From 02443ab8c9314134c9cd58946121726e4cd9a5c1 Mon Sep 17 00:00:00 2001 From: Edward Cree Date: Wed, 20 Jul 2022 19:33:47 +0100 Subject: sfc: support passing a representor to the EF100 TX path A non-null efv in __ef100_enqueue_skb() indicates that the packet is from that representor, should be transmitted with a suitable option descriptor (to instruct the switch to deliver it to the representee), and should not be accounted to the parent PF's stats or BQL. Signed-off-by: Edward Cree Signed-off-by: David S. Miller --- drivers/net/ethernet/sfc/ef100_tx.c | 84 +++++++++++++++++++++++++++++++++++-- 1 file changed, 80 insertions(+), 4 deletions(-) (limited to 'drivers/net/ethernet/sfc/ef100_tx.c') diff --git a/drivers/net/ethernet/sfc/ef100_tx.c b/drivers/net/ethernet/sfc/ef100_tx.c index 26ef51d6b542..102ddc7e206a 100644 --- a/drivers/net/ethernet/sfc/ef100_tx.c +++ b/drivers/net/ethernet/sfc/ef100_tx.c @@ -254,7 +254,8 @@ static void ef100_make_tso_desc(struct efx_nic *efx, static void ef100_tx_make_descriptors(struct efx_tx_queue *tx_queue, const struct sk_buff *skb, - unsigned int segment_count) + unsigned int segment_count, + struct efx_rep *efv) { unsigned int old_write_count = tx_queue->write_count; unsigned int new_write_count = old_write_count; @@ -272,6 +273,20 @@ static void ef100_tx_make_descriptors(struct efx_tx_queue *tx_queue, else next_desc_type = ESE_GZ_TX_DESC_TYPE_SEND; + if (unlikely(efv)) { + /* Create TX override descriptor */ + write_ptr = new_write_count & tx_queue->ptr_mask; + txd = ef100_tx_desc(tx_queue, write_ptr); + ++new_write_count; + + tx_queue->packet_write_count = new_write_count; + EFX_POPULATE_OWORD_3(*txd, + ESF_GZ_TX_DESC_TYPE, ESE_GZ_TX_DESC_TYPE_PREFIX, + ESF_GZ_TX_PREFIX_EGRESS_MPORT, efv->mport, + ESF_GZ_TX_PREFIX_EGRESS_MPORT_EN, 1); + nr_descs--; + } + /* if it's a raw write (such as XDP) then always SEND single frames */ if (!skb) nr_descs = 1; @@ -306,6 +321,9 @@ static void ef100_tx_make_descriptors(struct efx_tx_queue *tx_queue, /* if it's a raw write (such as XDP) then always SEND */ next_desc_type = skb ? ESE_GZ_TX_DESC_TYPE_SEG : ESE_GZ_TX_DESC_TYPE_SEND; + /* mark as an EFV buffer if applicable */ + if (unlikely(efv)) + buffer->flags |= EFX_TX_BUF_EFV; } while (new_write_count != tx_queue->insert_count); @@ -324,7 +342,7 @@ static void ef100_tx_make_descriptors(struct efx_tx_queue *tx_queue, void ef100_tx_write(struct efx_tx_queue *tx_queue) { - ef100_tx_make_descriptors(tx_queue, NULL, 0); + ef100_tx_make_descriptors(tx_queue, NULL, 0, NULL); ef100_tx_push_buffers(tx_queue); } @@ -350,6 +368,12 @@ void ef100_ev_tx(struct efx_channel *channel, const efx_qword_t *p_event) * function will free the SKB. */ int ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb) +{ + return __ef100_enqueue_skb(tx_queue, skb, NULL); +} + +int __ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb, + struct efx_rep *efv) { unsigned int old_insert_count = tx_queue->insert_count; struct efx_nic *efx = tx_queue->efx; @@ -376,16 +400,64 @@ int ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb) return 0; } + if (unlikely(efv)) { + struct efx_tx_buffer *buffer = __efx_tx_queue_get_insert_buffer(tx_queue); + + /* Drop representor packets if the queue is stopped. + * We currently don't assert backoff to representors so this is + * to make sure representor traffic can't starve the main + * net device. + * And, of course, if there are no TX descriptors left. + */ + if (netif_tx_queue_stopped(tx_queue->core_txq) || + unlikely(efx_tx_buffer_in_use(buffer))) { + atomic64_inc(&efv->stats.tx_errors); + rc = -ENOSPC; + goto err; + } + + /* Also drop representor traffic if it could cause us to + * stop the queue. If we assert backoff and we haven't + * received traffic on the main net device recently then the + * TX watchdog can go off erroneously. + */ + fill_level = efx_channel_tx_old_fill_level(tx_queue->channel); + fill_level += efx_tx_max_skb_descs(efx); + if (fill_level > efx->txq_stop_thresh) { + struct efx_tx_queue *txq2; + + /* Refresh cached fill level and re-check */ + efx_for_each_channel_tx_queue(txq2, tx_queue->channel) + txq2->old_read_count = READ_ONCE(txq2->read_count); + + fill_level = efx_channel_tx_old_fill_level(tx_queue->channel); + fill_level += efx_tx_max_skb_descs(efx); + if (fill_level > efx->txq_stop_thresh) { + atomic64_inc(&efv->stats.tx_errors); + rc = -ENOSPC; + goto err; + } + } + + buffer->flags = EFX_TX_BUF_OPTION | EFX_TX_BUF_EFV; + tx_queue->insert_count++; + } + /* Map for DMA and create descriptors */ rc = efx_tx_map_data(tx_queue, skb, segments); if (rc) goto err; - ef100_tx_make_descriptors(tx_queue, skb, segments); + ef100_tx_make_descriptors(tx_queue, skb, segments, efv); fill_level = efx_channel_tx_old_fill_level(tx_queue->channel); if (fill_level > efx->txq_stop_thresh) { struct efx_tx_queue *txq2; + /* Because of checks above, representor traffic should + * not be able to stop the queue. + */ + WARN_ON(efv); + netif_tx_stop_queue(tx_queue->core_txq); /* Re-read after a memory barrier in case we've raced with * the completion path. Otherwise there's a danger we'll never @@ -404,8 +476,12 @@ int ef100_enqueue_skb(struct efx_tx_queue *tx_queue, struct sk_buff *skb) /* If xmit_more then we don't need to push the doorbell, unless there * are 256 descriptors already queued in which case we have to push to * ensure we never push more than 256 at once. + * + * Always push for representor traffic, and don't account it to parent + * PF netdevice's BQL. */ - if (__netdev_tx_sent_queue(tx_queue->core_txq, skb->len, xmit_more) || + if (unlikely(efv) || + __netdev_tx_sent_queue(tx_queue->core_txq, skb->len, xmit_more) || tx_queue->write_count - tx_queue->notify_count > 255) ef100_tx_push_buffers(tx_queue); -- cgit v1.2.3