From f5823fe6897c444265ef3919d8684b647eef904f Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:43 +0300 Subject: [PATCH 01/12] qed: Add ll2 option to limit the number of bds per packet iWARP uses 3 ll2 connections, the maximum number of bds is known during connection setup. This patch modifies the static array in the ll2_tx_packet descriptor to be a flexible array and significantlly reduces memory size. In addition, some redundant fields in the ll2_tx_packet were removed, which also contributed to decreasing the descriptor size. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_ll2.c | 29 ++++++++++++++++------- drivers/net/ethernet/qlogic/qed/qed_ll2.h | 9 +++---- 2 files changed, 24 insertions(+), 14 deletions(-) diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c index 250afa5486cf..75af40a7690a 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c @@ -1105,6 +1105,7 @@ static int qed_ll2_acquire_connection_tx(struct qed_hwfn *p_hwfn, struct qed_ll2_info *p_ll2_info) { struct qed_ll2_tx_packet *p_descq; + u32 desc_size; u32 capacity; int rc = 0; @@ -1122,13 +1123,17 @@ static int qed_ll2_acquire_connection_tx(struct qed_hwfn *p_hwfn, goto out; capacity = qed_chain_get_capacity(&p_ll2_info->tx_queue.txq_chain); - p_descq = kcalloc(capacity, sizeof(struct qed_ll2_tx_packet), - GFP_KERNEL); + /* First element is part of the packet, rest are flexibly added */ + desc_size = (sizeof(*p_descq) + + (p_ll2_info->input.tx_max_bds_per_packet - 1) * + sizeof(p_descq->bds_set)); + + p_descq = kcalloc(capacity, desc_size, GFP_KERNEL); if (!p_descq) { rc = -ENOMEM; goto out; } - p_ll2_info->tx_queue.descq_array = p_descq; + p_ll2_info->tx_queue.descq_mem = p_descq; DP_VERBOSE(p_hwfn, QED_MSG_LL2, "Allocated LL2 Txq [Type %08x] with 0x%08x buffers\n", @@ -1359,11 +1364,13 @@ int qed_ll2_establish_connection(void *cxt, u8 connection_handle) { struct qed_hwfn *p_hwfn = cxt; struct qed_ll2_info *p_ll2_conn; + struct qed_ll2_tx_packet *p_pkt; struct qed_ll2_rx_queue *p_rx; struct qed_ll2_tx_queue *p_tx; struct qed_ptt *p_ptt; int rc = -EINVAL; u32 i, capacity; + u32 desc_size; u8 qid; p_ptt = qed_ptt_acquire(p_hwfn); @@ -1397,9 +1404,15 @@ int qed_ll2_establish_connection(void *cxt, u8 connection_handle) INIT_LIST_HEAD(&p_tx->sending_descq); spin_lock_init(&p_tx->lock); capacity = qed_chain_get_capacity(&p_tx->txq_chain); - for (i = 0; i < capacity; i++) - list_add_tail(&p_tx->descq_array[i].list_entry, - &p_tx->free_descq); + /* First element is part of the packet, rest are flexibly added */ + desc_size = (sizeof(*p_pkt) + + (p_ll2_conn->input.tx_max_bds_per_packet - 1) * + sizeof(p_pkt->bds_set)); + + for (i = 0; i < capacity; i++) { + p_pkt = p_tx->descq_mem + desc_size * i; + list_add_tail(&p_pkt->list_entry, &p_tx->free_descq); + } p_tx->cur_completing_bd_idx = 0; p_tx->bds_idx = 0; p_tx->b_completing_packet = false; @@ -1698,7 +1711,7 @@ int qed_ll2_prepare_tx_packet(void *cxt, p_tx = &p_ll2_conn->tx_queue; p_tx_chain = &p_tx->txq_chain; - if (pkt->num_of_bds > CORE_LL2_TX_MAX_BDS_PER_PACKET) + if (pkt->num_of_bds > p_ll2_conn->input.tx_max_bds_per_packet) return -EIO; spin_lock_irqsave(&p_tx->lock, flags); @@ -1858,7 +1871,7 @@ void qed_ll2_release_connection(void *cxt, u8 connection_handle) qed_int_unregister_cb(p_hwfn, p_ll2_conn->tx_queue.tx_sb_index); } - kfree(p_ll2_conn->tx_queue.descq_array); + kfree(p_ll2_conn->tx_queue.descq_mem); qed_chain_free(p_hwfn->cdev, &p_ll2_conn->tx_queue.txq_chain); kfree(p_ll2_conn->rx_queue.descq_array); diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h index a822528e9c63..9bdd08f15c79 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h @@ -63,17 +63,14 @@ struct qed_ll2_rx_packet { struct qed_ll2_tx_packet { struct list_head list_entry; u16 bd_used; - u16 vlan; - u16 l4_hdr_offset_w; - u8 bd_flags; bool notify_fw; void *cookie; - + /* Flexible Array of bds_set determined by max_bds_per_packet */ struct { struct core_tx_bd *txq_bd; dma_addr_t tx_frag; u16 frag_len; - } bds_set[ETH_TX_MAX_BDS_PER_NON_LSO_PACKET]; + } bds_set[1]; }; struct qed_ll2_rx_queue { @@ -101,7 +98,7 @@ struct qed_ll2_tx_queue { struct list_head active_descq; struct list_head free_descq; struct list_head sending_descq; - struct qed_ll2_tx_packet *descq_array; + void *descq_mem; /* memory for variable sized qed_ll2_tx_packet*/ struct qed_ll2_tx_packet *cur_send_packet; struct qed_ll2_tx_packet cur_completing_packet; u16 cur_completing_bd_idx; From ed468ebee04ffba0231a8f50616bdb250752a891 Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:44 +0300 Subject: [PATCH 02/12] qed: Add ll2 ability of opening a secondary queue When more than one ll2 queue is opened ( that is not an OOO queue ) ll2 code does not have enough information to determine whether the queue is the main one or not, so a new field is added to the acquire input data to expose the control of determining whether the queue is the main queue or a secondary queue. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_ll2.c | 7 ++++++- drivers/net/ethernet/qlogic/qed/qed_ll2.h | 1 + include/linux/qed/qed_ll2_if.h | 1 + 3 files changed, 8 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c index 75af40a7690a..3c695da890df 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c @@ -894,7 +894,7 @@ static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn, p_ramrod->drop_ttl0_flg = p_ll2_conn->input.rx_drop_ttl0_flg; p_ramrod->inner_vlan_removal_en = p_ll2_conn->input.rx_vlan_removal_en; p_ramrod->queue_id = p_ll2_conn->queue_id; - p_ramrod->main_func_queue = (conn_type == QED_LL2_TYPE_OOO) ? 0 : 1; + p_ramrod->main_func_queue = p_ll2_conn->main_func_queue ? 1 : 0; if ((IS_MF_DEFAULT(p_hwfn) || IS_MF_SI(p_hwfn)) && p_ramrod->main_func_queue && (conn_type != QED_LL2_TYPE_ROCE) && @@ -1265,6 +1265,11 @@ int qed_ll2_acquire_connection(void *cxt, struct qed_ll2_acquire_data *data) p_ll2_info->tx_dest = (data->input.tx_dest == QED_LL2_TX_DEST_NW) ? CORE_TX_DEST_NW : CORE_TX_DEST_LB; + if (data->input.conn_type == QED_LL2_TYPE_OOO || + data->input.secondary_queue) + p_ll2_info->main_func_queue = false; + else + p_ll2_info->main_func_queue = true; /* Correct maximum number of Tx BDs */ p_tx_max = &p_ll2_info->input.tx_max_bds_per_packet; diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h index 9bdd08f15c79..f65817012e97 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.h +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.h @@ -121,6 +121,7 @@ struct qed_ll2_info { bool b_active; enum core_tx_dest tx_dest; u8 tx_stats_en; + bool main_func_queue; struct qed_ll2_rx_queue rx_queue; struct qed_ll2_tx_queue tx_queue; struct qed_ll2_cbs cbs; diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h index 89fa0bbd54f3..d7cca590b743 100644 --- a/include/linux/qed/qed_ll2_if.h +++ b/include/linux/qed/qed_ll2_if.h @@ -171,6 +171,7 @@ struct qed_ll2_acquire_data_inputs { enum qed_ll2_tx_dest tx_dest; enum qed_ll2_error_handle ai_err_packet_too_big; enum qed_ll2_error_handle ai_err_no_buf; + bool secondary_queue; u8 gsi_enable; }; From 77caa792f5d8e4ecc88eb1cf4b9c478c07e0ec57 Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:45 +0300 Subject: [PATCH 03/12] qed: Add ll2 option for dropping a tx packet The option of sending a packet on the ll2 and dropping it exists in hardware and was not used until now, thus not exposed. The iWARP unaligned MPA flow requires this functionality for flushing the tx queue. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_ll2.c | 16 ++++++++++++++-- include/linux/qed/qed_ll2_if.h | 1 + 2 files changed, 15 insertions(+), 2 deletions(-) diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c index 3c695da890df..ad67d36956e8 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c @@ -1597,8 +1597,20 @@ qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn, roce_flavor = (pkt->qed_roce_flavor == QED_LL2_ROCE) ? CORE_ROCE : CORE_RROCE; - tx_dest = (pkt->tx_dest == QED_LL2_TX_DEST_NW) ? CORE_TX_DEST_NW - : CORE_TX_DEST_LB; + switch (pkt->tx_dest) { + case QED_LL2_TX_DEST_NW: + tx_dest = CORE_TX_DEST_NW; + break; + case QED_LL2_TX_DEST_LB: + tx_dest = CORE_TX_DEST_LB; + break; + case QED_LL2_TX_DEST_DROP: + tx_dest = CORE_TX_DEST_DROP; + break; + default: + tx_dest = CORE_TX_DEST_LB; + break; + } start_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain); start_bd->nw_vlan_or_lb_echo = cpu_to_le16(pkt->vlan); diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h index d7cca590b743..95fdf02a3bbe 100644 --- a/include/linux/qed/qed_ll2_if.h +++ b/include/linux/qed/qed_ll2_if.h @@ -64,6 +64,7 @@ enum qed_ll2_roce_flavor_type { enum qed_ll2_tx_dest { QED_LL2_TX_DEST_NW, /* Light L2 TX Destination to the Network */ QED_LL2_TX_DEST_LB, /* Light L2 TX Destination to the Loopback */ + QED_LL2_TX_DEST_DROP, /* Light L2 Drop the TX packet */ QED_LL2_TX_DEST_MAX }; From 6df60fe703c348a507b0030b92c2947e68e1c589 Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:46 +0300 Subject: [PATCH 04/12] qed: Fix initialization of ll2 offload feature enable_ip_cksum, enable_l4_cksum, calc_ip_len were added in commit stated below but not passed through to FW. This was OK until now as it wasn't used, but is required for the iWARP unaligned flow Fixes:7c7973b2ae27 ("qed: LL2 to use packed information for tx") Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_ll2.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c index ad67d36956e8..6d144747111a 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c @@ -1621,6 +1621,9 @@ qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn, SET_FIELD(bd_data, CORE_TX_BD_DATA_START_BD, 0x1); SET_FIELD(bd_data, CORE_TX_BD_DATA_NBDS, pkt->num_of_bds); SET_FIELD(bd_data, CORE_TX_BD_DATA_ROCE_FLAV, roce_flavor); + SET_FIELD(bd_data, CORE_TX_BD_DATA_IP_CSUM, !!(pkt->enable_ip_cksum)); + SET_FIELD(bd_data, CORE_TX_BD_DATA_L4_CSUM, !!(pkt->enable_l4_cksum)); + SET_FIELD(bd_data, CORE_TX_BD_DATA_IP_LEN, !!(pkt->calc_ip_len)); start_bd->bd_data.as_bitfield = cpu_to_le16(bd_data); DMA_REGPAIR_LE(start_bd->addr, pkt->first_frag); start_bd->nbytes = cpu_to_le16(pkt->first_frag_len); From 89d65113097072de7936a2aea2f819818a7c987a Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:47 +0300 Subject: [PATCH 05/12] qed: Add the source of a packet sent on an iWARP ll2 connection When a packet is sent back to iWARP FW via the tx ll2 connection the FW needs to know the source of the packet. Whether it is OOO or unaligned MPA related. Since OOO is implemented entirely inside the ll2 code (and shared with iSCSI), packets are marked as IN_ORDER inside the ll2 code. For unaligned mpa the value will be determined in the iWARP code and sent on the pkt->vlan field. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_ll2.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c index 6d144747111a..8eb9645c880d 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c @@ -1613,7 +1613,12 @@ qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn, } start_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain); - start_bd->nw_vlan_or_lb_echo = cpu_to_le16(pkt->vlan); + if (QED_IS_IWARP_PERSONALITY(p_hwfn) && + p_ll2->input.conn_type == QED_LL2_TYPE_OOO) + start_bd->nw_vlan_or_lb_echo = + cpu_to_le16(IWARP_LL2_IN_ORDER_TX_QUEUE); + else + start_bd->nw_vlan_or_lb_echo = cpu_to_le16(pkt->vlan); SET_FIELD(start_bd->bitfield1, CORE_TX_BD_L4_HDR_OFFSET_W, cpu_to_le16(pkt->l4_hdr_offset_w)); SET_FIELD(start_bd->bitfield1, CORE_TX_BD_TX_DST, tx_dest); From 6f34a284f36399501fcc034dc4522a2d8d9fa6c9 Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:48 +0300 Subject: [PATCH 06/12] qed: Add LL2 slowpath handling For iWARP unaligned MPA flow, a slowpath event of flushing an MPA connection that entered an unaligned state is required. The flush ramrod is received on the ll2 queue, and a pre-registered callback function is called to handle the flush event. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_ll2.c | 40 +++++++++++++++++++++-- include/linux/qed/qed_ll2_if.h | 5 +++ 2 files changed, 43 insertions(+), 2 deletions(-) diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c index 8eb9645c880d..047f556ca62e 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_ll2.c +++ b/drivers/net/ethernet/qlogic/qed/qed_ll2.c @@ -422,6 +422,41 @@ static void qed_ll2_rxq_parse_reg(struct qed_hwfn *p_hwfn, data->u.placement_offset = p_cqe->rx_cqe_fp.placement_offset; } +static int +qed_ll2_handle_slowpath(struct qed_hwfn *p_hwfn, + struct qed_ll2_info *p_ll2_conn, + union core_rx_cqe_union *p_cqe, + unsigned long *p_lock_flags) +{ + struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue; + struct core_rx_slow_path_cqe *sp_cqe; + + sp_cqe = &p_cqe->rx_cqe_sp; + if (sp_cqe->ramrod_cmd_id != CORE_RAMROD_RX_QUEUE_FLUSH) { + DP_NOTICE(p_hwfn, + "LL2 - unexpected Rx CQE slowpath ramrod_cmd_id:%d\n", + sp_cqe->ramrod_cmd_id); + return -EINVAL; + } + + if (!p_ll2_conn->cbs.slowpath_cb) { + DP_NOTICE(p_hwfn, + "LL2 - received RX_QUEUE_FLUSH but no callback was provided\n"); + return -EINVAL; + } + + spin_unlock_irqrestore(&p_rx->lock, *p_lock_flags); + + p_ll2_conn->cbs.slowpath_cb(p_ll2_conn->cbs.cookie, + p_ll2_conn->my_id, + le32_to_cpu(sp_cqe->opaque_data.data[0]), + le32_to_cpu(sp_cqe->opaque_data.data[1])); + + spin_lock_irqsave(&p_rx->lock, *p_lock_flags); + + return 0; +} + static int qed_ll2_rxq_handle_completion(struct qed_hwfn *p_hwfn, struct qed_ll2_info *p_ll2_conn, @@ -495,8 +530,8 @@ static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie) switch (cqe->rx_cqe_sp.type) { case CORE_RX_CQE_TYPE_SLOW_PATH: - DP_NOTICE(p_hwfn, "LL2 - unexpected Rx CQE slowpath\n"); - rc = -EINVAL; + rc = qed_ll2_handle_slowpath(p_hwfn, p_ll2_conn, + cqe, &flags); break; case CORE_RX_CQE_TYPE_GSI_OFFLOAD: case CORE_RX_CQE_TYPE_REGULAR: @@ -1214,6 +1249,7 @@ qed_ll2_set_cbs(struct qed_ll2_info *p_ll2_info, const struct qed_ll2_cbs *cbs) p_ll2_info->cbs.rx_release_cb = cbs->rx_release_cb; p_ll2_info->cbs.tx_comp_cb = cbs->tx_comp_cb; p_ll2_info->cbs.tx_release_cb = cbs->tx_release_cb; + p_ll2_info->cbs.slowpath_cb = cbs->slowpath_cb; p_ll2_info->cbs.cookie = cbs->cookie; return 0; diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h index 95fdf02a3bbe..e755954d85fd 100644 --- a/include/linux/qed/qed_ll2_if.h +++ b/include/linux/qed/qed_ll2_if.h @@ -151,11 +151,16 @@ void (*qed_ll2_release_tx_packet_cb)(void *cxt, dma_addr_t first_frag_addr, bool b_last_fragment, bool b_last_packet); +typedef +void (*qed_ll2_slowpath_cb)(void *cxt, u8 connection_handle, + u32 opaque_data_0, u32 opaque_data_1); + struct qed_ll2_cbs { qed_ll2_complete_rx_packet_cb rx_comp_cb; qed_ll2_release_rx_packet_cb rx_release_cb; qed_ll2_complete_tx_packet_cb tx_comp_cb; qed_ll2_release_tx_packet_cb tx_release_cb; + qed_ll2_slowpath_cb slowpath_cb; void *cookie; }; From ae3488ff37dc4f21985111f442d26a8805e56d45 Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:49 +0300 Subject: [PATCH 07/12] qed: Add ll2 connection for processing unaligned MPA packets This patch adds only the establishment and termination of the ll2 connection that handles unaligned MPA packets. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_iwarp.c | 65 +++++++++++++++++++++ drivers/net/ethernet/qlogic/qed/qed_iwarp.h | 1 + 2 files changed, 66 insertions(+) diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c index 8fc9c811f6e3..f413621a67b0 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c @@ -1713,6 +1713,19 @@ qed_iwarp_parse_rx_pkt(struct qed_hwfn *p_hwfn, return 0; } +/* fpdu can be fragmented over maximum 3 bds: header, partial mpa, unaligned */ +#define QED_IWARP_MAX_BDS_PER_FPDU 3 +static void +qed_iwarp_ll2_comp_mpa_pkt(void *cxt, struct qed_ll2_comp_rx_data *data) +{ + struct qed_iwarp_info *iwarp_info; + struct qed_hwfn *p_hwfn = cxt; + + iwarp_info = &p_hwfn->p_rdma_info->iwarp; + qed_iwarp_ll2_post_rx(p_hwfn, data->cookie, + iwarp_info->ll2_mpa_handle); +} + static void qed_iwarp_ll2_comp_syn_pkt(void *cxt, struct qed_ll2_comp_rx_data *data) { @@ -1877,6 +1890,13 @@ static void qed_iwarp_ll2_rel_tx_pkt(void *cxt, u8 connection_handle, kfree(buffer); } +void +qed_iwarp_ll2_slowpath(void *cxt, + u8 connection_handle, + u32 opaque_data_0, u32 opaque_data_1) +{ +} + static int qed_iwarp_ll2_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt) { struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp; @@ -1902,6 +1922,16 @@ static int qed_iwarp_ll2_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt) iwarp_info->ll2_ooo_handle = QED_IWARP_HANDLE_INVAL; } + if (iwarp_info->ll2_mpa_handle != QED_IWARP_HANDLE_INVAL) { + rc = qed_ll2_terminate_connection(p_hwfn, + iwarp_info->ll2_mpa_handle); + if (rc) + DP_INFO(p_hwfn, "Failed to terminate mpa connection\n"); + + qed_ll2_release_connection(p_hwfn, iwarp_info->ll2_mpa_handle); + iwarp_info->ll2_mpa_handle = QED_IWARP_HANDLE_INVAL; + } + qed_llh_remove_mac_filter(p_hwfn, p_ptt, p_hwfn->p_rdma_info->iwarp.mac_addr); return rc; @@ -1953,12 +1983,14 @@ qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn, struct qed_iwarp_info *iwarp_info; struct qed_ll2_acquire_data data; struct qed_ll2_cbs cbs; + u32 mpa_buff_size; u16 n_ooo_bufs; int rc = 0; iwarp_info = &p_hwfn->p_rdma_info->iwarp; iwarp_info->ll2_syn_handle = QED_IWARP_HANDLE_INVAL; iwarp_info->ll2_ooo_handle = QED_IWARP_HANDLE_INVAL; + iwarp_info->ll2_mpa_handle = QED_IWARP_HANDLE_INVAL; iwarp_info->max_mtu = params->max_mtu; @@ -2029,6 +2061,39 @@ qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn, if (rc) goto err; + /* Start Unaligned MPA connection */ + cbs.rx_comp_cb = qed_iwarp_ll2_comp_mpa_pkt; + cbs.slowpath_cb = qed_iwarp_ll2_slowpath; + + memset(&data, 0, sizeof(data)); + data.input.conn_type = QED_LL2_TYPE_IWARP; + data.input.mtu = params->max_mtu; + /* FW requires that once a packet arrives OOO, it must have at + * least 2 rx buffers available on the unaligned connection + * for handling the case that it is a partial fpdu. + */ + data.input.rx_num_desc = n_ooo_bufs * 2; + data.input.tx_num_desc = data.input.rx_num_desc; + data.input.tx_max_bds_per_packet = QED_IWARP_MAX_BDS_PER_FPDU; + data.p_connection_handle = &iwarp_info->ll2_mpa_handle; + data.input.secondary_queue = true; + data.cbs = &cbs; + + rc = qed_ll2_acquire_connection(p_hwfn, &data); + if (rc) + goto err; + + rc = qed_ll2_establish_connection(p_hwfn, iwarp_info->ll2_mpa_handle); + if (rc) + goto err; + + mpa_buff_size = QED_IWARP_MAX_BUF_SIZE(params->max_mtu); + rc = qed_iwarp_ll2_alloc_buffers(p_hwfn, + data.input.rx_num_desc, + mpa_buff_size, + iwarp_info->ll2_mpa_handle); + if (rc) + goto err; return rc; err: qed_iwarp_ll2_stop(p_hwfn, p_ptt); diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h index 9e2bfde894df..9d33a1fa1758 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h @@ -73,6 +73,7 @@ struct qed_iwarp_info { u8 tcp_flags; u8 ll2_syn_handle; u8 ll2_ooo_handle; + u8 ll2_mpa_handle; u8 peer2peer; enum mpa_negotiation_mode mpa_rev; enum mpa_rtr_type rtr_type; From fcb39f6c10b24d2d16d4c2bdb4c256bc21b8a131 Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:50 +0300 Subject: [PATCH 08/12] qed: Add mpa buffer descriptors for storing and processing mpa fpdus The mpa buff is a descriptor for iwarp ll2 buffers that contains additional information required for aligining fpdu's. In some cases, an additional packet will arrive which will complete the alignment of a fpdu, but we won't be able to post the fpdu due to insufficient place on the tx ring. In this case we can't loose the data and require storing it for later. Processing is therefore done in two places, during rx completion, where we initialize a mpa buffer descriptor and add it to the pending list, and during tx-completion, since we free up an entry in the tx chain we can process any pending mpa packets. The mpa buff descriptors are pre-allocated since we have to ensure that we won't reach a state where we can't store an incoming unaligned packet. All packets received on the ll2 MUST be processed by the driver at some stage. Since they are preallocated, we hold a free list. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_iwarp.c | 116 ++++++++++++++++++++ drivers/net/ethernet/qlogic/qed/qed_iwarp.h | 11 ++ 2 files changed, 127 insertions(+) diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c index f413621a67b0..efd4861c72e2 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c @@ -1415,7 +1415,10 @@ int qed_iwarp_alloc(struct qed_hwfn *p_hwfn) void qed_iwarp_resc_free(struct qed_hwfn *p_hwfn) { + struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp; + qed_rdma_bmap_free(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, 1); + kfree(iwarp_info->mpa_bufs); } int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams) @@ -1715,13 +1718,104 @@ qed_iwarp_parse_rx_pkt(struct qed_hwfn *p_hwfn, /* fpdu can be fragmented over maximum 3 bds: header, partial mpa, unaligned */ #define QED_IWARP_MAX_BDS_PER_FPDU 3 +static void +qed_iwarp_mpa_get_data(struct qed_hwfn *p_hwfn, + struct unaligned_opaque_data *curr_pkt, + u32 opaque_data0, u32 opaque_data1) +{ + u64 opaque_data; + + opaque_data = HILO_64(opaque_data1, opaque_data0); + *curr_pkt = *((struct unaligned_opaque_data *)&opaque_data); + + curr_pkt->first_mpa_offset = curr_pkt->tcp_payload_offset + + le16_to_cpu(curr_pkt->first_mpa_offset); + curr_pkt->cid = le32_to_cpu(curr_pkt->cid); +} + +/* This function is called when an unaligned or incomplete MPA packet arrives + * driver needs to align the packet, perhaps using previous data and send + * it down to FW once it is aligned. + */ +static int +qed_iwarp_process_mpa_pkt(struct qed_hwfn *p_hwfn, + struct qed_iwarp_ll2_mpa_buf *mpa_buf) +{ + struct qed_iwarp_ll2_buff *buf = mpa_buf->ll2_buf; + int rc = -EINVAL; + + qed_iwarp_ll2_post_rx(p_hwfn, + buf, + p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle); + return rc; +} + +static void qed_iwarp_process_pending_pkts(struct qed_hwfn *p_hwfn) +{ + struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp; + struct qed_iwarp_ll2_mpa_buf *mpa_buf = NULL; + int rc; + + while (!list_empty(&iwarp_info->mpa_buf_pending_list)) { + mpa_buf = list_first_entry(&iwarp_info->mpa_buf_pending_list, + struct qed_iwarp_ll2_mpa_buf, + list_entry); + + rc = qed_iwarp_process_mpa_pkt(p_hwfn, mpa_buf); + + /* busy means break and continue processing later, don't + * remove the buf from the pending list. + */ + if (rc == -EBUSY) + break; + + list_del(&mpa_buf->list_entry); + list_add_tail(&mpa_buf->list_entry, &iwarp_info->mpa_buf_list); + + if (rc) { /* different error, don't continue */ + DP_NOTICE(p_hwfn, "process pkts failed rc=%d\n", rc); + break; + } + } +} + static void qed_iwarp_ll2_comp_mpa_pkt(void *cxt, struct qed_ll2_comp_rx_data *data) { + struct qed_iwarp_ll2_mpa_buf *mpa_buf; struct qed_iwarp_info *iwarp_info; struct qed_hwfn *p_hwfn = cxt; iwarp_info = &p_hwfn->p_rdma_info->iwarp; + mpa_buf = list_first_entry(&iwarp_info->mpa_buf_list, + struct qed_iwarp_ll2_mpa_buf, list_entry); + if (!mpa_buf) { + DP_ERR(p_hwfn, "No free mpa buf\n"); + goto err; + } + + list_del(&mpa_buf->list_entry); + qed_iwarp_mpa_get_data(p_hwfn, &mpa_buf->data, + data->opaque_data_0, data->opaque_data_1); + + DP_VERBOSE(p_hwfn, + QED_MSG_RDMA, + "LL2 MPA CompRx payload_len:0x%x\tfirst_mpa_offset:0x%x\ttcp_payload_offset:0x%x\tflags:0x%x\tcid:0x%x\n", + data->length.packet_length, mpa_buf->data.first_mpa_offset, + mpa_buf->data.tcp_payload_offset, mpa_buf->data.flags, + mpa_buf->data.cid); + + mpa_buf->ll2_buf = data->cookie; + mpa_buf->tcp_payload_len = data->length.packet_length - + mpa_buf->data.first_mpa_offset; + mpa_buf->data.first_mpa_offset += data->u.placement_offset; + mpa_buf->placement_offset = data->u.placement_offset; + + list_add_tail(&mpa_buf->list_entry, &iwarp_info->mpa_buf_pending_list); + + qed_iwarp_process_pending_pkts(p_hwfn); + return; +err: qed_iwarp_ll2_post_rx(p_hwfn, data->cookie, iwarp_info->ll2_mpa_handle); } @@ -1872,6 +1966,11 @@ static void qed_iwarp_ll2_comp_tx_pkt(void *cxt, u8 connection_handle, /* this was originally an rx packet, post it back */ qed_iwarp_ll2_post_rx(p_hwfn, buffer, connection_handle); + + if (connection_handle == p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle) + qed_iwarp_process_pending_pkts(p_hwfn); + + return; } static void qed_iwarp_ll2_rel_tx_pkt(void *cxt, u8 connection_handle, @@ -1986,6 +2085,7 @@ qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn, u32 mpa_buff_size; u16 n_ooo_bufs; int rc = 0; + int i; iwarp_info = &p_hwfn->p_rdma_info->iwarp; iwarp_info->ll2_syn_handle = QED_IWARP_HANDLE_INVAL; @@ -2094,6 +2194,22 @@ qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn, iwarp_info->ll2_mpa_handle); if (rc) goto err; + /* The mpa_bufs array serves for pending RX packets received on the + * mpa ll2 that don't have place on the tx ring and require later + * processing. We can't fail on allocation of such a struct therefore + * we allocate enough to take care of all rx packets + */ + iwarp_info->mpa_bufs = kcalloc(data.input.rx_num_desc, + sizeof(*iwarp_info->mpa_bufs), + GFP_KERNEL); + if (!iwarp_info->mpa_bufs) + goto err; + + INIT_LIST_HEAD(&iwarp_info->mpa_buf_pending_list); + INIT_LIST_HEAD(&iwarp_info->mpa_buf_list); + for (i = 0; i < data.input.rx_num_desc; i++) + list_add_tail(&iwarp_info->mpa_bufs[i].list_entry, + &iwarp_info->mpa_buf_list); return rc; err: qed_iwarp_ll2_stop(p_hwfn, p_ptt); diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h index 9d33a1fa1758..2c53fe46345c 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h @@ -60,10 +60,20 @@ struct qed_iwarp_ll2_buff { u32 buff_size; }; +struct qed_iwarp_ll2_mpa_buf { + struct list_head list_entry; + struct qed_iwarp_ll2_buff *ll2_buf; + struct unaligned_opaque_data data; + u16 tcp_payload_len; + u8 placement_offset; +}; + struct qed_iwarp_info { struct list_head listen_list; /* qed_iwarp_listener */ struct list_head ep_list; /* qed_iwarp_ep */ struct list_head ep_free_list; /* pre-allocated ep's */ + struct list_head mpa_buf_list; /* list of mpa_bufs */ + struct list_head mpa_buf_pending_list; spinlock_t iw_lock; /* for iwarp resources */ spinlock_t qp_lock; /* for teardown races */ u32 rcv_wnd_scale; @@ -77,6 +87,7 @@ struct qed_iwarp_info { u8 peer2peer; enum mpa_negotiation_mode mpa_rev; enum mpa_rtr_type rtr_type; + struct qed_iwarp_ll2_mpa_buf *mpa_bufs; }; enum qed_iwarp_ep_state { From 469981b17a4f8ddac91837bd74ebc98578f2ddbf Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:51 +0300 Subject: [PATCH 09/12] qed: Add unaligned and packed packet processing The fpdu data structure is preallocated per connection. Each connection stores the current status of the connection: either nothing pending, or there is a partial fpdu that is waiting for the rest of the fpdu (incomplete bytes != 0). The same structure is also used for splitting a packet when there are packed fpdus. The structure is initialized with all data required for sending the fpdu back to the FW. A fpdu will always be spanned across a maximum of 3 tx bds. One for the header, one for the partial fdpu received and one for the remainder (unaligned) packet. In case of packed fpdu's, two fragments are used, one for the header and one for the data. Corner cases are not handled in the patch for clarity, and will be added as a separate patch. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_iwarp.c | 257 ++++++++++++++++++++ drivers/net/ethernet/qlogic/qed/qed_iwarp.h | 13 + 2 files changed, 270 insertions(+) diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c index efd4861c72e2..83b147fdacde 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c @@ -1419,6 +1419,7 @@ void qed_iwarp_resc_free(struct qed_hwfn *p_hwfn) qed_rdma_bmap_free(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, 1); kfree(iwarp_info->mpa_bufs); + kfree(iwarp_info->partial_fpdus); } int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams) @@ -1716,8 +1717,170 @@ qed_iwarp_parse_rx_pkt(struct qed_hwfn *p_hwfn, return 0; } +static struct qed_iwarp_fpdu *qed_iwarp_get_curr_fpdu(struct qed_hwfn *p_hwfn, + u16 cid) +{ + struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp; + struct qed_iwarp_fpdu *partial_fpdu; + u32 idx; + + idx = cid - qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_IWARP); + if (idx >= iwarp_info->max_num_partial_fpdus) { + DP_ERR(p_hwfn, "Invalid cid %x max_num_partial_fpdus=%x\n", cid, + iwarp_info->max_num_partial_fpdus); + return NULL; + } + + partial_fpdu = &iwarp_info->partial_fpdus[idx]; + + return partial_fpdu; +} + +enum qed_iwarp_mpa_pkt_type { + QED_IWARP_MPA_PKT_PACKED, + QED_IWARP_MPA_PKT_PARTIAL, + QED_IWARP_MPA_PKT_UNALIGNED +}; + +#define QED_IWARP_MPA_FPDU_LENGTH_SIZE (2) +#define QED_IWARP_MPA_CRC32_DIGEST_SIZE (4) + +/* Pad to multiple of 4 */ +#define QED_IWARP_PDU_DATA_LEN_WITH_PAD(data_len) ALIGN(data_len, 4) +#define QED_IWARP_FPDU_LEN_WITH_PAD(_mpa_len) \ + (QED_IWARP_PDU_DATA_LEN_WITH_PAD((_mpa_len) + \ + QED_IWARP_MPA_FPDU_LENGTH_SIZE) + \ + QED_IWARP_MPA_CRC32_DIGEST_SIZE) + /* fpdu can be fragmented over maximum 3 bds: header, partial mpa, unaligned */ #define QED_IWARP_MAX_BDS_PER_FPDU 3 + +char *pkt_type_str[] = { + "QED_IWARP_MPA_PKT_PACKED", + "QED_IWARP_MPA_PKT_PARTIAL", + "QED_IWARP_MPA_PKT_UNALIGNED" +}; + +static enum qed_iwarp_mpa_pkt_type +qed_iwarp_mpa_classify(struct qed_hwfn *p_hwfn, + struct qed_iwarp_fpdu *fpdu, + u16 tcp_payload_len, u8 *mpa_data) +{ + enum qed_iwarp_mpa_pkt_type pkt_type; + u16 mpa_len; + + if (fpdu->incomplete_bytes) { + pkt_type = QED_IWARP_MPA_PKT_UNALIGNED; + goto out; + } + + mpa_len = ntohs(*((u16 *)(mpa_data))); + fpdu->fpdu_length = QED_IWARP_FPDU_LEN_WITH_PAD(mpa_len); + + if (fpdu->fpdu_length <= tcp_payload_len) + pkt_type = QED_IWARP_MPA_PKT_PACKED; + else + pkt_type = QED_IWARP_MPA_PKT_PARTIAL; + +out: + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, + "MPA_ALIGN: %s: fpdu_length=0x%x tcp_payload_len:0x%x\n", + pkt_type_str[pkt_type], fpdu->fpdu_length, tcp_payload_len); + + return pkt_type; +} + +static void +qed_iwarp_init_fpdu(struct qed_iwarp_ll2_buff *buf, + struct qed_iwarp_fpdu *fpdu, + struct unaligned_opaque_data *pkt_data, + u16 tcp_payload_size, u8 placement_offset) +{ + fpdu->mpa_buf = buf; + fpdu->pkt_hdr = buf->data_phys_addr + placement_offset; + fpdu->pkt_hdr_size = pkt_data->tcp_payload_offset; + fpdu->mpa_frag = buf->data_phys_addr + pkt_data->first_mpa_offset; + fpdu->mpa_frag_virt = (u8 *)(buf->data) + pkt_data->first_mpa_offset; + + if (tcp_payload_size < fpdu->fpdu_length) + fpdu->incomplete_bytes = fpdu->fpdu_length - tcp_payload_size; + else + fpdu->incomplete_bytes = 0; /* complete fpdu */ + + fpdu->mpa_frag_len = fpdu->fpdu_length - fpdu->incomplete_bytes; +} + +static int +qed_iwarp_send_fpdu(struct qed_hwfn *p_hwfn, + struct qed_iwarp_fpdu *fpdu, + struct unaligned_opaque_data *curr_pkt, + struct qed_iwarp_ll2_buff *buf, + u16 tcp_payload_size, enum qed_iwarp_mpa_pkt_type pkt_type) +{ + struct qed_ll2_tx_pkt_info tx_pkt; + u8 ll2_handle; + int rc; + + memset(&tx_pkt, 0, sizeof(tx_pkt)); + + /* An unaligned packet means it's split over two tcp segments. So the + * complete packet requires 3 bds, one for the header, one for the + * part of the fpdu of the first tcp segment, and the last fragment + * will point to the remainder of the fpdu. A packed pdu, requires only + * two bds, one for the header and one for the data. + */ + tx_pkt.num_of_bds = (pkt_type == QED_IWARP_MPA_PKT_UNALIGNED) ? 3 : 2; + tx_pkt.tx_dest = QED_LL2_TX_DEST_LB; + tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2; /* offset in words */ + + /* Send the mpa_buf only with the last fpdu (in case of packed) */ + if (pkt_type == QED_IWARP_MPA_PKT_UNALIGNED || + tcp_payload_size <= fpdu->fpdu_length) + tx_pkt.cookie = fpdu->mpa_buf; + + tx_pkt.first_frag = fpdu->pkt_hdr; + tx_pkt.first_frag_len = fpdu->pkt_hdr_size; + tx_pkt.enable_ip_cksum = true; + tx_pkt.enable_l4_cksum = true; + tx_pkt.calc_ip_len = true; + /* vlan overload with enum iwarp_ll2_tx_queues */ + tx_pkt.vlan = IWARP_LL2_ALIGNED_TX_QUEUE; + + ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle; + + /* Set first fragment to header */ + rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true); + if (rc) + goto out; + + /* Set second fragment to first part of packet */ + rc = qed_ll2_set_fragment_of_tx_packet(p_hwfn, ll2_handle, + fpdu->mpa_frag, + fpdu->mpa_frag_len); + if (rc) + goto out; + + if (!fpdu->incomplete_bytes) + goto out; + + /* Set third fragment to second part of the packet */ + rc = qed_ll2_set_fragment_of_tx_packet(p_hwfn, + ll2_handle, + buf->data_phys_addr + + curr_pkt->first_mpa_offset, + fpdu->incomplete_bytes); +out: + DP_VERBOSE(p_hwfn, + QED_MSG_RDMA, + "MPA_ALIGN: Sent FPDU num_bds=%d first_frag_len=%x, mpa_frag_len=0x%x, incomplete_bytes:0x%x rc=%d\n", + tx_pkt.num_of_bds, + tx_pkt.first_frag_len, + fpdu->mpa_frag_len, + fpdu->incomplete_bytes, rc); + + return rc; +} + static void qed_iwarp_mpa_get_data(struct qed_hwfn *p_hwfn, struct unaligned_opaque_data *curr_pkt, @@ -1741,9 +1904,79 @@ static int qed_iwarp_process_mpa_pkt(struct qed_hwfn *p_hwfn, struct qed_iwarp_ll2_mpa_buf *mpa_buf) { + struct unaligned_opaque_data *curr_pkt = &mpa_buf->data; struct qed_iwarp_ll2_buff *buf = mpa_buf->ll2_buf; + enum qed_iwarp_mpa_pkt_type pkt_type; + struct qed_iwarp_fpdu *fpdu; int rc = -EINVAL; + u8 *mpa_data; + fpdu = qed_iwarp_get_curr_fpdu(p_hwfn, curr_pkt->cid & 0xffff); + if (!fpdu) { /* something corrupt with cid, post rx back */ + DP_ERR(p_hwfn, "Invalid cid, drop and post back to rx cid=%x\n", + curr_pkt->cid); + goto err; + } + + do { + mpa_data = ((u8 *)(buf->data) + curr_pkt->first_mpa_offset); + + pkt_type = qed_iwarp_mpa_classify(p_hwfn, fpdu, + mpa_buf->tcp_payload_len, + mpa_data); + + switch (pkt_type) { + case QED_IWARP_MPA_PKT_PARTIAL: + qed_iwarp_init_fpdu(buf, fpdu, + curr_pkt, + mpa_buf->tcp_payload_len, + mpa_buf->placement_offset); + + mpa_buf->tcp_payload_len = 0; + break; + case QED_IWARP_MPA_PKT_PACKED: + qed_iwarp_init_fpdu(buf, fpdu, + curr_pkt, + mpa_buf->tcp_payload_len, + mpa_buf->placement_offset); + + rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf, + mpa_buf->tcp_payload_len, + pkt_type); + if (rc) { + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, + "Can't send FPDU:reset rc=%d\n", rc); + memset(fpdu, 0, sizeof(*fpdu)); + break; + } + + mpa_buf->tcp_payload_len -= fpdu->fpdu_length; + curr_pkt->first_mpa_offset += fpdu->fpdu_length; + break; + case QED_IWARP_MPA_PKT_UNALIGNED: + rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf, + mpa_buf->tcp_payload_len, + pkt_type); + if (rc) { + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, + "Can't send FPDU:delay rc=%d\n", rc); + /* don't reset fpdu -> we need it for next + * classify + */ + break; + } + + mpa_buf->tcp_payload_len -= fpdu->incomplete_bytes; + curr_pkt->first_mpa_offset += fpdu->incomplete_bytes; + /* The framed PDU was sent - no more incomplete bytes */ + fpdu->incomplete_bytes = 0; + break; + } + } while (mpa_buf->tcp_payload_len && !rc); + + return rc; + +err: qed_iwarp_ll2_post_rx(p_hwfn, buf, p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle); @@ -1989,11 +2222,27 @@ static void qed_iwarp_ll2_rel_tx_pkt(void *cxt, u8 connection_handle, kfree(buffer); } +/* The only slowpath for iwarp ll2 is unalign flush. When this completion + * is received, need to reset the FPDU. + */ void qed_iwarp_ll2_slowpath(void *cxt, u8 connection_handle, u32 opaque_data_0, u32 opaque_data_1) { + struct unaligned_opaque_data unalign_data; + struct qed_hwfn *p_hwfn = cxt; + struct qed_iwarp_fpdu *fpdu; + + qed_iwarp_mpa_get_data(p_hwfn, &unalign_data, + opaque_data_0, opaque_data_1); + + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "(0x%x) Flush fpdu\n", + unalign_data.cid); + + fpdu = qed_iwarp_get_curr_fpdu(p_hwfn, (u16)unalign_data.cid); + if (fpdu) + memset(fpdu, 0, sizeof(*fpdu)); } static int qed_iwarp_ll2_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt) @@ -2194,6 +2443,14 @@ qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn, iwarp_info->ll2_mpa_handle); if (rc) goto err; + + iwarp_info->partial_fpdus = kcalloc((u16)p_hwfn->p_rdma_info->num_qps, + sizeof(*iwarp_info->partial_fpdus), + GFP_KERNEL); + if (!iwarp_info->partial_fpdus) + goto err; + + iwarp_info->max_num_partial_fpdus = (u16)p_hwfn->p_rdma_info->num_qps; /* The mpa_bufs array serves for pending RX packets received on the * mpa ll2 that don't have place on the tx ring and require later * processing. We can't fail on allocation of such a struct therefore diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h index 2c53fe46345c..858755cafd2b 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h @@ -68,6 +68,17 @@ struct qed_iwarp_ll2_mpa_buf { u8 placement_offset; }; +struct qed_iwarp_fpdu { + struct qed_iwarp_ll2_buff *mpa_buf; + void *mpa_frag_virt; + dma_addr_t mpa_frag; + dma_addr_t pkt_hdr; + u16 mpa_frag_len; + u16 fpdu_length; + u16 incomplete_bytes; + u8 pkt_hdr_size; +}; + struct qed_iwarp_info { struct list_head listen_list; /* qed_iwarp_listener */ struct list_head ep_list; /* qed_iwarp_ep */ @@ -87,7 +98,9 @@ struct qed_iwarp_info { u8 peer2peer; enum mpa_negotiation_mode mpa_rev; enum mpa_rtr_type rtr_type; + struct qed_iwarp_fpdu *partial_fpdus; struct qed_iwarp_ll2_mpa_buf *mpa_bufs; + u16 max_num_partial_fpdus; }; enum qed_iwarp_ep_state { From d531038eeb6dd25dbf88402f932bf0ea524de82e Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:52 +0300 Subject: [PATCH 10/12] qed: Add support for freeing two ll2 buffers for corner cases When posting a packet on the ll2 tx, we can provide a cookie that will be returned upon tx completion. This cookie is the ll2 iwarp buffer which is then reposted to the rx ring. Part of the unaligned mpa flow is determining when a buffer can be reposted. Each buffer needs to be sent only once as a cookie for on the tx ring. In packed fpdu case, only the last packet will be sent with the buffer, meaning we need to handle the case that a cookie can be NULL on tx complete. In addition, when a fpdu splits over two buffers, but there are no more fpdus on the second buffer, two buffers need to be provided as a cookie. To avoid changing the ll2 interface to provide two cookies, we introduce a piggy buf pointer, relevant for iWARP only, that holds a pointer to a second buffer that needs to be released during tx completion. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_iwarp.c | 25 +++++++++++++++++++++ drivers/net/ethernet/qlogic/qed/qed_iwarp.h | 1 + 2 files changed, 26 insertions(+) diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c index 83b147fdacde..8b17369af9ef 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c @@ -1846,6 +1846,12 @@ qed_iwarp_send_fpdu(struct qed_hwfn *p_hwfn, /* vlan overload with enum iwarp_ll2_tx_queues */ tx_pkt.vlan = IWARP_LL2_ALIGNED_TX_QUEUE; + /* special case of unaligned packet and not packed, need to send + * both buffers as cookie to release. + */ + if (tcp_payload_size == fpdu->incomplete_bytes) + fpdu->mpa_buf->piggy_buf = buf; + ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle; /* Set first fragment to header */ @@ -2195,9 +2201,19 @@ static void qed_iwarp_ll2_comp_tx_pkt(void *cxt, u8 connection_handle, bool b_last_fragment, bool b_last_packet) { struct qed_iwarp_ll2_buff *buffer = cookie; + struct qed_iwarp_ll2_buff *piggy; struct qed_hwfn *p_hwfn = cxt; + if (!buffer) /* can happen in packed mpa unaligned... */ + return; + /* this was originally an rx packet, post it back */ + piggy = buffer->piggy_buf; + if (piggy) { + buffer->piggy_buf = NULL; + qed_iwarp_ll2_post_rx(p_hwfn, piggy, connection_handle); + } + qed_iwarp_ll2_post_rx(p_hwfn, buffer, connection_handle); if (connection_handle == p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle) @@ -2216,6 +2232,15 @@ static void qed_iwarp_ll2_rel_tx_pkt(void *cxt, u8 connection_handle, if (!buffer) return; + if (buffer->piggy_buf) { + dma_free_coherent(&p_hwfn->cdev->pdev->dev, + buffer->piggy_buf->buff_size, + buffer->piggy_buf->data, + buffer->piggy_buf->data_phys_addr); + + kfree(buffer->piggy_buf); + } + dma_free_coherent(&p_hwfn->cdev->pdev->dev, buffer->buff_size, buffer->data, buffer->data_phys_addr); diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h index 858755cafd2b..58db51af26bd 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h @@ -55,6 +55,7 @@ enum qed_iwarp_qp_state qed_roce2iwarp_state(enum qed_roce_qp_state state); #define QED_IWARP_HANDLE_INVAL (0xff) struct qed_iwarp_ll2_buff { + struct qed_iwarp_ll2_buff *piggy_buf; void *data; dma_addr_t data_phys_addr; u32 buff_size; From c7d1d839999476aac0d7e16732722285a9c30cce Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:53 +0300 Subject: [PATCH 11/12] qed: Add support for MPA header being split over two tcp packets There is a special case where an MPA header is split over to tcp packets, in this case we need to wait for the next packet to get the fpdu length. We use the incomplete_bytes to mark this fpdu as a "special" one which requires updating the length with the next packet Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_iwarp.c | 36 ++++++++++++++++++++- drivers/net/ethernet/qlogic/qed/qed_iwarp.h | 6 ++++ 2 files changed, 41 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c index 8b17369af9ef..299494225f44 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c @@ -1742,6 +1742,7 @@ enum qed_iwarp_mpa_pkt_type { QED_IWARP_MPA_PKT_UNALIGNED }; +#define QED_IWARP_INVALID_FPDU_LENGTH 0xffff #define QED_IWARP_MPA_FPDU_LENGTH_SIZE (2) #define QED_IWARP_MPA_CRC32_DIGEST_SIZE (4) @@ -1774,6 +1775,15 @@ qed_iwarp_mpa_classify(struct qed_hwfn *p_hwfn, goto out; } + /* special case of one byte remaining... + * lower byte will be read next packet + */ + if (tcp_payload_len == 1) { + fpdu->fpdu_length = *mpa_data << BITS_PER_BYTE; + pkt_type = QED_IWARP_MPA_PKT_PARTIAL; + goto out; + } + mpa_len = ntohs(*((u16 *)(mpa_data))); fpdu->fpdu_length = QED_IWARP_FPDU_LEN_WITH_PAD(mpa_len); @@ -1802,7 +1812,9 @@ qed_iwarp_init_fpdu(struct qed_iwarp_ll2_buff *buf, fpdu->mpa_frag = buf->data_phys_addr + pkt_data->first_mpa_offset; fpdu->mpa_frag_virt = (u8 *)(buf->data) + pkt_data->first_mpa_offset; - if (tcp_payload_size < fpdu->fpdu_length) + if (tcp_payload_size == 1) + fpdu->incomplete_bytes = QED_IWARP_INVALID_FPDU_LENGTH; + else if (tcp_payload_size < fpdu->fpdu_length) fpdu->incomplete_bytes = fpdu->fpdu_length - tcp_payload_size; else fpdu->incomplete_bytes = 0; /* complete fpdu */ @@ -1810,6 +1822,27 @@ qed_iwarp_init_fpdu(struct qed_iwarp_ll2_buff *buf, fpdu->mpa_frag_len = fpdu->fpdu_length - fpdu->incomplete_bytes; } +static void +qed_iwarp_update_fpdu_length(struct qed_hwfn *p_hwfn, + struct qed_iwarp_fpdu *fpdu, u8 *mpa_data) +{ + u16 mpa_len; + + /* Update incomplete packets if needed */ + if (fpdu->incomplete_bytes == QED_IWARP_INVALID_FPDU_LENGTH) { + /* Missing lower byte is now available */ + mpa_len = fpdu->fpdu_length | *mpa_data; + fpdu->fpdu_length = QED_IWARP_FPDU_LEN_WITH_PAD(mpa_len); + fpdu->mpa_frag_len = fpdu->fpdu_length; + /* one byte of hdr */ + fpdu->incomplete_bytes = fpdu->fpdu_length - 1; + DP_VERBOSE(p_hwfn, + QED_MSG_RDMA, + "MPA_ALIGN: Partial header mpa_len=%x fpdu_length=%x incomplete_bytes=%x\n", + mpa_len, fpdu->fpdu_length, fpdu->incomplete_bytes); + } +} + static int qed_iwarp_send_fpdu(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu, @@ -1960,6 +1993,7 @@ qed_iwarp_process_mpa_pkt(struct qed_hwfn *p_hwfn, curr_pkt->first_mpa_offset += fpdu->fpdu_length; break; case QED_IWARP_MPA_PKT_UNALIGNED: + qed_iwarp_update_fpdu_length(p_hwfn, fpdu, mpa_data); rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf, mpa_buf->tcp_payload_len, pkt_type); diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h index 58db51af26bd..c58793a47774 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h @@ -69,6 +69,12 @@ struct qed_iwarp_ll2_mpa_buf { u8 placement_offset; }; +/* In some cases a fpdu will arrive with only one byte of the header, in this + * case the fpdu_length will be partial (contain only higher byte and + * incomplete bytes will contain the invalid value + */ +#define QED_IWARP_INVALID_INCOMPLETE_BYTES 0xffff + struct qed_iwarp_fpdu { struct qed_iwarp_ll2_buff *mpa_buf; void *mpa_frag_virt; From 1e28eaad07ea1e2d6537586529e87cbc1d698ffd Mon Sep 17 00:00:00 2001 From: Michal Kalderon Date: Mon, 9 Oct 2017 12:37:54 +0300 Subject: [PATCH 12/12] qed: Add iWARP support for fpdu spanned over more than two tcp packets We continue to maintain a maximum of three buffers per fpdu, to ensure that there are enough buffers for additional unaligned mpa packets. To support this, if a fpdu is split over more than two tcp packets, we use an intermediate buffer to copy the data to the previous buffer, then we can release the data. We need an intermediate buffer as the initial buffer partial packet could be located at the end of the packet, not leaving room for additional data. This is a corner case, and will usually not be the case. Signed-off-by: Michal Kalderon Signed-off-by: Ariel Elior Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qed/qed_iwarp.c | 193 ++++++++++++++++++++ drivers/net/ethernet/qlogic/qed/qed_iwarp.h | 1 + 2 files changed, 194 insertions(+) diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c index 299494225f44..b2b1f87864ef 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c @@ -1420,6 +1420,7 @@ void qed_iwarp_resc_free(struct qed_hwfn *p_hwfn) qed_rdma_bmap_free(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, 1); kfree(iwarp_info->mpa_bufs); kfree(iwarp_info->partial_fpdus); + kfree(iwarp_info->mpa_intermediate_buf); } int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams) @@ -1762,6 +1763,11 @@ char *pkt_type_str[] = { "QED_IWARP_MPA_PKT_UNALIGNED" }; +static int +qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn, + struct qed_iwarp_fpdu *fpdu, + struct qed_iwarp_ll2_buff *buf); + static enum qed_iwarp_mpa_pkt_type qed_iwarp_mpa_classify(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu, @@ -1822,6 +1828,68 @@ qed_iwarp_init_fpdu(struct qed_iwarp_ll2_buff *buf, fpdu->mpa_frag_len = fpdu->fpdu_length - fpdu->incomplete_bytes; } +static int +qed_iwarp_cp_pkt(struct qed_hwfn *p_hwfn, + struct qed_iwarp_fpdu *fpdu, + struct unaligned_opaque_data *pkt_data, + struct qed_iwarp_ll2_buff *buf, u16 tcp_payload_size) +{ + u8 *tmp_buf = p_hwfn->p_rdma_info->iwarp.mpa_intermediate_buf; + int rc; + + /* need to copy the data from the partial packet stored in fpdu + * to the new buf, for this we also need to move the data currently + * placed on the buf. The assumption is that the buffer is big enough + * since fpdu_length <= mss, we use an intermediate buffer since + * we may need to copy the new data to an overlapping location + */ + if ((fpdu->mpa_frag_len + tcp_payload_size) > (u16)buf->buff_size) { + DP_ERR(p_hwfn, + "MPA ALIGN: Unexpected: buffer is not large enough for split fpdu buff_size = %d mpa_frag_len = %d, tcp_payload_size = %d, incomplete_bytes = %d\n", + buf->buff_size, fpdu->mpa_frag_len, + tcp_payload_size, fpdu->incomplete_bytes); + return -EINVAL; + } + + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, + "MPA ALIGN Copying fpdu: [%p, %d] [%p, %d]\n", + fpdu->mpa_frag_virt, fpdu->mpa_frag_len, + (u8 *)(buf->data) + pkt_data->first_mpa_offset, + tcp_payload_size); + + memcpy(tmp_buf, fpdu->mpa_frag_virt, fpdu->mpa_frag_len); + memcpy(tmp_buf + fpdu->mpa_frag_len, + (u8 *)(buf->data) + pkt_data->first_mpa_offset, + tcp_payload_size); + + rc = qed_iwarp_recycle_pkt(p_hwfn, fpdu, fpdu->mpa_buf); + if (rc) + return rc; + + /* If we managed to post the buffer copy the data to the new buffer + * o/w this will occur in the next round... + */ + memcpy((u8 *)(buf->data), tmp_buf, + fpdu->mpa_frag_len + tcp_payload_size); + + fpdu->mpa_buf = buf; + /* fpdu->pkt_hdr remains as is */ + /* fpdu->mpa_frag is overridden with new buf */ + fpdu->mpa_frag = buf->data_phys_addr; + fpdu->mpa_frag_virt = buf->data; + fpdu->mpa_frag_len += tcp_payload_size; + + fpdu->incomplete_bytes -= tcp_payload_size; + + DP_VERBOSE(p_hwfn, + QED_MSG_RDMA, + "MPA ALIGN: split fpdu buff_size = %d mpa_frag_len = %d, tcp_payload_size = %d, incomplete_bytes = %d\n", + buf->buff_size, fpdu->mpa_frag_len, tcp_payload_size, + fpdu->incomplete_bytes); + + return 0; +} + static void qed_iwarp_update_fpdu_length(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu, u8 *mpa_data) @@ -1843,6 +1911,90 @@ qed_iwarp_update_fpdu_length(struct qed_hwfn *p_hwfn, } } +#define QED_IWARP_IS_RIGHT_EDGE(_curr_pkt) \ + (GET_FIELD((_curr_pkt)->flags, \ + UNALIGNED_OPAQUE_DATA_PKT_REACHED_WIN_RIGHT_EDGE)) + +/* This function is used to recycle a buffer using the ll2 drop option. It + * uses the mechanism to ensure that all buffers posted to tx before this one + * were completed. The buffer sent here will be sent as a cookie in the tx + * completion function and can then be reposted to rx chain when done. The flow + * that requires this is the flow where a FPDU splits over more than 3 tcp + * segments. In this case the driver needs to re-post a rx buffer instead of + * the one received, but driver can't simply repost a buffer it copied from + * as there is a case where the buffer was originally a packed FPDU, and is + * partially posted to FW. Driver needs to ensure FW is done with it. + */ +static int +qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn, + struct qed_iwarp_fpdu *fpdu, + struct qed_iwarp_ll2_buff *buf) +{ + struct qed_ll2_tx_pkt_info tx_pkt; + u8 ll2_handle; + int rc; + + memset(&tx_pkt, 0, sizeof(tx_pkt)); + tx_pkt.num_of_bds = 1; + tx_pkt.tx_dest = QED_LL2_TX_DEST_DROP; + tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2; + tx_pkt.first_frag = fpdu->pkt_hdr; + tx_pkt.first_frag_len = fpdu->pkt_hdr_size; + buf->piggy_buf = NULL; + tx_pkt.cookie = buf; + + ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle; + + rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true); + if (rc) + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, + "Can't drop packet rc=%d\n", rc); + + DP_VERBOSE(p_hwfn, + QED_MSG_RDMA, + "MPA_ALIGN: send drop tx packet [%lx, 0x%x], buf=%p, rc=%d\n", + (unsigned long int)tx_pkt.first_frag, + tx_pkt.first_frag_len, buf, rc); + + return rc; +} + +static int +qed_iwarp_win_right_edge(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu) +{ + struct qed_ll2_tx_pkt_info tx_pkt; + u8 ll2_handle; + int rc; + + memset(&tx_pkt, 0, sizeof(tx_pkt)); + tx_pkt.num_of_bds = 1; + tx_pkt.tx_dest = QED_LL2_TX_DEST_LB; + tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2; + + tx_pkt.first_frag = fpdu->pkt_hdr; + tx_pkt.first_frag_len = fpdu->pkt_hdr_size; + tx_pkt.enable_ip_cksum = true; + tx_pkt.enable_l4_cksum = true; + tx_pkt.calc_ip_len = true; + /* vlan overload with enum iwarp_ll2_tx_queues */ + tx_pkt.vlan = IWARP_LL2_ALIGNED_RIGHT_TRIMMED_TX_QUEUE; + + ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle; + + rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true); + if (rc) + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, + "Can't send right edge rc=%d\n", rc); + DP_VERBOSE(p_hwfn, + QED_MSG_RDMA, + "MPA_ALIGN: Sent right edge FPDU num_bds=%d [%lx, 0x%x], rc=%d\n", + tx_pkt.num_of_bds, + (unsigned long int)tx_pkt.first_frag, + tx_pkt.first_frag_len, rc); + + return rc; +} + static int qed_iwarp_send_fpdu(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu, @@ -1971,6 +2123,20 @@ qed_iwarp_process_mpa_pkt(struct qed_hwfn *p_hwfn, mpa_buf->tcp_payload_len, mpa_buf->placement_offset); + if (!QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) { + mpa_buf->tcp_payload_len = 0; + break; + } + + rc = qed_iwarp_win_right_edge(p_hwfn, fpdu); + + if (rc) { + DP_VERBOSE(p_hwfn, QED_MSG_RDMA, + "Can't send FPDU:reset rc=%d\n", rc); + memset(fpdu, 0, sizeof(*fpdu)); + break; + } + mpa_buf->tcp_payload_len = 0; break; case QED_IWARP_MPA_PKT_PACKED: @@ -1994,6 +2160,28 @@ qed_iwarp_process_mpa_pkt(struct qed_hwfn *p_hwfn, break; case QED_IWARP_MPA_PKT_UNALIGNED: qed_iwarp_update_fpdu_length(p_hwfn, fpdu, mpa_data); + if (mpa_buf->tcp_payload_len < fpdu->incomplete_bytes) { + /* special handling of fpdu split over more + * than 2 segments + */ + if (QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) { + rc = qed_iwarp_win_right_edge(p_hwfn, + fpdu); + /* packet will be re-processed later */ + if (rc) + return rc; + } + + rc = qed_iwarp_cp_pkt(p_hwfn, fpdu, curr_pkt, + buf, + mpa_buf->tcp_payload_len); + if (rc) /* packet will be re-processed later */ + return rc; + + mpa_buf->tcp_payload_len = 0; + break; + } + rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf, mpa_buf->tcp_payload_len, pkt_type); @@ -2510,6 +2698,11 @@ qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn, goto err; iwarp_info->max_num_partial_fpdus = (u16)p_hwfn->p_rdma_info->num_qps; + + iwarp_info->mpa_intermediate_buf = kzalloc(mpa_buff_size, GFP_KERNEL); + if (!iwarp_info->mpa_intermediate_buf) + goto err; + /* The mpa_bufs array serves for pending RX packets received on the * mpa ll2 that don't have place on the tx ring and require later * processing. We can't fail on allocation of such a struct therefore diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h index c58793a47774..c1ecd743305f 100644 --- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h +++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h @@ -107,6 +107,7 @@ struct qed_iwarp_info { enum mpa_rtr_type rtr_type; struct qed_iwarp_fpdu *partial_fpdus; struct qed_iwarp_ll2_mpa_buf *mpa_bufs; + u8 *mpa_intermediate_buf; u16 max_num_partial_fpdus; };