dim: pass dim_sample to net_dim() by reference

net_dim() is currently passed a struct dim_sample argument by value.
struct dim_sample is 24 bytes. Since this is greater 16 bytes, x86-64
passes it on the stack. All callers have already initialized dim_sample
on the stack, so passing it by value requires pushing a duplicated copy
to the stack. Either witing to the stack and immediately reading it, or
perhaps dereferencing addresses relative to the stack pointer in a chain
of push instructions, seems to perform quite poorly.

In a heavy TCP workload, mlx5e_handle_rx_dim() consumes 3% of CPU time,
94% of which is attributed to the first push instruction to copy
dim_sample on the stack for the call to net_dim():
// Call ktime_get()
  0.26 |4ead2:   call   4ead7 <mlx5e_handle_rx_dim+0x47>
// Pass the address of struct dim in %rdi
       |4ead7:   lea    0x3d0(%rbx),%rdi
// Set dim_sample.pkt_ctr
       |4eade:   mov    %r13d,0x8(%rsp)
// Set dim_sample.byte_ctr
       |4eae3:   mov    %r12d,0xc(%rsp)
// Set dim_sample.event_ctr
  0.15 |4eae8:   mov    %bp,0x10(%rsp)
// Duplicate dim_sample on the stack
 94.16 |4eaed:   push   0x10(%rsp)
  2.79 |4eaf1:   push   0x10(%rsp)
  0.07 |4eaf5:   push   %rax
// Call net_dim()
  0.21 |4eaf6:   call   4eafb <mlx5e_handle_rx_dim+0x6b>

To allow the caller to reuse the struct dim_sample already on the stack,
pass the struct dim_sample by reference to net_dim().

Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Shannon Nelson <shannon.nelson@amd.com>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Reviewed-by: Arthur Kiyanovski <akiyano@amazon.com>
Reviewed-by: Louis Peens <louis.peens@corigine.com>
Link: https://patch.msgid.link/20241031002326.3426181-2-csander@purestorage.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Documentation/networking/net_dim.rst

@@ -156,7 +156,7 @@ usage is not complete but it should make the outline of the usage clear.
 		          my_entity->bytes,
 		          &dim_sample);
 	/* Call net DIM */
-	net_dim(&my_entity->dim, dim_sample);
+	net_dim(&my_entity->dim, &dim_sample);
 	...
   }
 

drivers/net/ethernet/amazon/ena/ena_netdev.c

@@ -1383,7 +1383,7 @@ static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
 			  rx_ring->rx_stats.bytes,
 			  &dim_sample);
 
-	net_dim(&ena_napi->dim, dim_sample);
+	net_dim(&ena_napi->dim, &dim_sample);
 
 	rx_ring->per_napi_packets = 0;
 }

drivers/net/ethernet/broadcom/bcmsysport.c

@@ -1029,7 +1029,7 @@ static int bcm_sysport_poll(struct napi_struct *napi, int budget)
 	if (priv->dim.use_dim) {
 		dim_update_sample(priv->dim.event_ctr, priv->dim.packets,
 				  priv->dim.bytes, &dim_sample);
-		net_dim(&priv->dim.dim, dim_sample);
+		net_dim(&priv->dim.dim, &dim_sample);
 	}
 
 	return work_done;

drivers/net/ethernet/broadcom/bnxt/bnxt.c

@@ -3102,7 +3102,7 @@ static int bnxt_poll(struct napi_struct *napi, int budget)
 				  cpr->rx_packets,
 				  cpr->rx_bytes,
 				  &dim_sample);
-		net_dim(&cpr->dim, dim_sample);
+		net_dim(&cpr->dim, &dim_sample);
 	}
 	return work_done;
 }
@@ -3233,7 +3233,7 @@ static int bnxt_poll_p5(struct napi_struct *napi, int budget)
 				  cpr_rx->rx_packets,
 				  cpr_rx->rx_bytes,
 				  &dim_sample);
-		net_dim(&cpr->dim, dim_sample);
+		net_dim(&cpr->dim, &dim_sample);
 	}
 	return work_done;
 }

drivers/net/ethernet/broadcom/genet/bcmgenet.c

@@ -2405,7 +2405,7 @@ static int bcmgenet_rx_poll(struct napi_struct *napi, int budget)
 	if (ring->dim.use_dim) {
 		dim_update_sample(ring->dim.event_ctr, ring->dim.packets,
 				  ring->dim.bytes, &dim_sample);
-		net_dim(&ring->dim.dim, dim_sample);
+		net_dim(&ring->dim.dim, &dim_sample);
 	}
 
 	return work_done;

drivers/net/ethernet/freescale/enetc/enetc.c

@@ -718,7 +718,7 @@ static void enetc_rx_net_dim(struct enetc_int_vector *v)
 			  v->rx_ring.stats.packets,
 			  v->rx_ring.stats.bytes,
 			  &dim_sample);
-	net_dim(&v->rx_dim, dim_sample);
+	net_dim(&v->rx_dim, &dim_sample);
 }
 
 static int enetc_bd_ready_count(struct enetc_bdr *tx_ring, int ci)

