From ac7100ba93428a26cde8e47dfcfcfbfbfcce66de Mon Sep 17 00:00:00 2001 From: Ben Hutchings Date: Mon, 14 Feb 2011 19:02:23 +0000 Subject: [PATCH 1/5] sch_mqprio: Always set num_tc to 0 in mqprio_destroy() All the cleanup code in mqprio_destroy() is currently conditional on priv->qdiscs being non-null, but that condition should only apply to the per-queue qdisc cleanup. We should always set the number of traffic classes back to 0 here. Signed-off-by: Ben Hutchings --- net/sched/sch_mqprio.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/net/sched/sch_mqprio.c b/net/sched/sch_mqprio.c index effd4ee0e880..ace37f9f1cd0 100644 --- a/net/sched/sch_mqprio.c +++ b/net/sched/sch_mqprio.c @@ -29,18 +29,18 @@ static void mqprio_destroy(struct Qdisc *sch) struct mqprio_sched *priv = qdisc_priv(sch); unsigned int ntx; - if (!priv->qdiscs) - return; - - for (ntx = 0; ntx < dev->num_tx_queues && priv->qdiscs[ntx]; ntx++) - qdisc_destroy(priv->qdiscs[ntx]); + if (priv->qdiscs) { + for (ntx = 0; + ntx < dev->num_tx_queues && priv->qdiscs[ntx]; + ntx++) + qdisc_destroy(priv->qdiscs[ntx]); + kfree(priv->qdiscs); + } if (priv->hw_owned && dev->netdev_ops->ndo_setup_tc) dev->netdev_ops->ndo_setup_tc(dev, 0); else netdev_set_num_tc(dev, 0); - - kfree(priv->qdiscs); } static int mqprio_parse_opt(struct net_device *dev, struct tc_mqprio_qopt *qopt) From 5c56580b74e57e56f30e3c5bbc9d7ab487858497 Mon Sep 17 00:00:00 2001 From: Ben Hutchings Date: Tue, 15 Feb 2011 19:39:21 +0000 Subject: [PATCH 2/5] net: Adjust TX queue kobjects if number of queues changes during unregister If the root qdisc for a net device is mqprio, and the driver's ndo_setup_tc() operation dynamically adds and remvoes TX queues, netif_set_real_num_tx_queues() will be called during device unregistration to remove the extra TX queues when the qdisc is destroyed. Currently this causes the corresponding kobjects to be leaked, and the device's reference count never drops to 0. Signed-off-by: Ben Hutchings --- net/core/dev.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/net/core/dev.c b/net/core/dev.c index 6392ea0a5910..30c71f9b0419 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -1648,7 +1648,8 @@ int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) if (txq < 1 || txq > dev->num_tx_queues) return -EINVAL; - if (dev->reg_state == NETREG_REGISTERED) { + if (dev->reg_state == NETREG_REGISTERED || + dev->reg_state == NETREG_UNREGISTERING) { ASSERT_RTNL(); rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, From 60031fcc17057e21566ed34ba23e93fffda1aa27 Mon Sep 17 00:00:00 2001 From: Ben Hutchings Date: Wed, 12 Jan 2011 18:39:40 +0000 Subject: [PATCH 3/5] sfc: Move TX queue core queue mapping into tx.c efx_hard_start_xmit() needs to implement a mapping which is the inverse of tx_queue::core_txq. Move the initialisation of tx_queue::core_txq next to efx_hard_start_xmit() to make the connection more obvious. Signed-off-by: Ben Hutchings --- drivers/net/sfc/efx.c | 6 ++---- drivers/net/sfc/efx.h | 1 + drivers/net/sfc/tx.c | 7 +++++++ 3 files changed, 10 insertions(+), 4 deletions(-) diff --git a/drivers/net/sfc/efx.c b/drivers/net/sfc/efx.c