On Thu, 10 Dec 2020 16:18:53 -0800 Ivan Babrou wrote:
> Queue sharing behaviour already exists in the out-of-tree sfc driver,
> available under xdp_alloc_tx_resources module parameter.
>
> This avoids the following issue on machines with many cpus:
>
> Insufficient resources for 12 XDP event queues (24 other channels, max 32)
>
> Which in turn triggers EINVAL on XDP processing:
>
> sfc 0000:86:00.0 ext0: XDP TX failed (-22)
>
> Signed-off-by: Ivan Babrou <i...@cloudflare.com>
Please drop the references to the out-of-tree driver, the source of the
idea is not that relevant, what matters is the motivation and trade
offs.
Your patch got mangled by your email client, please try sending it with
git send-email.
> diff --git a/drivers/net/ethernet/sfc/efx_channels.c
> b/drivers/net/ethernet/sfc/efx_channels.c
> index a4a626e9cd9a..1bfeee283ea9 100644
> --- a/drivers/net/ethernet/sfc/efx_channels.c
> +++ b/drivers/net/ethernet/sfc/efx_channels.c
> @@ -17,6 +17,7 @@
> #include "rx_common.h"
> #include "nic.h"
> #include "sriov.h"
> +#include "workarounds.h"
>
> /* This is the first interrupt mode to try out of:
> * 0 => MSI-X
> @@ -137,6 +138,7 @@ static int efx_allocate_msix_channels(struct efx_nic *efx,
> {
> unsigned int n_channels = parallelism;
> int vec_count;
> + int tx_per_ev;
> int n_xdp_tx;
> int n_xdp_ev;
>
> @@ -149,9 +151,9 @@ static int efx_allocate_msix_channels(struct efx_nic *efx,
> * multiple tx queues, assuming tx and ev queues are both
> * maximum size.
> */
> -
> + tx_per_ev = EFX_MAX_EVQ_SIZE / EFX_TXQ_MAX_ENT(efx);
> n_xdp_tx = num_possible_cpus();
> - n_xdp_ev = DIV_ROUND_UP(n_xdp_tx, EFX_MAX_TXQ_PER_CHANNEL);
> + n_xdp_ev = DIV_ROUND_UP(n_xdp_tx, tx_per_ev);
>
> vec_count = pci_msix_vec_count(efx->pci_dev);
> if (vec_count < 0)
> --
> 2.29.2
