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 "en.h"
35 #include "en/txrx.h"
36 #include "en/xdp.h"
37 #include "en/xsk/rx.h"
38 #include "en/xsk/tx.h"
39 #include "en_accel/ktls_txrx.h"
40
mlx5e_channel_no_affinity_change(struct mlx5e_channel * c)41 static inline bool mlx5e_channel_no_affinity_change(struct mlx5e_channel *c)
42 {
43 int current_cpu = smp_processor_id();
44
45 return cpumask_test_cpu(current_cpu, c->aff_mask);
46 }
47
mlx5e_handle_tx_dim(struct mlx5e_txqsq * sq)48 static void mlx5e_handle_tx_dim(struct mlx5e_txqsq *sq)
49 {
50 struct mlx5e_sq_stats *stats = sq->stats;
51 struct dim_sample dim_sample = {};
52
53 if (unlikely(!test_bit(MLX5E_SQ_STATE_AM, &sq->state)))
54 return;
55
56 dim_update_sample(sq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
57 net_dim(&sq->dim, dim_sample);
58 }
59
mlx5e_handle_rx_dim(struct mlx5e_rq * rq)60 static void mlx5e_handle_rx_dim(struct mlx5e_rq *rq)
61 {
62 struct mlx5e_rq_stats *stats = rq->stats;
63 struct dim_sample dim_sample = {};
64
65 if (unlikely(!test_bit(MLX5E_RQ_STATE_AM, &rq->state)))
66 return;
67
68 dim_update_sample(rq->cq.event_ctr, stats->packets, stats->bytes, &dim_sample);
69 net_dim(&rq->dim, dim_sample);
70 }
71
mlx5e_trigger_irq(struct mlx5e_icosq * sq)72 void mlx5e_trigger_irq(struct mlx5e_icosq *sq)
73 {
74 struct mlx5_wq_cyc *wq = &sq->wq;
75 struct mlx5e_tx_wqe *nopwqe;
76 u16 pi = mlx5_wq_cyc_ctr2ix(wq, sq->pc);
77
78 sq->db.wqe_info[pi] = (struct mlx5e_icosq_wqe_info) {
79 .wqe_type = MLX5E_ICOSQ_WQE_NOP,
80 .num_wqebbs = 1,
81 };
82
83 nopwqe = mlx5e_post_nop(wq, sq->sqn, &sq->pc);
84 mlx5e_notify_hw(wq, sq->pc, sq->uar_map, &nopwqe->ctrl);
85 }
86
mlx5e_napi_xsk_post(struct mlx5e_xdpsq * xsksq,struct mlx5e_rq * xskrq)87 static bool mlx5e_napi_xsk_post(struct mlx5e_xdpsq *xsksq, struct mlx5e_rq *xskrq)
88 {
89 bool busy_xsk = false, xsk_rx_alloc_err;
90
91 /* Handle the race between the application querying need_wakeup and the
92 * driver setting it:
93 * 1. Update need_wakeup both before and after the TX. If it goes to
94 * "yes", it can only happen with the first update.
95 * 2. If the application queried need_wakeup before we set it, the
96 * packets will be transmitted anyway, even w/o a wakeup.
97 * 3. Give a chance to clear need_wakeup after new packets were queued
98 * for TX.
99 */
100 mlx5e_xsk_update_tx_wakeup(xsksq);
101 busy_xsk |= mlx5e_xsk_tx(xsksq, MLX5E_TX_XSK_POLL_BUDGET);
102 mlx5e_xsk_update_tx_wakeup(xsksq);
103
104 xsk_rx_alloc_err = INDIRECT_CALL_2(xskrq->post_wqes,
105 mlx5e_post_rx_mpwqes,
106 mlx5e_post_rx_wqes,
107 xskrq);
108 busy_xsk |= mlx5e_xsk_update_rx_wakeup(xskrq, xsk_rx_alloc_err);
109
110 return busy_xsk;
111 }
112
mlx5e_napi_poll(struct napi_struct * napi,int budget)113 int mlx5e_napi_poll(struct napi_struct *napi, int budget)
114 {
115 struct mlx5e_channel *c = container_of(napi, struct mlx5e_channel,
116 napi);
117 struct mlx5e_ch_stats *ch_stats = c->stats;
118 struct mlx5e_xdpsq *xsksq = &c->xsksq;
119 struct mlx5e_txqsq __rcu **qos_sqs;
120 struct mlx5e_rq *xskrq = &c->xskrq;
121 struct mlx5e_rq *rq = &c->rq;
122 bool aff_change = false;
123 bool busy_xsk = false;
124 bool busy = false;
125 int work_done = 0;
126 u16 qos_sqs_size;
127 bool xsk_open;
128 int i;
129
130 rcu_read_lock();
131
132 qos_sqs = rcu_dereference(c->qos_sqs);
133
134 xsk_open = test_bit(MLX5E_CHANNEL_STATE_XSK, c->state);
135
136 ch_stats->poll++;
137
138 for (i = 0; i < c->num_tc; i++)
139 busy |= mlx5e_poll_tx_cq(&c->sq[i].cq, budget);
140
141 if (unlikely(qos_sqs)) {
142 smp_rmb(); /* Pairs with mlx5e_qos_alloc_queues. */
143 qos_sqs_size = READ_ONCE(c->qos_sqs_size);
144
145 for (i = 0; i < qos_sqs_size; i++) {
146 struct mlx5e_txqsq *sq = rcu_dereference(qos_sqs[i]);
147
148 if (sq)
149 busy |= mlx5e_poll_tx_cq(&sq->cq, budget);
150 }
151 }
152
153 /* budget=0 means we may be in IRQ context, do as little as possible */
154 if (unlikely(!budget))
155 goto out;
156
157 busy |= mlx5e_poll_xdpsq_cq(&c->xdpsq.cq);
158
159 if (c->xdp)
160 busy |= mlx5e_poll_xdpsq_cq(&c->rq_xdpsq.cq);
161
162 if (xsk_open)
163 work_done = mlx5e_poll_rx_cq(&xskrq->cq, budget);
164
165 if (likely(budget - work_done))
166 work_done += mlx5e_poll_rx_cq(&rq->cq, budget - work_done);
167
168 busy |= work_done == budget;
169
170 mlx5e_poll_ico_cq(&c->icosq.cq);
171 if (mlx5e_poll_ico_cq(&c->async_icosq.cq))
172 /* Don't clear the flag if nothing was polled to prevent
173 * queueing more WQEs and overflowing the async ICOSQ.