index 002bac743843..c559bc372fc1 100644 --- a/drivers/net/sfc/efx.c +++ b/drivers/net/sfc/efx.c @@ -1910,10 +1910,8 @@ static int efx_register_netdev(struct efx_nic *efx) efx_for_each_channel(channel, efx) { struct efx_tx_queue *tx_queue; - efx_for_each_channel_tx_queue(tx_queue, channel) { - tx_queue->core_txq = netdev_get_tx_queue( - efx->net_dev, tx_queue->queue / EFX_TXQ_TYPES); - } + efx_for_each_channel_tx_queue(tx_queue, channel) + efx_init_tx_queue_core_txq(tx_queue); } /* Always start with carrier off; PHY events will detect the link */ diff --git a/drivers/net/sfc/efx.h b/drivers/net/sfc/efx.h index d43a7e5212b1..116207045068 100644 --- a/drivers/net/sfc/efx.h +++ b/drivers/net/sfc/efx.h @@ -29,6 +29,7 @@ extern int efx_probe_tx_queue(struct efx_tx_queue *tx_queue); extern void efx_remove_tx_queue(struct efx_tx_queue *tx_queue); extern void efx_init_tx_queue(struct efx_tx_queue *tx_queue); +extern void efx_init_tx_queue_core_txq(struct efx_tx_queue *tx_queue); extern void efx_fini_tx_queue(struct efx_tx_queue *tx_queue); extern void efx_release_tx_buffers(struct efx_tx_queue *tx_queue); extern netdev_tx_t diff --git a/drivers/net/sfc/tx.c b/drivers/net/sfc/tx.c index 2f5e9da657bf..7e463fb19fb9 100644 --- a/drivers/net/sfc/tx.c +++ b/drivers/net/sfc/tx.c @@ -347,6 +347,13 @@ netdev_tx_t efx_hard_start_xmit(struct sk_buff *skb, return efx_enqueue_skb(tx_queue, skb); } +void efx_init_tx_queue_core_txq(struct efx_tx_queue *tx_queue) +{ + /* Must be inverse of queue lookup in efx_hard_start_xmit() */ + tx_queue->core_txq = netdev_get_tx_queue( + tx_queue->efx->net_dev, tx_queue->queue / EFX_TXQ_TYPES); +} + void efx_xmit_done(struct efx_tx_queue *tx_queue, unsigned int index) { unsigned fill_level; From 525da9072c28df815bff64bf00f3b11ab88face8 Mon Sep 17 00:00:00 2001 From: Ben Hutchings Date: Mon, 7 Feb 2011 23:04:38 +0000 Subject: [PATCH 4/5] sfc: Distinguish queue lookup from test for queue existence efx_channel_get_{rx,tx}_queue() currently return NULL if the channel isn't used for traffic in that direction. In most cases this is a bug, but some callers rely on it as an existence test. Add existence test functions efx_channel_has_{rx_queue,tx_queues}() and use them as appropriate. Change efx_channel_get_{rx,tx}_queue() to assert that the requested queue exists. Remove now-redundant initialisation from efx_set_channels(). Signed-off-by: Ben Hutchings --- drivers/net/sfc/efx.c | 17 ++-------------- drivers/net/sfc/ethtool.c | 6 +++--- drivers/net/sfc/net_driver.h | 39 ++++++++++++++++++++++++++---------- 3 files changed, 33 insertions(+), 29 deletions(-) diff --git a/drivers/net/sfc/efx.c b/drivers/net/sfc/efx.c index c559bc372fc1..6189d3066018 100644 --- a/drivers/net/sfc/efx.c +++ b/drivers/net/sfc/efx.c @@ -1271,21 +1271,8 @@ static void efx_remove_interrupts(struct efx_nic *efx) static void efx_set_channels(struct efx_nic *efx) { - struct efx_channel *channel; - struct efx_tx_queue *tx_queue; - efx->tx_channel_offset = separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0; - - /* Channel pointers were set in efx_init_struct() but we now - * need to clear them for TX queues in any