drivers/net/ethernet/hisilicon/hns3/hns3_enet.c

@@ -4478,7 +4478,7 @@ static void hns3_update_rx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
 
 	dim_update_sample(tqp_vector->event_cnt, rx_group->total_packets,
 			  rx_group->total_bytes, &sample);
-	net_dim(&rx_group->dim, sample);
+	net_dim(&rx_group->dim, &sample);
 }
 
 static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
@@ -4491,7 +4491,7 @@ static void hns3_update_tx_int_coalesce(struct hns3_enet_tqp_vector *tqp_vector)
 
 	dim_update_sample(tqp_vector->event_cnt, tx_group->total_packets,
 			  tx_group->total_bytes, &sample);
-	net_dim(&tx_group->dim, sample);
+	net_dim(&tx_group->dim, &sample);
 }
 
 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)

drivers/net/ethernet/intel/ice/ice_txrx.c

@@ -1352,14 +1352,14 @@ static void ice_net_dim(struct ice_q_vector *q_vector)
 		struct dim_sample dim_sample;
 
 		__ice_update_sample(q_vector, tx, &dim_sample, true);
-		net_dim(&tx->dim, dim_sample);
+		net_dim(&tx->dim, &dim_sample);
 	}
 
 	if (ITR_IS_DYNAMIC(rx)) {
 		struct dim_sample dim_sample;
 
 		__ice_update_sample(q_vector, rx, &dim_sample, false);
-		net_dim(&rx->dim, dim_sample);
+		net_dim(&rx->dim, &dim_sample);
 	}
 }
 

drivers/net/ethernet/intel/idpf/idpf_txrx.c

@@ -3679,7 +3679,7 @@ static void idpf_net_dim(struct idpf_q_vector *q_vector)
 
 	idpf_update_dim_sample(q_vector, &dim_sample, &q_vector->tx_dim,
 			       packets, bytes);
-	net_dim(&q_vector->tx_dim, dim_sample);
+	net_dim(&q_vector->tx_dim, &dim_sample);
 
 check_rx_itr:
 	if (!IDPF_ITR_IS_DYNAMIC(q_vector->rx_intr_mode))
@@ -3698,7 +3698,7 @@ static void idpf_net_dim(struct idpf_q_vector *q_vector)
 
 	idpf_update_dim_sample(q_vector, &dim_sample, &q_vector->rx_dim,
 			       packets, bytes);
-	net_dim(&q_vector->rx_dim, dim_sample);
+	net_dim(&q_vector->rx_dim, &dim_sample);
 }
 
 /**

drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c

@@ -527,7 +527,7 @@ static void otx2_adjust_adaptive_coalese(struct otx2_nic *pfvf, struct otx2_cq_p
 			  rx_frames + tx_frames,
 			  rx_bytes + tx_bytes,
 			  &dim_sample);
-	net_dim(&cq_poll->dim, dim_sample);
+	net_dim(&cq_poll->dim, &dim_sample);
 }
 
 int otx2_napi_handler(struct napi_struct *napi, int budget)

drivers/net/ethernet/mediatek/mtk_eth_soc.c

@@ -2227,7 +2227,7 @@ static int mtk_poll_rx(struct napi_struct *napi, int budget,
 	eth->rx_bytes += bytes;
 	dim_update_sample(eth->rx_events, eth->rx_packets, eth->rx_bytes,
 			  &dim_sample);
-	net_dim(&eth->rx_dim, dim_sample);
+	net_dim(&eth->rx_dim, &dim_sample);
 
 	if (xdp_flush)
 		xdp_do_flush();
@@ -2377,7 +2377,7 @@ static int mtk_poll_tx(struct mtk_eth *eth, int budget)
 
 	dim_update_sample(eth->tx_events, eth->tx_packets, eth->tx_bytes,
 			  &dim_sample);
-	net_dim(&eth->tx_dim, dim_sample);
+	net_dim(&eth->tx_dim, &dim_sample);
 
 	if (mtk_queue_stopped(eth) &&
 	    (atomic_read(&ring->free_count) > ring->thresh))

drivers/net/ethernet/mellanox/mlx5/core/en_txrx.c

@@ -55,7 +55,7 @@ static void mlx5e_handle_tx_dim(struct mlx5e_txqsq *sq)
 		return;
 
 	dim_update_sample(sq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
-	net_dim(sq->dim, dim_sample);
+	net_dim(sq->dim, &dim_sample);
 }
 
 static void mlx5e_handle_rx_dim(struct mlx5e_rq *rq)
@@ -67,7 +67,7 @@ static void mlx5e_handle_rx_dim(struct mlx5e_rq *rq)
 		return;
 
 	dim_update_sample(rq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
-	net_dim(rq->dim, dim_sample);
+	net_dim(rq->dim, &dim_sample);
 }
 
 void mlx5e_trigger_irq(struct mlx5e_icosq *sq)

drivers/net/ethernet/netronome/nfp/nfd3/dp.c

@@ -1179,7 +1179,7 @@ int nfp_nfd3_poll(struct napi_struct *napi, int budget)
 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
 
