| Message ID | 20220622091447.243101-1-ciara.loftus@intel.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net-next] i40e: xsk: read the XDP program once per NAPI | expand |
On Wed, Jun 22, 2022 at 4:15 AM Ciara Loftus <ciara.loftus@intel.com> wrote: > > Similar to how it's done in the ice driver since 'eb087cd82864 ("ice: > propagate xdp_ring onto rx_ring")', read the XDP program once per NAPI > instead of once per descriptor cleaned. I measured an improvement in > throughput of 2% for the AF_XDP xdpsock l2fwd benchmark in busy polling > mode on my platform. > Should the same improvement be made to i40e_run_xdp/i40e_clean_rx_irq for the non-AF_XDP case? > Signed-off-by: Ciara Loftus <ciara.loftus@intel.com> > --- > drivers/net/ethernet/intel/i40e/i40e_xsk.c | 16 +++++++++------- > 1 file changed, 9 insertions(+), 7 deletions(-) > > diff --git a/drivers/net/ethernet/intel/i40e/i40e_xsk.c b/drivers/net/ethernet/intel/i40e/i40e_xsk.c > index af3e7e6afc85..2f422c61ac11 100644 > --- a/drivers/net/ethernet/intel/i40e/i40e_xsk.c > +++ b/drivers/net/ethernet/intel/i40e/i40e_xsk.c > @@ -146,17 +146,13 @@ int i40e_xsk_pool_setup(struct i40e_vsi *vsi, struct xsk_buff_pool *pool, > * > * Returns any of I40E_XDP_{PASS, CONSUMED, TX, REDIR} > **/ > -static int i40e_run_xdp_zc(struct i40e_ring *rx_ring, struct xdp_buff *xdp) > +static int i40e_run_xdp_zc(struct i40e_ring *rx_ring, struct xdp_buff *xdp, > + struct bpf_prog *xdp_prog) > { > int err, result = I40E_XDP_PASS; > struct i40e_ring *xdp_ring; > - struct bpf_prog *xdp_prog; > u32 act; > > - /* NB! xdp_prog will always be !NULL, due to the fact that > - * this path is enabled by setting an XDP program. > - */ > - xdp_prog = READ_ONCE(rx_ring->xdp_prog); > act = bpf_prog_run_xdp(xdp_prog, xdp); > > if (likely(act == XDP_REDIRECT)) { > @@ -339,9 +335,15 @@ int i40e_clean_rx_irq_zc(struct i40e_ring *rx_ring, int budget) > u16 next_to_clean = rx_ring->next_to_clean; > u16 count_mask = rx_ring->count - 1; > unsigned int xdp_res, xdp_xmit = 0; > + struct bpf_prog *xdp_prog; > bool failure = false; > u16 cleaned_count; > > + /* NB! xdp_prog will always be !NULL, due to the fact that > + * this path is enabled by setting an XDP program. > + */ > + xdp_prog = READ_ONCE(rx_ring->xdp_prog); > + > while (likely(total_rx_packets < (unsigned int)budget)) { > union i40e_rx_desc *rx_desc; > unsigned int rx_packets; > @@ -378,7 +380,7 @@ int i40e_clean_rx_irq_zc(struct i40e_ring *rx_ring, int budget) > xsk_buff_set_size(bi, size); > xsk_buff_dma_sync_for_cpu(bi, rx_ring->xsk_pool); > > - xdp_res = i40e_run_xdp_zc(rx_ring, bi); > + xdp_res = i40e_run_xdp_zc(rx_ring, bi, xdp_prog); > i40e_handle_xdp_result_zc(rx_ring, bi, rx_desc, &rx_packets, > &rx_bytes, size, xdp_res, &failure); > if (failure) > -- > 2.25.1 >
On Wed, 22 Jun 2022 09:14:47 +0000 Ciara Loftus wrote: > Similar to how it's done in the ice driver since 'eb087cd82864 ("ice: > propagate xdp_ring onto rx_ring")', read the XDP program once per NAPI > instead of once per descriptor cleaned. I measured an improvement in > throughput of 2% for the AF_XDP xdpsock l2fwd benchmark in busy polling > mode on my platform. drivers/net/ethernet/intel/i40e/i40e_xsk.c:151: warning: Function parameter or member 'xdp_prog' not described in 'i40e_run_xdp_zc'
diff --git a/drivers/net/ethernet/intel/i40e/i40e_xsk.c b/drivers/net/ethernet/intel/i40e/i40e_xsk.c index af3e7e6afc85..2f422c61ac11 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_xsk.c +++ b/drivers/net/ethernet/intel/i40e/i40e_xsk.c @@ -146,17 +146,13 @@ int i40e_xsk_pool_setup(struct i40e_vsi *vsi, struct xsk_buff_pool *pool, * * Returns any of I40E_XDP_{PASS, CONSUMED, TX, REDIR} **/ -static int i40e_run_xdp_zc(struct i40e_ring *rx_ring, struct xdp_buff *xdp) +static int i40e_run_xdp_zc(struct i40e_ring *rx_ring, struct xdp_buff *xdp, + struct bpf_prog *xdp_prog) { int err, result = I40E_XDP_PASS; struct i40e_ring *xdp_ring; - struct bpf_prog *xdp_prog; u32 act; - /* NB! xdp_prog will always be !NULL, due to the fact that - * this path is enabled by setting an XDP program. - */ - xdp_prog = READ_ONCE(rx_ring->xdp_prog); act = bpf_prog_run_xdp(xdp_prog, xdp); if (likely(act == XDP_REDIRECT)) { @@ -339,9 +335,15 @@ int i40e_clean_rx_irq_zc(struct i40e_ring *rx_ring, int budget) u16 next_to_clean = rx_ring->next_to_clean; u16 count_mask = rx_ring->count - 1; unsigned int xdp_res, xdp_xmit = 0; + struct bpf_prog *xdp_prog; bool failure = false; u16 cleaned_count; + /* NB! xdp_prog will always be !NULL, due to the fact that + * this path is enabled by setting an XDP program. + */ + xdp_prog = READ_ONCE(rx_ring->xdp_prog); + while (likely(total_rx_packets < (unsigned int)budget)) { union i40e_rx_desc *rx_desc; unsigned int rx_packets; @@ -378,7 +380,7 @@ int i40e_clean_rx_irq_zc(struct i40e_ring *rx_ring, int budget) xsk_buff_set_size(bi, size); xsk_buff_dma_sync_for_cpu(bi, rx_ring->xsk_pool); - xdp_res = i40e_run_xdp_zc(rx_ring, bi); + xdp_res = i40e_run_xdp_zc(rx_ring, bi, xdp_prog); i40e_handle_xdp_result_zc(rx_ring, bi, rx_desc, &rx_packets, &rx_bytes, size, xdp_res, &failure); if (failure)
Similar to how it's done in the ice driver since 'eb087cd82864 ("ice: propagate xdp_ring onto rx_ring")', read the XDP program once per NAPI instead of once per descriptor cleaned. I measured an improvement in throughput of 2% for the AF_XDP xdpsock l2fwd benchmark in busy polling mode on my platform. Signed-off-by: Ciara Loftus <ciara.loftus@intel.com> --- drivers/net/ethernet/intel/i40e/i40e_xsk.c | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-)