RX-only channels. */ - efx_for_each_channel(channel, efx) { - if (channel->channel - efx->tx_channel_offset >= - efx->n_tx_channels) { - efx_for_each_channel_tx_queue(tx_queue, channel) - tx_queue->channel = NULL; - } - } } static int efx_probe_nic(struct efx_nic *efx) @@ -1531,9 +1518,9 @@ void efx_init_irq_moderation(struct efx_nic *efx, int tx_usecs, int rx_usecs, efx->irq_rx_adaptive = rx_adaptive; efx->irq_rx_moderation = rx_ticks; efx_for_each_channel(channel, efx) { - if (efx_channel_get_rx_queue(channel)) + if (efx_channel_has_rx_queue(channel)) channel->irq_moderation = rx_ticks; - else if (efx_channel_get_tx_queue(channel, 0)) + else if (efx_channel_has_tx_queues(channel)) channel->irq_moderation = tx_ticks; } } diff --git a/drivers/net/sfc/ethtool.c b/drivers/net/sfc/ethtool.c index 713969accdbd..272cfe724e1b 100644 --- a/drivers/net/sfc/ethtool.c +++ b/drivers/net/sfc/ethtool.c @@ -631,7 +631,7 @@ static int efx_ethtool_get_coalesce(struct net_device *net_dev, /* Find lowest IRQ moderation across all used TX queues */ coalesce->tx_coalesce_usecs_irq = ~((u32) 0); efx_for_each_channel(channel, efx) { - if (!efx_channel_get_tx_queue(channel, 0)) + if (!efx_channel_has_tx_queues(channel)) continue; if (channel->irq_moderation < coalesce->tx_coalesce_usecs_irq) { if (channel->channel < efx->n_rx_channels) @@ -676,8 +676,8 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev, /* If the channel is shared only allow RX parameters to be set */ efx_for_each_channel(channel, efx) { - if (efx_channel_get_rx_queue(channel) && - efx_channel_get_tx_queue(channel, 0) && + if (efx_channel_has_rx_queue(channel) && + efx_channel_has_tx_queues(channel) && tx_usecs) { netif_err(efx, drv, efx->net_dev, "Channel is shared. " "Only RX coalescing may be set\n"); diff --git a/drivers/net/sfc/net_driver.h b/drivers/net/sfc/net_driver.h index c65270241d2d..77b7ce451519 100644 --- a/drivers/net/sfc/net_driver.h +++ b/drivers/net/sfc/net_driver.h @@ -938,19 +938,28 @@ efx_get_tx_queue(struct efx_nic *efx, unsigned index, unsigned type) return &efx->channel[efx->tx_channel_offset + index]->tx_queue[type]; } +static inline bool efx_channel_has_tx_queues(struct efx_channel *channel) +{ + return channel->channel - channel->efx->tx_channel_offset < + channel->efx->n_tx_channels; +} + static inline struct efx_tx_queue * efx_channel_get_tx_queue(struct efx_channel *channel, unsigned type) { - struct efx_tx_queue *tx_queue = channel->tx_queue; - EFX_BUG_ON_PARANOID(type >= EFX_TXQ_TYPES); - return tx_queue->channel ? tx_queue + type : NULL; + EFX_BUG_ON_PARANOID(!efx_channel_has_tx_queues(channel) || + type >= EFX_TXQ_TYPES); + return &channel->tx_queue[type]; } /* Iterate over all TX queues belonging to a channel */ #define efx_for_each_channel_tx_queue(_tx_queue, _channel) \ - for (_tx_queue = efx_channel_get_tx_queue(channel, 0); \ - _tx_queue && _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES; \ - _tx_queue++) + if (!efx_channel_has_tx_queues(_channel)) \ + ; \ + else \ + for (_tx_queue = (_channel)->tx_queue; \ + _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES; \ + _tx_queue++) static inline