174 */
175 clear_bit(MLX5E_SQ_STATE_PENDING_XSK_TX, &c->async_icosq.state);
176
177 /* Keep after async ICOSQ CQ poll */
178 if (unlikely(mlx5e_ktls_rx_pending_resync_list(c, budget)))
179 busy |= mlx5e_ktls_rx_handle_resync_list(c, budget);
180
181 busy |= INDIRECT_CALL_2(rq->post_wqes,
182 mlx5e_post_rx_mpwqes,
183 mlx5e_post_rx_wqes,
184 rq);
185 if (xsk_open) {
186 busy |= mlx5e_poll_xdpsq_cq(&xsksq->cq);
187 busy_xsk |= mlx5e_napi_xsk_post(xsksq, xskrq);
188 }
189
190 busy |= busy_xsk;
191
192 if (busy) {
193 if (likely(mlx5e_channel_no_affinity_change(c))) {
194 work_done = budget;
195 goto out;
196 }
197 ch_stats->aff_change++;
198 aff_change = true;
199 if (budget && work_done == budget)
200 work_done--;
201 }
202
203 if (unlikely(!napi_complete_done(napi, work_done)))
204 goto out;
205
206 ch_stats->arm++;
207
208 for (i = 0; i < c->num_tc; i++) {
209 mlx5e_handle_tx_dim(&c->sq[i]);
210 mlx5e_cq_arm(&c->sq[i].cq);
211 }
212 if (unlikely(qos_sqs)) {
213 for (i = 0; i < qos_sqs_size; i++) {
214 struct mlx5e_txqsq *sq = rcu_dereference(qos_sqs[i]);
215
216 if (sq) {
217 mlx5e_handle_tx_dim(sq);
218 mlx5e_cq_arm(&sq->cq);
219 }
220 }
221 }
222
223 mlx5e_handle_rx_dim(rq);
224
225 mlx5e_cq_arm(&rq->cq);
226 mlx5e_cq_arm(&c->icosq.cq);
227 mlx5e_cq_arm(&c->async_icosq.cq);
228 mlx5e_cq_arm(&c->xdpsq.cq);
229
230 if (xsk_open) {
231 mlx5e_handle_rx_dim(xskrq);
232 mlx5e_cq_arm(&xsksq->cq);
233 mlx5e_cq_arm(&xskrq->cq);
234 }
235
236 if (unlikely(aff_change && busy_xsk)) {
237 mlx5e_trigger_irq(&c->icosq);
238 ch_stats->force_irq++;
239 }
240
241 out:
242 rcu_read_unlock();
243
244 return work_done;
245 }
246
mlx5e_completion_event(struct mlx5_core_cq * mcq,struct mlx5_eqe * eqe)247 void mlx5e_completion_event(struct mlx5_core_cq *mcq, struct mlx5_eqe *eqe)
248 {
249 struct mlx5e_cq *cq = container_of(mcq, struct mlx5e_cq, mcq);
250
251 napi_schedule(cq->napi);
252 cq->event_ctr++;
253 cq->ch_stats->events++;
254 }
255
mlx5e_cq_error_event(struct mlx5_core_cq * mcq,enum mlx5_event event)256 void mlx5e_cq_error_event(struct mlx5_core_cq *mcq, enum mlx5_event event)
257 {
258 struct mlx5e_cq *cq = container_of(mcq, struct mlx5e_cq, mcq);
259 struct net_device *netdev = cq->netdev;
260
261 netdev_err(netdev, "%s: cqn=0x%.6x event=0x%.2x\n",
262 __func__, mcq->cqn, event);
263 }
264