From aa3ac822689f5c4d8961a8707edba9b6bf4631a3 Mon Sep 17 00:00:00 2001 From: Alexander Duyck Date: Sat, 20 Sep 2014 19:50:42 -0400 Subject: [PATCH] fm10k: Add support for multiple queues This patch takes the driver from supporting a single queue to supporting multiple queues. The upper queue limit for the PF is 128 queues and the upper limit for the VF is (128 / num_vfs) rounded down to nearest power of 2. Signed-off-by: Alexander Duyck Signed-off-by: Jeff Kirsher --- drivers/net/ethernet/intel/fm10k/fm10k.h | 1 + .../net/ethernet/intel/fm10k/fm10k_ethtool.c | 57 +++++++ drivers/net/ethernet/intel/fm10k/fm10k_main.c | 148 ++++++++++++++++++ .../net/ethernet/intel/fm10k/fm10k_netdev.c | 41 +++++ 4 files changed, 247 insertions(+) diff --git a/drivers/net/ethernet/intel/fm10k/fm10k.h b/drivers/net/ethernet/intel/fm10k/fm10k.h index d8b0d343bf9e..4b6f4064c4fc 100644 --- a/drivers/net/ethernet/intel/fm10k/fm10k.h +++ b/drivers/net/ethernet/intel/fm10k/fm10k.h @@ -422,6 +422,7 @@ void fm10k_unmap_and_free_tx_resource(struct fm10k_ring *, struct fm10k_tx_buffer *); void fm10k_restore_rx_state(struct fm10k_intfc *); void fm10k_reset_rx_state(struct fm10k_intfc *); +int fm10k_setup_tc(struct net_device *dev, u8 tc); int fm10k_open(struct net_device *netdev); int fm10k_close(struct net_device *netdev); diff --git a/drivers/net/ethernet/intel/fm10k/fm10k_ethtool.c b/drivers/net/ethernet/intel/fm10k/fm10k_ethtool.c index a88c75c70b91..54e8ebd9fbe4 100644 --- a/drivers/net/ethernet/intel/fm10k/fm10k_ethtool.c +++ b/drivers/net/ethernet/intel/fm10k/fm10k_ethtool.c @@ -838,6 +838,61 @@ static int fm10k_set_rssh(struct net_device *netdev, const u32 *indir, return 0; } +static unsigned int fm10k_max_channels(struct net_device *dev) +{ + struct fm10k_intfc *interface = netdev_priv(dev); + unsigned int max_combined = interface->hw.mac.max_queues; + u8 tcs = netdev_get_num_tc(dev); + + /* For QoS report channels per traffic class */ + if (tcs > 1) + max_combined = 1 << (fls(max_combined / tcs) - 1); + + return max_combined; +} + +static void fm10k_get_channels(struct net_device *dev, + struct ethtool_channels *ch) +{ + struct fm10k_intfc *interface = netdev_priv(dev); + struct fm10k_hw *hw = &interface->hw; + + /* report maximum channels */ + ch->max_combined = fm10k_max_channels(dev); + + /* report info for other vector */ + ch->max_other = NON_Q_VECTORS(hw); + ch->other_count = ch->max_other; + + /* record RSS queues */ + ch->combined_count = interface->ring_feature[RING_F_RSS].indices; +} + +static int fm10k_set_channels(struct net_device *dev, + struct ethtool_channels *ch) +{ + struct fm10k_intfc *interface = netdev_priv(dev); + unsigned int count = ch->combined_count; + struct fm10k_hw *hw = &interface->hw; + + /* verify they are not requesting separate vectors */ + if (!count || ch->rx_count || ch->tx_count) + return -EINVAL; + + /* verify other_count has not changed */ + if (ch->other_count != NON_Q_VECTORS(hw)) + return -EINVAL; + + /* verify the number of channels does not exceed hardware limits */ + if (count > fm10k_max_channels(dev)) + return -EINVAL; + + interface->ring_feature[RING_F_RSS].limit = count; + + /* use setup TC to update any traffic class queue mapping */ + return fm10k_setup_tc(dev, netdev_get_num_tc(dev)); +} + static const struct ethtool_ops fm10k_ethtool_ops = { .get_strings = fm10k_get_strings, .get_sset_count = fm10k_get_sset_count, @@ -860,6 +915,8 @@ static const struct ethtool_ops fm10k_ethtool_ops = { .get_rxfh_key_size = fm10k_get_rssrk_size, .get_rxfh = fm10k_get_rssh, .set_rxfh = fm10k_set_rssh, + .get_channels = fm10k_get_channels, + .set_channels = fm10k_set_channels, }; void fm10k_set_ethtool_ops(struct net_device *dev) diff --git a/drivers/net/ethernet/intel/fm10k/fm10k_main.c b/drivers/net/ethernet/intel/fm10k/fm10k_main.c index f7220d841336..dae82d29af52 100644 --- a/drivers/net/ethernet/intel/fm10k/fm10k_main.c +++ b/drivers/net/ethernet/intel/fm10k/fm10k_main.c @@ -1085,6 +1085,81 @@ static int fm10k_poll(struct napi_struct *napi, int budget) return 0; } +/** + * fm10k_set_qos_queues: Allocate queues for a QOS-enabled device + * @interface: board private structure to initialize + * + * When QoS (Quality of Service) is enabled, allocate queues for + * each traffic class. If multiqueue isn't available,then abort QoS + * initialization. + * + * This function handles all combinations of Qos and RSS. + * + **/ +static bool fm10k_set_qos_queues(struct fm10k_intfc *interface) +{ + struct net_device *dev = interface->netdev; + struct fm10k_ring_feature *f; + int rss_i, i; + int pcs; + + /* Map queue offset and counts onto allocated tx queues */ + pcs = netdev_get_num_tc(dev); + + if (pcs <= 1) + return false; + + /* set QoS mask and indices */ + f = &interface->ring_feature[RING_F_QOS]; + f->indices = pcs; + f->mask = (1 << fls(pcs - 1)) - 1; + + /* determine the upper limit for our current DCB mode */ + rss_i = interface->hw.mac.max_queues / pcs; + rss_i = 1 << (fls(rss_i) - 1); + + /* set RSS mask and indices */ + f = &interface->ring_feature[RING_F_RSS]; + rss_i = min_t(u16, rss_i, f->limit); + f->indices = rss_i; + f->mask = (1 << fls(rss_i - 1)) - 1; + + /* configure pause class to queue mapping */ + for (i = 0; i < pcs; i++) + netdev_set_tc_queue(dev, i, rss_i, rss_i * i); + + interface->num_rx_queues = rss_i * pcs; + interface->num_tx_queues = rss_i * pcs; + + return true; +} + +/** + * fm10k_set_rss_queues: Allocate queues for RSS + * @interface: board private structure to initialize + * + * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try + * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU. + * + **/ +static bool fm10k_set_rss_queues(struct fm10k_intfc *interface) +{ + struct fm10k_ring_feature *f; + u16 rss_i; + + f = &interface->ring_feature[RING_F_RSS]; + rss_i = min_t(u16, interface->hw.mac.max_queues, f->limit); + + /* record indices and power of 2 mask for RSS */ + f->indices = rss_i; + f->mask = (1 << fls(rss_i - 1)) - 1; + + interface->num_rx_queues = rss_i; + interface->num_tx_queues = rss_i; + + return true; +} + /** * fm10k_set_num_queues: Allocate queues for device, feature dependent * @interface: board private structure to initialize @@ -1101,6 +1176,11 @@ static void fm10k_set_num_queues(struct fm10k_intfc *interface) /* Start with base case */ interface->num_rx_queues = 1; interface->num_tx_queues = 1; + + if (fm10k_set_qos_queues(interface)) + return; + + fm10k_set_rss_queues(interface); } /** @@ -1381,6 +1461,71 @@ static int fm10k_init_msix_capability(struct fm10k_intfc *interface) return 0; } +/** + * fm10k_cache_ring_qos - Descriptor ring to register mapping for QoS + * @interface: Interface structure continaining rings and devices + * + * Cache the descriptor ring offsets for Qos + **/ +static bool fm10k_cache_ring_qos(struct fm10k_intfc *interface) +{ + struct net_device *dev = interface->netdev; + int pc, offset, rss_i, i, q_idx; + u16 pc_stride = interface->ring_feature[RING_F_QOS].mask + 1; + u8 num_pcs = netdev_get_num_tc(dev); + + if (num_pcs <= 1) + return false; + + rss_i = interface->ring_feature[RING_F_RSS].indices; + + for (pc = 0, offset = 0; pc < num_pcs; pc++, offset += rss_i) { + q_idx = pc; + for (i = 0; i < rss_i; i++) { + interface->tx_ring[offset + i]->reg_idx = q_idx; + interface->tx_ring[offset + i]->qos_pc = pc; + interface->rx_ring[offset + i]->reg_idx = q_idx; + interface->rx_ring[offset + i]->qos_pc = pc; + q_idx += pc_stride; + } + } + + return true; +} + +/** + * fm10k_cache_ring_rss - Descriptor ring to register mapping for RSS + * @interface: Interface structure continaining rings and devices + * + * Cache the descriptor ring offsets for RSS + **/ +static void fm10k_cache_ring_rss(struct fm10k_intfc *interface) +{ + int i; + + for (i = 0; i < interface->num_rx_queues; i++) + interface->rx_ring[i]->reg_idx = i; + + for (i = 0; i < interface->num_tx_queues; i++) + interface->tx_ring[i]->reg_idx = i; +} + +/** + * fm10k_assign_rings - Map rings to network devices + * @interface: Interface structure containing rings and devices + * + * This function is meant to go though and configure both the network + * devices so that they contain rings, and configure the rings so that + * they function with their network devices. + **/ +static void fm10k_assign_rings(struct fm10k_intfc *interface) +{ + if (fm10k_cache_ring_qos(interface)) + return; + + fm10k_cache_ring_rss(interface); +} + static void fm10k_init_reta(struct fm10k_intfc *interface) { u16 i, rss_i = interface->ring_feature[RING_F_RSS].indices; @@ -1448,6 +1593,9 @@ int fm10k_init_queueing_scheme(struct fm10k_intfc *interface) if (err) return err; + /* Map rings to devices, and map devices to physical queues */ + fm10k_assign_rings(interface); + /* Initialize RSS redirection table */ fm10k_init_reta(interface); diff --git a/drivers/net/ethernet/intel/fm10k/fm10k_netdev.c b/drivers/net/ethernet/intel/fm10k/fm10k_netdev.c index b9b1459b4697..2cda9f99edc3 100644 --- a/drivers/net/ethernet/intel/fm10k/fm10k_netdev.c +++ b/drivers/net/ethernet/intel/fm10k/fm10k_netdev.c @@ -973,6 +973,46 @@ static struct rtnl_link_stats64 *fm10k_get_stats64(struct net_device *netdev, return stats; } +int fm10k_setup_tc(struct net_device *dev, u8 tc) +{ + struct fm10k_intfc *interface = netdev_priv(dev); + + /* Currently only the PF supports priority classes */ + if (tc && (interface->hw.mac.type != fm10k_mac_pf)) + return -EINVAL; + + /* Hardware supports up to 8 traffic classes */ + if (tc > 8) + return -EINVAL; + + /* Hardware has to reinitialize queues to match packet + * buffer alignment. Unfortunately, the hardware is not + * flexible enough to do this dynamically. + */ + if (netif_running(dev)) + fm10k_close(dev); + + fm10k_mbx_free_irq(interface); + + fm10k_clear_queueing_scheme(interface); + + /* we expect the prio_tc map to be repopulated later */ + netdev_reset_tc(dev); + netdev_set_num_tc(dev, tc); + + fm10k_init_queueing_scheme(interface); + + fm10k_mbx_request_irq(interface); + + if (netif_running(dev)) + fm10k_open(dev); + + /* flag to indicate SWPRI has yet to be updated */ + interface->flags |= FM10K_FLAG_SWPRI_CONFIG; + + return 0; +} + static const struct net_device_ops fm10k_netdev_ops = { .ndo_open = fm10k_open, .ndo_stop = fm10k_close, @@ -985,6 +1025,7 @@ static const struct net_device_ops fm10k_netdev_ops = { .ndo_vlan_rx_kill_vid = fm10k_vlan_rx_kill_vid, .ndo_set_rx_mode = fm10k_set_rx_mode, .ndo_get_stats64 = fm10k_get_stats64, + .ndo_setup_tc = fm10k_setup_tc, }; #define DEFAULT_DEBUG_LEVEL_SHIFT 3 -- 2.20.1