struct efx_rx_queue * efx_get_rx_queue(struct efx_nic *efx, unsigned index) @@ -959,18 +968,26 @@ efx_get_rx_queue(struct efx_nic *efx, unsigned index) return &efx->channel[index]->rx_queue; } +static inline bool efx_channel_has_rx_queue(struct efx_channel *channel) +{ + return channel->channel < channel->efx->n_rx_channels; +} + static inline struct efx_rx_queue * efx_channel_get_rx_queue(struct efx_channel *channel) { - return channel->channel < channel->efx->n_rx_channels ? - &channel->rx_queue : NULL; + EFX_BUG_ON_PARANOID(!efx_channel_has_rx_queue(channel)); + return &channel->rx_queue; } /* Iterate over all RX queues belonging to a channel */ #define efx_for_each_channel_rx_queue(_rx_queue, _channel) \ - for (_rx_queue = efx_channel_get_rx_queue(channel); \ - _rx_queue; \ - _rx_queue = NULL) + if (!efx_channel_has_rx_queue(_channel)) \ + ; \ + else \ + for (_rx_queue = &(_channel)->rx_queue; \ + _rx_queue; \ + _rx_queue = NULL) static inline struct efx_channel * efx_rx_queue_channel(struct efx_rx_queue *rx_queue) From 94b274bf5fba6c75b922c8a23ad4b5639a168780 Mon Sep 17 00:00:00 2001 From: Ben Hutchings Date: Mon, 10 Jan 2011 21:18:20 +0000 Subject: [PATCH 5/5] sfc: Add TX queues for high-priority traffic Implement the ndo_setup_tc() operation with 2 traffic classes. Current Solarstorm controllers do not implement TX queue priority, but they do allow queues to be 'paced' with an enforced delay between packets. Paced and unpaced queues are scheduled in round-robin within two separate hardware bins (paced queues with a large delay may be placed into a third bin temporarily, but we won't use that). If there are queues in both bins, the TX scheduler will alternate between them. If we make high-priority queues unpaced and best-effort queues paced, and high-priority queues are mostly empty, a single high-priority queue can then instantly take 50% of the packet rate regardless of how many of the best-effort queues have descriptors outstanding. We do not actually want an enforced delay between packets on best- effort queues, so we set the pace value to a reserved value that actually results in a delay of 0. Signed-off-by: Ben Hutchings --- drivers/net/sfc/efx.c | 8 ++-- drivers/net/sfc/efx.h | 1 + drivers/net/sfc/net_driver.h | 29 +++++++++--- drivers/net/sfc/nic.c | 51 +++++++++++++++------ drivers/net/sfc/regs.h | 6 +++ drivers/net/sfc/selftest.c | 2 +- drivers/net/sfc/tx.c | 87 +++++++++++++++++++++++++++++++++--- 7 files changed, 156 insertions(+), 28 deletions(-) diff --git a/drivers/net/sfc/efx.c b/drivers/net/sfc/efx.c index 6189d3066018..d4e04256730b 100644 --- a/drivers/net/sfc/efx.c +++ b/drivers/net/sfc/efx.c @@ -673,7 +673,7 @@ static void efx_fini_channels(struct efx_nic *efx) efx_for_each_channel_rx_queue(rx_queue, channel) efx_fini_rx_queue(rx_queue); - efx_for_each_channel_tx_queue(tx_queue, channel) + efx_for_each_possible_channel_tx_queue(tx_queue, channel) efx_fini_tx_queue(tx_queue); efx_fini_eventq(channel); } @@ -689,7 +689,7 @@ static void efx_remove_channel(struct efx_channel *channel) efx_for_each_channel_rx_queue(rx_queue, channel) efx_remove_rx_queue(rx_queue); - efx_for_each_channel_tx_queue(tx_queue, channel) + efx_for_each_possible_channel_tx_queue(tx_queue, channel) efx_remove_tx_queue(tx_queue); efx_remove_eventq(channel); } @@ -1836,6 +1836,7 @@ static const struct net_device_ops efx_netdev_ops = { #ifdef CONFIG_NET_POLL_CONTROLLER .ndo_poll_controller = efx_netpoll, #endif + .ndo_setup_tc = efx_setup_tc, }; static void efx_update_name(struct efx_nic *efx) @@ -2386,7 +2387,8 @@ static int __devinit efx_pci_probe(struct pci_dev *pci_dev, int i, rc; /* Allocate and initialise a struct net_device and struct efx_nic */ - net_dev = alloc_etherdev_mq(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES); + net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES, + EFX_MAX_RX_QUEUES); if (!net_dev) return -ENOMEM; net_dev->features |= (type->offload_features | NETIF_F_SG | diff --git a/drivers/net/sfc/efx.h b/drivers/net/sfc/efx.h index 116207045068..0cb198a64a63 100644 --- a/drivers/net/sfc/efx.h +++ b/drivers/net/sfc/efx.h @@ -37,6 +37,7 @@ efx_hard_start_xmit(struct sk_buff *skb, struct net_device *net_dev); extern netdev_tx_t efx_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb); extern void efx_xmit_done(struct efx_tx_queue *tx_queue, unsigned int index); +extern int efx_setup_tc(struct net_device *net_dev, u8 num_tc); /* RX */ extern int efx_probe_rx_queue(struct efx_rx_queue *rx_queue); diff --git a/drivers/net/sfc/net_driver.h b/drivers/net/sfc/net_driver.h index 77b7ce451519..96e22ad34970 100644 --- a/drivers/net/sfc/net_driver.h +++ b/drivers/net/sfc/net_driver.h @@ -63,10 +63,12 @@ /* Checksum generation is a per-queue option in hardware, so each * queue visible to the networking core is backed by two hardware TX * queues. */ -#define EFX_MAX_CORE_TX_QUEUES EFX_MAX_CHANNELS -#define EFX_TXQ_TYPE_OFFLOAD 1 -#define EFX_TXQ_TYPES 2 -#define EFX_MAX_TX_QUEUES (EFX_TXQ_TYPES * EFX_MAX_CORE_TX_QUEUES) +#define EFX_MAX_TX_TC 2 +#define EFX_MAX_CORE_TX_QUEUES (EFX_MAX_TX_TC * EFX_MAX_CHANNELS) +#define EFX_TXQ_TYPE_OFFLOAD 1 /* flag */ +#define EFX_TXQ_TYPE_HIGHPRI 2 /* flag */ +#define EFX_TXQ_TYPES 4 +#define EFX_MAX_TX_QUEUES (EFX_TXQ_TYPES * EFX_MAX_CHANNELS) /** * struct efx_special_buffer - An Efx special buffer @@ -140,6 +142,7 @@ struct efx_tx_buffer { * @buffer: The software buffer ring * @txd: The hardware descriptor ring * @ptr_mask: The size of the ring minus 1. + * @initialised: Has hardware queue been initialised? * @flushed: Used when handling queue flushing * @read_count: Current read pointer. * This is the number of buffers that have been removed from both rings. @@ -182,6 +185,7 @@ struct efx_tx_queue { struct efx_tx_buffer *buffer; struct efx_special_buffer txd; unsigned int ptr_mask; + bool initialised; enum efx_flush_state flushed; /* Members used mainly on the completion path */ @@ -377,7 +381,7 @@ struct efx_channel { bool rx_pkt_csummed; struct efx_rx_queue rx_queue; - struct efx_tx_queue tx_queue[2]; + struct efx_tx_queue tx_queue[EFX_TXQ_TYPES]; }; enum efx_led_mode { @@ -952,15 +956,28 @@ efx_channel_get_tx_queue(struct efx_channel *channel, unsigned type) return &channel->tx_queue[type]; } +static inline bool efx_tx_queue_used(struct efx_tx_queue *tx_queue) +{ + return !