 		dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample);
-		net_dim(&r_vec->rx_dim, dim_sample);
+		net_dim(&r_vec->rx_dim, &dim_sample);
 	}
 
 	if (r_vec->nfp_net->tx_coalesce_adapt_on && r_vec->tx_ring) {
@@ -1194,7 +1194,7 @@ int nfp_nfd3_poll(struct napi_struct *napi, int budget)
 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
 
 		dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample);
-		net_dim(&r_vec->tx_dim, dim_sample);
+		net_dim(&r_vec->tx_dim, &dim_sample);
 	}
 
 	return pkts_polled;

drivers/net/ethernet/netronome/nfp/nfdk/dp.c

@@ -1289,7 +1289,7 @@ int nfp_nfdk_poll(struct napi_struct *napi, int budget)
 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
 
 		dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample);
-		net_dim(&r_vec->rx_dim, dim_sample);
+		net_dim(&r_vec->rx_dim, &dim_sample);
 	}
 
 	if (r_vec->nfp_net->tx_coalesce_adapt_on && r_vec->tx_ring) {
@@ -1304,7 +1304,7 @@ int nfp_nfdk_poll(struct napi_struct *napi, int budget)
 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
 
 		dim_update_sample(r_vec->event_ctr, pkts, bytes, &dim_sample);
-		net_dim(&r_vec->tx_dim, dim_sample);
+		net_dim(&r_vec->tx_dim, &dim_sample);
 	}
 
 	return pkts_polled;

drivers/net/ethernet/pensando/ionic/ionic_txrx.c

@@ -928,7 +928,7 @@ static void ionic_dim_update(struct ionic_qcq *qcq, int napi_mode)
 	dim_update_sample(qcq->cq.bound_intr->rearm_count,
 			  pkts, bytes, &dim_sample);
 
-	net_dim(&qcq->dim, dim_sample);
+	net_dim(&qcq->dim, &dim_sample);
 }
 
 int ionic_tx_napi(struct napi_struct *napi, int budget)

drivers/net/virtio_net.c

@@ -2804,7 +2804,7 @@ static void virtnet_rx_dim_update(struct virtnet_info *vi, struct receive_queue
 			  u64_stats_read(&rq->stats.bytes),
 			  &cur_sample);
 
-	net_dim(&rq->dim, cur_sample);
+	net_dim(&rq->dim, &cur_sample);
 	rq->packets_in_napi = 0;
 }
 

drivers/soc/fsl/dpio/dpio-service.c

@@ -891,7 +891,7 @@ void dpaa2_io_update_net_dim(struct dpaa2_io *d, __u64 frames, __u64 bytes)
 	d->frames += frames;
 
 	dim_update_sample(d->event_ctr, d->frames, d->bytes, &dim_sample);
-	net_dim(&d->rx_dim, dim_sample);
+	net_dim(&d->rx_dim, &dim_sample);
 
 	spin_unlock(&d->dim_lock);
 }

include/linux/dim.h

@@ -425,7 +425,7 @@ struct dim_cq_moder net_dim_get_def_tx_moderation(u8 cq_period_mode);
  * This is the main logic of the algorithm, where data is processed in order
  * to decide on next required action.
  */
-void net_dim(struct dim *dim, struct dim_sample end_sample);
+void net_dim(struct dim *dim, const struct dim_sample *end_sample);
 
 /* RDMA DIM */
 

lib/dim/net_dim.c

@@ -347,7 +347,7 @@ static bool net_dim_decision(struct dim_stats *curr_stats, struct dim *dim)
 	return dim->profile_ix != prev_ix;
 }
 
-void net_dim(struct dim *dim, struct dim_sample end_sample)
+void net_dim(struct dim *dim, const struct dim_sample *end_sample)
 {
 	struct dim_stats curr_stats;
 	u16 nevents;
@@ -355,11 +355,11 @@ void net_dim(struct dim *dim, struct dim_sample end_sample)
 	switch (dim->state) {
 	case DIM_MEASURE_IN_PROGRESS:
 		nevents = BIT_GAP(BITS_PER_TYPE(u16),
-				  end_sample.event_ctr,
+				  end_sample->event_ctr,
 				  dim->start_sample.event_ctr);
 		if (nevents < DIM_NEVENTS)
 			break;
-		if (!dim_calc_stats(&dim->start_sample, &end_sample, &curr_stats))
+		if (!dim_calc_stats(&dim->start_sample, end_sample, &curr_stats))
 			break;
 		if (net_dim_decision(&curr_stats, dim)) {
 			dim->state = DIM_APPLY_NEW_PROFILE;
@@ -368,8 +368,8 @@ void net_dim(struct dim *dim, struct dim_sample end_sample)
 		}
 		fallthrough;
 	case DIM_START_MEASURE:
-		dim_update_sample(end_sample.event_ctr, end_sample.pkt_ctr,
-				  end_sample.byte_ctr, &dim->start_sample);
+		dim_update_sample(end_sample->event_ctr, end_sample->pkt_ctr,
+				  end_sample->byte_ctr, &dim->start_sample);
 		dim->state = DIM_MEASURE_IN_PROGRESS;
 		break;
 	case DIM_APPLY_NEW_PROFILE: