From: Alexander Duyck Date: Fri, 26 Aug 2011 07:45:20 +0000 (+0000) Subject: igb: Use node specific allocations for the q_vectors and rings X-Git-Tag: MMI-PSA29.97-13-9~18194^2~105^2~9 X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=81c2fc22323f461aee30cf7028a79eb67426e4b6;p=GitHub%2FMotorolaMobilityLLC%2Fkernel-slsi.git igb: Use node specific allocations for the q_vectors and rings This change is meant to update the ring and vector allocations so that they are per node instead of allocating everything on the node that ifconfig/modprobe is called on. By doing this we can cut down significantly on cross node traffic. Signed-off-by: Alexander Duyck Tested-by: Aaron Brown Signed-off-by: Jeff Kirsher --- diff --git a/drivers/net/ethernet/intel/igb/igb.h b/drivers/net/ethernet/intel/igb/igb.h index de35c02876aa..9e4bed37d9be 100644 --- a/drivers/net/ethernet/intel/igb/igb.h +++ b/drivers/net/ethernet/intel/igb/igb.h @@ -185,6 +185,8 @@ struct igb_q_vector { u16 cpu; u16 tx_work_limit; + int numa_node; + u16 itr_val; u8 set_itr; void __iomem *itr_register; @@ -232,6 +234,7 @@ struct igb_ring { }; /* Items past this point are only used during ring alloc / free */ dma_addr_t dma; /* phys address of the ring */ + int numa_node; /* node to alloc ring memory on */ }; #define IGB_RING_FLAG_RX_CSUM 0x00000001 /* RX CSUM enabled */ @@ -341,6 +344,7 @@ struct igb_adapter { int vf_rate_link_speed; u32 rss_queues; u32 wvbr; + int node; }; #define IGB_FLAG_HAS_MSI (1 << 0) diff --git a/drivers/net/ethernet/intel/igb/igb_main.c b/drivers/net/ethernet/intel/igb/igb_main.c index 1c234f03b21c..287be855107a 100644 --- a/drivers/net/ethernet/intel/igb/igb_main.c +++ b/drivers/net/ethernet/intel/igb/igb_main.c @@ -687,41 +687,68 @@ static int igb_alloc_queues(struct igb_adapter *adapter) { struct igb_ring *ring; int i; + int orig_node = adapter->node; for (i = 0; i < adapter->num_tx_queues; i++) { - ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL); + if (orig_node == -1) { + int cur_node = next_online_node(adapter->node); + if (cur_node == MAX_NUMNODES) + cur_node = first_online_node; + adapter->node = cur_node; + } + ring = kzalloc_node(sizeof(struct igb_ring), GFP_KERNEL, + adapter->node); + if (!ring) + ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL); if (!ring) goto err; ring->count = adapter->tx_ring_count; ring->queue_index = i; ring->dev = &adapter->pdev->dev; ring->netdev = adapter->netdev; + ring->numa_node = adapter->node; /* For 82575, context index must be unique per ring. */ if (adapter->hw.mac.type == e1000_82575) ring->flags = IGB_RING_FLAG_TX_CTX_IDX; adapter->tx_ring[i] = ring; } + /* Restore the adapter's original node */ + adapter->node = orig_node; for (i = 0; i < adapter->num_rx_queues; i++) { - ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL); + if (orig_node == -1) { + int cur_node = next_online_node(adapter->node); + if (cur_node == MAX_NUMNODES) + cur_node = first_online_node; + adapter->node = cur_node; + } + ring = kzalloc_node(sizeof(struct igb_ring), GFP_KERNEL, + adapter->node); + if (!ring) + ring = kzalloc(sizeof(struct igb_ring), GFP_KERNEL); if (!ring) goto err; ring->count = adapter->rx_ring_count; ring->queue_index = i; ring->dev = &adapter->pdev->dev; ring->netdev = adapter->netdev; + ring->numa_node = adapter->node; ring->flags = IGB_RING_FLAG_RX_CSUM; /* enable rx checksum */ /* set flag indicating ring supports SCTP checksum offload */ if (adapter->hw.mac.type >= e1000_82576) ring->flags |= IGB_RING_FLAG_RX_SCTP_CSUM; adapter->rx_ring[i] = ring; } + /* Restore the adapter's original node */ + adapter->node = orig_node; igb_cache_ring_register(adapter); return 0; err: + /* Restore the adapter's original node */ + adapter->node = orig_node; igb_free_queues(adapter); return -ENOMEM; @@ -1087,9 +1114,24 @@ static int igb_alloc_q_vectors(struct igb_adapter *adapter) struct igb_q_vector *q_vector; struct e1000_hw *hw = &adapter->hw; int v_idx; + int orig_node = adapter->node; for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) { - q_vector = kzalloc(sizeof(struct igb_q_vector), GFP_KERNEL); + if ((adapter->num_q_vectors == (adapter->num_rx_queues + + adapter->num_tx_queues)) && + (adapter->num_rx_queues == v_idx)) + adapter->node = orig_node; + if (orig_node == -1) { + int cur_node = next_online_node(adapter->node); + if (cur_node == MAX_NUMNODES) + cur_node = first_online_node; + adapter->node = cur_node; + } + q_vector = kzalloc_node(sizeof(struct igb_q_vector), GFP_KERNEL, + adapter->node); + if (!q_vector) + q_vector = kzalloc(sizeof(struct igb_q_vector), + GFP_KERNEL); if (!q_vector) goto err_out; q_vector->adapter = adapter; @@ -1098,9 +1140,14 @@ static int igb_alloc_q_vectors(struct igb_adapter *adapter) netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll, 64); adapter->q_vector[v_idx] = q_vector; } + /* Restore the adapter's original node */ + adapter->node = orig_node; + return 0; err_out: + /* Restore the adapter's original node */ + adapter->node = orig_node; igb_free_q_vectors(adapter); return -ENOMEM; } @@ -2409,6 +2456,8 @@ static int __devinit igb_sw_init(struct igb_adapter *adapter) VLAN_HLEN; adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; + adapter->node = -1; + spin_lock_init(&adapter->stats64_lock); #ifdef CONFIG_PCI_IOV switch (hw->mac.type) { @@ -2579,10 +2628,13 @@ static int igb_close(struct net_device *netdev) int igb_setup_tx_resources(struct igb_ring *tx_ring) { struct device *dev = tx_ring->dev; + int orig_node = dev_to_node(dev); int size; size = sizeof(struct igb_tx_buffer) * tx_ring->count; - tx_ring->tx_buffer_info = vzalloc(size); + tx_ring->tx_buffer_info = vzalloc_node(size, tx_ring->numa_node); + if (!tx_ring->tx_buffer_info) + tx_ring->tx_buffer_info = vzalloc(size); if (!tx_ring->tx_buffer_info) goto err; @@ -2590,16 +2642,24 @@ int igb_setup_tx_resources(struct igb_ring *tx_ring) tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc); tx_ring->size = ALIGN(tx_ring->size, 4096); + set_dev_node(dev, tx_ring->numa_node); tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, &tx_ring->dma, GFP_KERNEL); + set_dev_node(dev, orig_node); + if (!tx_ring->desc) + tx_ring->desc = dma_alloc_coherent(dev, + tx_ring->size, + &tx_ring->dma, + GFP_KERNEL); if (!tx_ring->desc) goto err; tx_ring->next_to_use = 0; tx_ring->next_to_clean = 0; + return 0; err: @@ -2722,10 +2782,13 @@ static void igb_configure_tx(struct igb_adapter *adapter) int igb_setup_rx_resources(struct igb_ring *rx_ring) { struct device *dev = rx_ring->dev; + int orig_node = dev_to_node(dev); int size, desc_len; size = sizeof(struct igb_rx_buffer) * rx_ring->count; - rx_ring->rx_buffer_info = vzalloc(size); + rx_ring->rx_buffer_info = vzalloc_node(size, rx_ring->numa_node); + if (!rx_ring->rx_buffer_info) + rx_ring->rx_buffer_info = vzalloc(size); if (!rx_ring->rx_buffer_info) goto err; @@ -2735,10 +2798,17 @@ int igb_setup_rx_resources(struct igb_ring *rx_ring) rx_ring->size = rx_ring->count * desc_len; rx_ring->size = ALIGN(rx_ring->size, 4096); + set_dev_node(dev, rx_ring->numa_node); rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, &rx_ring->dma, GFP_KERNEL); + set_dev_node(dev, orig_node); + if (!rx_ring->desc) + rx_ring->desc = dma_alloc_coherent(dev, + rx_ring->size, + &rx_ring->dma, + GFP_KERNEL); if (!rx_ring->desc) goto err;