(tx_queue->efx->net_dev->num_tc < 2 && + tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI); +} + /* Iterate over all TX queues belonging to a channel */ #define efx_for_each_channel_tx_queue(_tx_queue, _channel) \ if (!efx_channel_has_tx_queues(_channel)) \ ; \ else \ for (_tx_queue = (_channel)->tx_queue; \ - _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES; \ + _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES && \ + efx_tx_queue_used(_tx_queue); \ _tx_queue++) +/* Iterate over all possible TX queues belonging to a channel */ +#define efx_for_each_possible_channel_tx_queue(_tx_queue, _channel) \ + for (_tx_queue = (_channel)->tx_queue; \ + _tx_queue < (_channel)->tx_queue + EFX_TXQ_TYPES; \ + _tx_queue++) + static inline struct efx_rx_queue * efx_get_rx_queue(struct efx_nic *efx, unsigned index) { diff --git a/drivers/net/sfc/nic.c b/drivers/net/sfc/nic.c index da386599ab68..1d0b8b6f25c4 100644 --- a/drivers/net/sfc/nic.c +++ b/drivers/net/sfc/nic.c @@ -445,8 +445,8 @@ int efx_nic_probe_tx(struct efx_tx_queue *tx_queue) void efx_nic_init_tx(struct efx_tx_queue *tx_queue) { - efx_oword_t tx_desc_ptr; struct efx_nic *efx = tx_queue->efx; + efx_oword_t reg; tx_queue->flushed = FLUSH_NONE; @@ -454,7 +454,7 @@ void efx_nic_init_tx(struct efx_tx_queue *tx_queue) efx_init_special_buffer(efx, &tx_queue->txd); /* Push TX descriptor ring to card */ - EFX_POPULATE_OWORD_10(tx_desc_ptr, + EFX_POPULATE_OWORD_10(reg, FRF_AZ_TX_DESCQ_EN, 1, FRF_AZ_TX_ISCSI_DDIG_EN, 0, FRF_AZ_TX_ISCSI_HDIG_EN, 0, @@ -470,17 +470,15 @@ void efx_nic_init_tx(struct efx_tx_queue *tx_queue) if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) { int csum = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD; - EFX_SET_OWORD_FIELD(tx_desc_ptr, FRF_BZ_TX_IP_CHKSM_DIS, !csum); - EFX_SET_OWORD_FIELD(tx_desc_ptr, FRF_BZ_TX_TCP_CHKSM_DIS, + EFX_SET_OWORD_FIELD(reg, FRF_BZ_TX_IP_CHKSM_DIS, !csum); + EFX_SET_OWORD_FIELD(reg, FRF_BZ_TX_TCP_CHKSM_DIS, !csum); } - efx_writeo_table(efx, &tx_desc_ptr, efx->type->txd_ptr_tbl_base, + efx_writeo_table(efx, ®, efx->type->txd_ptr_tbl_base, tx_queue->queue); if (efx_nic_rev(efx) < EFX_REV_FALCON_B0) { - efx_oword_t reg; - /* Only 128 bits in this register */ BUILD_BUG_ON(EFX_MAX_TX_QUEUES > 128); @@ -491,6 +489,16 @@ void efx_nic_init_tx(struct efx_tx_queue *tx_queue) set_bit_le(tx_queue->queue, (void *)®); efx_writeo(efx, ®, FR_AA_TX_CHKSM_CFG); } + + if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) { + EFX_POPULATE_OWORD_1(reg, + FRF_BZ_TX_PACE, + (tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI) ? + FFE_BZ_TX_PACE_OFF : + FFE_BZ_TX_PACE_RESERVED); + efx_writeo_table(efx, ®, FR_BZ_TX_PACE_TBL, + tx_queue->queue); + } } static void efx_flush_tx_queue(struct efx_tx_queue *tx_queue) @@ -1238,8 +1246,10 @@ int efx_nic_flush_queues(struct efx_nic *efx) /* Flush all tx queues in parallel */ efx_for_each_channel(channel, efx) { - efx_for_each_channel_tx_queue(tx_queue, channel) - efx_flush_tx_queue(tx_queue); + efx_for_each_possible_channel_tx_queue(tx_queue, channel) { + if (tx_queue->initialised) + efx_flush_tx_queue(tx_queue); + } } /* The hardware supports four concurrent rx flushes, each of which may @@ -1262,8 +1272,9 @@ int efx_nic_flush_queues(struct efx_nic *efx) ++rx_pending; } } - efx_for_each_channel_tx_queue(tx_queue, channel) { - if (tx_queue->flushed != FLUSH_DONE) + efx_for_each_possible_channel_tx_queue(tx_queue, channel) { + if (tx_queue->initialised && + tx_queue->flushed != FLUSH_DONE) ++tx_pending; } } @@ -1278,8 +1289,9 @@ int efx_nic_flush_queues(struct efx_nic *efx) /* Mark the queues as all flushed. We're going to return failure * leading to a reset, or fake up success anyway */ efx_for_each_channel(channel, efx) { - efx_for_each_channel_tx_queue(tx_queue, channel) { - if (tx_queue->flushed != FLUSH_DONE) + efx_for_each_possible_channel_tx_queue(tx_queue, channel) { + if (tx_queue->initialised && + tx_queue->flushed != FLUSH_DONE) netif_err(efx, hw, efx->net_dev, "tx queue %d flush command timed out\n", tx_queue->queue); @@ -1682,6 +1694,19 @@ void efx_nic_init_common(struct efx_nic *efx) if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) EFX_SET_OWORD_FIELD(temp, FRF_BZ_TX_FLUSH_MIN_LEN_EN, 1); efx_writeo(efx, &temp, FR_AZ_TX_RESERVED); + + if (efx_nic_rev(efx) >= EFX_REV_FALCON_B0) { + EFX_POPULATE_OWORD_4(temp, + /* Default values */ + FRF_BZ_TX_PACE_SB_NOT_AF, 0x15, + FRF_BZ_TX_PACE_SB_AF, 0xb, + FRF_BZ_TX_PACE_FB_BASE, 0, + /* Allow large pace values in the + * fast bin. */ + FRF_BZ_TX_PACE_BIN_TH, + FFE_BZ_TX_PACE_RESERVED); + efx_writeo(efx, &temp, FR_BZ_TX_PACE); + } } /* Register dump */ diff --git a/drivers/net/sfc/regs.h b/drivers/net/sfc/regs.h index 96430ed81c36..8227de62014f 100644 --- a/drivers/net/sfc/regs.h +++ b/drivers/net/sfc/regs.h @@ -2907,6 +2907,12 @@ #define FRF_CZ_TMFT_SRC_MAC_HI_LBN 44 #define FRF_CZ_TMFT_SRC_MAC_HI_WIDTH 16 +/* TX_PACE_TBL */ +/* Values >20 are documented as reserved, but will result in a queue going + * into the fast bin with a pace value of zero. */ +#define FFE_BZ_TX_PACE_OFF 0 +#define FFE_BZ_TX_PACE_RESERVED 21 + /* DRIVER_EV */ /* Sub-fields of an RX flush completion event */ #define FSF_AZ_DRIVER_EV_RX_FLUSH_FAIL_LBN 12 diff --git a/drivers/net/sfc/selftest.c b/drivers/net/sfc/selftest.c index 0ebfb99f1299..f936892aa423 100644 --- a/drivers/net/sfc/selftest.c +++ b/drivers/net/sfc/selftest.c @@ -644,7 +644,7 @@ static int efx_test_loopbacks(struct efx_nic *efx, struct efx_self_tests *tests, goto out; } - /* Test both types of TX queue */ + /* Test all enabled types of TX queue */ efx_for_each_channel_tx_queue(tx_queue, channel) { state->offload_csum = (tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD); diff --git a/drivers/net/sfc/tx.c b/drivers/net/sfc/tx.c index 7e463fb19fb9..1a51653bb92b 100644 --- a/drivers/net/sfc/tx.c +++ b/drivers/net/sfc/tx.c @@ -336,22 +336,89 @@ netdev_tx_t efx_hard_start_xmit(struct sk_buff *skb, { struct efx_nic *efx = netdev_priv(net_dev); struct efx_tx_queue *tx_queue; + unsigned index, type; if (unlikely(efx->port_inhibited)) return NETDEV_TX_BUSY; - tx_queue = efx_get_tx_queue(efx, skb_get_queue_mapping(skb), - skb->ip_summed == CHECKSUM_PARTIAL ? - EFX_TXQ_TYPE_OFFLOAD : 0); + index = skb_get_queue_mapping(skb); + type = skb->ip_summed == CHECKSUM_PARTIAL ? EFX_TXQ_TYPE_OFFLOAD : 0; + if (index >= efx->n_tx_channels) { + index -= efx->n_tx_channels; + type |= EFX_TXQ_TYPE_HIGHPRI; + } + tx_queue = efx_get_tx_queue(efx, index, type); return efx_enqueue_skb(tx_queue, skb); } void efx_init_tx_queue_core_txq(struct efx_tx_queue *tx_queue) { + struct efx_nic *efx = tx_queue->efx; + /* Must be inverse of queue lookup in efx_hard_start_xmit() */ - tx_queue->core_txq = netdev_get_tx_queue( - tx_queue->efx->net_dev, tx_queue->queue / EFX_TXQ_TYPES); + tx_queue->core_txq = + netdev_get_tx_queue(efx->net_dev, + tx_queue->queue / EFX_TXQ_TYPES + + ((tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI) ? + efx->n_tx_channels : 0)); +} + +int efx_setup_tc(struct net_device *net_dev, u8 num_tc) +{ + struct efx_nic *efx = netdev_priv(net_dev); + struct efx_channel *channel; + struct efx_tx_queue *tx_queue; + unsigned tc; + int rc; + + if (efx_nic_rev(efx) < EFX_REV_FALCON_B0 || num_tc > EFX_MAX_TX_TC) + return -EINVAL; + + if (num_tc == net_dev->num_tc) + return 0; + + for (tc = 0; tc < num_tc; tc++) { + net_dev->tc_to_txq[tc].offset = tc * efx->n_tx_channels; + net_dev->tc_to_txq[tc].count = efx->n_tx_channels; + } + + if (num_tc > net_dev->num_tc) { + /* Initialise high-priority queues as necessary */ + efx_for_each_channel(channel, efx) { + efx_for_each_possible_channel_tx_queue(tx_queue, + channel) { + if (!(tx_queue->queue & EFX_TXQ_TYPE_HIGHPRI)) + continue; + if (!tx_queue->buffer) { + rc = efx_probe_tx_queue(tx_queue); + if (rc) + return rc; + } + if (!tx_queue->initialised) + efx_init_tx_queue(tx_queue); + efx_init_tx_queue_core_txq(tx_queue); + } + } + } else { + /* Reduce number of classes before number of queues */ + net_dev->num_tc = num_tc; + } + + rc = netif_set_real_num_tx_queues(net_dev, + max_t(int, num_tc, 1) * + efx->n_tx_channels); + if (rc) + return rc; + + /* Do not destroy high-priority queues when they become + * unused. We would have to flush them first, and it is + * fairly difficult to flush a subset of TX queues. Leave + * it to efx_fini_channels(). + */ + + net_dev->num_tc = num_tc; + return 0; } void efx_xmit_done(struct efx_tx_queue *tx_queue, unsigned int index) @@ -437,6 +504,8 @@ void efx_init_tx_queue(struct efx_tx_queue *tx_queue) /* Set up TX descriptor ring */ efx_nic_init_tx(tx_queue); + + tx_queue->initialised = true; } void efx_release_tx_buffers(struct efx_tx_queue *tx_queue) @@ -459,9 +528,14 @@ void efx_release_tx_buffers(struct efx_tx_queue *tx_queue) void efx_fini_tx_queue(struct efx_tx_queue *tx_queue) { + if (!tx_queue->initialised) + return; + netif_dbg(tx_queue->efx, drv, tx_queue->efx->net_dev, "shutting down TX queue %d\n", tx_queue->queue); + tx_queue->initialised = false; + /* Flush TX queue, remove descriptor ring */ efx_nic_fini_tx(tx_queue); @@ -473,6 +547,9 @@ void efx_fini_tx_queue(struct efx_tx_queue *tx_queue) void efx_remove_tx_queue(struct efx_tx_queue *tx_queue) { + if (!tx_queue->buffer) + return; + netif_dbg(tx_queue->efx, drv, tx_queue->efx->net_dev, "destroying TX queue %d\n", tx_queue->queue); efx_nic_remove_tx(tx_queue);