zc = xs->umem->zc;
> xs->sg = !!(xs->umem->flags & XDP_UMEM_SG_FLAG);
> xs->queue_id = qid;
> + xs->tx_budget_cache = 0;
This is not needed. The struct is zeroed at allocation.
> xp_add_xsk(xs->pool, xs);
>
> out_unlock:
> --
> 2.20.1
>
>
