1 /* 2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32 33 #include <linux/irq.h> 34 #include <net/xdp_sock_drv.h> 35 #include "en.h" 36 #include "en/txrx.h" 37 #include "en/xdp.h" 38 #include "en/xsk/rx.h" 39 #include "en/xsk/tx.h" 40 #include "en_accel/ktls_txrx.h" 41 42 static inline bool mlx5e_channel_no_affinity_change(struct mlx5e_channel *c) 43 { 44 int current_cpu = smp_processor_id(); 45 46 return cpumask_test_cpu(current_cpu, c->aff_mask); 47 } 48 49 static void mlx5e_handle_tx_dim(struct mlx5e_txqsq *sq) 50 { 51 struct mlx5e_sq_stats *stats = sq->stats; 52 struct dim_sample dim_sample = {}; 53 54 if (unlikely(!test_bit(MLX5E_SQ_STATE_DIM, &sq->state))) 55 return; 56 57 dim_update_sample(sq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample); 58 net_dim(sq->dim, &dim_sample); 59 } 60 61 static void mlx5e_handle_rx_dim(struct mlx5e_rq *rq) 62 { 63 struct mlx5e_rq_stats *stats = rq->stats; 64 struct dim_sample dim_sample = {}; 65 66 if (unlikely(!test_bit(MLX5E_RQ_STATE_DIM, &rq->state))) 67 return; 68 69 dim_update_sample(rq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample); 70 net_dim(rq->dim, &dim_sample); 71 } 72 73 void mlx5e_trigger_irq(struct mlx5e_icosq *sq) 74 { 75 struct mlx5_wq_cyc *wq = &sq->wq; 76 struct mlx5e_tx_wqe *nopwqe; 77 u16 pi = mlx5_wq_cyc_ctr2ix(wq, sq->pc); 78 79 sq->db.wqe_info[pi] = (struct mlx5e_icosq_wqe_info) { 80 .wqe_type = MLX5E_ICOSQ_WQE_NOP, 81 .num_wqebbs = 1, 82 }; 83 84 nopwqe = mlx5e_post_nop(wq, sq->sqn, &sq->pc); 85 mlx5e_notify_hw(wq, sq->pc, sq->uar_map, &nopwqe->ctrl); 86 } 87 88 static bool mlx5e_napi_xsk_post(struct mlx5e_xdpsq *xsksq, struct mlx5e_rq *xskrq) 89 { 90 bool need_wakeup = xsk_uses_need_wakeup(xskrq->xsk_pool); 91 bool busy_xsk = false, xsk_rx_alloc_err; 92 93 /* If SQ is empty, there are no TX completions to trigger NAPI, so set 94 * need_wakeup. Do it before queuing packets for TX to avoid race 95 * condition with userspace. 96 */ 97 if (need_wakeup && xsksq->pc == xsksq->cc) 98 xsk_set_tx_need_wakeup(xsksq->xsk_pool); 99 busy_xsk |= mlx5e_xsk_tx(xsksq, MLX5E_TX_XSK_POLL_BUDGET); 100 /* If we queued some packets for TX, no need for wakeup anymore. */ 101 if (need_wakeup && xsksq->pc != xsksq->cc) 102 xsk_clear_tx_need_wakeup(xsksq->xsk_pool); 103 104 /* If WQ is empty, RX won't trigger NAPI, so set need_wakeup. Do it 105 * before refilling to avoid race condition with userspace. 106 */ 107 if (need_wakeup && !mlx5e_rqwq_get_cur_sz(xskrq)) 108 xsk_set_rx_need_wakeup(xskrq->xsk_pool); 109 xsk_rx_alloc_err = INDIRECT_CALL_2(xskrq->post_wqes, 110 mlx5e_post_rx_mpwqes, 111 mlx5e_post_rx_wqes, 112 xskrq); 113 /* Ask for wakeup if WQ is not full after refill. */ 114 if (!need_wakeup) 115 busy_xsk |= xsk_rx_alloc_err; 116 else if (xsk_rx_alloc_err) 117 xsk_set_rx_need_wakeup(xskrq->xsk_pool); 118 else 119 xsk_clear_rx_need_wakeup(xskrq->xsk_pool); 120 121 return busy_xsk; 122 } 123 124 int mlx5e_napi_poll(struct napi_struct *napi, int budget) 125 { 126 struct mlx5e_channel *c = container_of(napi, struct mlx5e_channel, 127 napi); 128 struct mlx5e_icosq *aicosq = c->async_icosq; 129 struct mlx5e_ch_stats *ch_stats = c->stats; 130 struct mlx5e_xdpsq *xsksq = &c->xsksq; 131 struct mlx5e_txqsq __rcu **qos_sqs; 132 struct mlx5e_rq *xskrq = &c->xskrq; 133 struct mlx5e_rq *rq = &c->rq; 134 bool aff_change = false; 135 bool busy_xsk = false; 136 bool busy = false; 137 int work_done = 0; 138 u16 qos_sqs_size; 139 bool xsk_open; 140 int i; 141 142 rcu_read_lock(); 143 144 qos_sqs = rcu_dereference(c->qos_sqs); 145 146 xsk_open = test_bit(MLX5E_CHANNEL_STATE_XSK, c->state); 147 148 ch_stats->poll++; 149 150 for (i = 0; i < c->num_tc; i++) 151 busy |= mlx5e_poll_tx_cq(&c->sq[i].cq, budget); 152 153 if (unlikely(qos_sqs)) { 154 smp_rmb(); /* Pairs with mlx5e_qos_alloc_queues. */ 155 qos_sqs_size = READ_ONCE(c->qos_sqs_size); 156 157 for (i = 0; i < qos_sqs_size; i++) { 158 struct mlx5e_txqsq *sq = rcu_dereference(qos_sqs[i]); 159 160 if (sq) 161 busy |= mlx5e_poll_tx_cq(&sq->cq, budget); 162 } 163 } 164 165 /* budget=0 means we may be in IRQ context, do as little as possible */ 166 if (unlikely(!budget)) 167 goto out; 168 169 if (c->xdp) { 170 if (c->xdpsq) 171 busy |= mlx5e_poll_xdpsq_cq(&c->xdpsq->cq); 172 busy |= mlx5e_poll_xdpsq_cq(&c->rq_xdpsq.cq); 173 if (xsk_open) 174 work_done += mlx5e_poll_rx_cq(&xskrq->cq, budget); 175 } 176 177 if (likely(budget - work_done)) 178 work_done += mlx5e_poll_rx_cq(&rq->cq, budget - work_done); 179 180 busy |= work_done == budget; 181 182 mlx5e_poll_ico_cq(&c->icosq.cq); 183 if (aicosq) { 184 if (mlx5e_poll_ico_cq(&aicosq->cq)) 185 /* Don't clear the flag if nothing was polled to prevent 186 * queueing more WQEs and overflowing the async ICOSQ. 187 */ 188 clear_bit(MLX5E_SQ_STATE_PENDING_XSK_TX, 189 &aicosq->state); 190 191 /* Keep after async ICOSQ CQ poll */ 192 if (unlikely(mlx5e_ktls_rx_pending_resync_list(aicosq, budget))) 193 busy |= mlx5e_ktls_rx_handle_resync_list(aicosq, 194 budget); 195 196 if (xsk_open) { 197 busy |= mlx5e_poll_xdpsq_cq(&xsksq->cq); 198 busy_xsk |= mlx5e_napi_xsk_post(xsksq, xskrq); 199 200 busy |= busy_xsk; 201 } 202 } 203 204 busy |= INDIRECT_CALL_2(rq->post_wqes, 205 mlx5e_post_rx_mpwqes, 206 mlx5e_post_rx_wqes, 207 rq); 208 209 if (busy) { 210 if (likely(mlx5e_channel_no_affinity_change(c))) { 211 work_done = budget; 212 goto out; 213 } 214 ch_stats->aff_change++; 215 aff_change = true; 216 if (work_done == budget) 217 work_done--; 218 } 219 220 if (unlikely(!napi_complete_done(napi, work_done))) 221 goto out; 222 223 ch_stats->arm++; 224 225 for (i = 0; i < c->num_tc; i++) { 226 mlx5e_handle_tx_dim(&c->sq[i]); 227 mlx5e_cq_arm(&c->sq[i].cq); 228 } 229 if (unlikely(qos_sqs)) { 230 for (i = 0; i < qos_sqs_size; i++) { 231 struct mlx5e_txqsq *sq = rcu_dereference(qos_sqs[i]); 232 233 if (sq) { 234 mlx5e_handle_tx_dim(sq); 235 mlx5e_cq_arm(&sq->cq); 236 } 237 } 238 } 239 240 mlx5e_handle_rx_dim(rq); 241 242 mlx5e_cq_arm(&rq->cq); 243 mlx5e_cq_arm(&c->icosq.cq); 244 if (aicosq) { 245 mlx5e_cq_arm(&aicosq->cq); 246 if (xsk_open) { 247 mlx5e_handle_rx_dim(xskrq); 248 mlx5e_cq_arm(&xsksq->cq); 249 mlx5e_cq_arm(&xskrq->cq); 250 } 251 if (c->xdpsq) 252 mlx5e_cq_arm(&c->xdpsq->cq); 253 } 254 255 if (unlikely(aff_change && busy_xsk)) { 256 mlx5e_trigger_napi_async_icosq(c); 257 ch_stats->force_irq++; 258 } 259 260 out: 261 rcu_read_unlock(); 262 263 return work_done; 264 } 265 266 void mlx5e_completion_event(struct mlx5_core_cq *mcq, struct mlx5_eqe *eqe) 267 { 268 struct mlx5e_cq *cq = container_of(mcq, struct mlx5e_cq, mcq); 269 270 napi_schedule(cq->napi); 271 cq->event_ctr++; 272 cq->ch_stats->events++; 273 } 274 275 void mlx5e_cq_error_event(struct mlx5_core_cq *mcq, enum mlx5_event event) 276 { 277 struct mlx5e_cq *cq = container_of(mcq, struct mlx5e_cq, mcq); 278 struct net_device *netdev = cq->netdev; 279 280 netdev_err(netdev, "%s: cqn=0x%.6x event=0x%.2x\n", 281 __func__, mcq->cqn, event); 282 } 283