2 * Copyright (c) 2016 Solarflare Communications Inc.
5 * This software was jointly developed between OKTET Labs (under contract
6 * for Solarflare) and Solarflare Communications, Inc.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions are met:
11 * 1. Redistributions of source code must retain the above copyright notice,
12 * this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright notice,
14 * this list of conditions and the following disclaimer in the documentation
15 * and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
19 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
20 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
21 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
22 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
26 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
27 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38 sfc_rx_qcheck_conf(struct sfc_adapter *sa,
39 const struct rte_eth_rxconf *rx_conf)
43 if (rx_conf->rx_thresh.pthresh != 0 ||
44 rx_conf->rx_thresh.hthresh != 0 ||
45 rx_conf->rx_thresh.wthresh != 0) {
47 "RxQ prefetch/host/writeback thresholds are not supported");
51 if (rx_conf->rx_free_thresh != 0) {
52 sfc_err(sa, "RxQ free threshold is not supported");
56 if (rx_conf->rx_drop_en == 0) {
57 sfc_err(sa, "RxQ drop disable is not supported");
61 if (rx_conf->rx_deferred_start != 0) {
62 sfc_err(sa, "RxQ deferred start is not supported");
70 sfc_rx_mbuf_data_alignment(struct rte_mempool *mb_pool)
75 /* The mbuf object itself is always cache line aligned */
76 order = rte_bsf32(RTE_CACHE_LINE_SIZE);
78 /* Data offset from mbuf object start */
79 data_off = sizeof(struct rte_mbuf) + rte_pktmbuf_priv_size(mb_pool) +
82 order = MIN(order, rte_bsf32(data_off));
84 return 1u << (order - 1);
88 sfc_rx_mb_pool_buf_size(struct sfc_adapter *sa, struct rte_mempool *mb_pool)
90 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
91 const uint32_t nic_align_start = MAX(1, encp->enc_rx_buf_align_start);
92 const uint32_t nic_align_end = MAX(1, encp->enc_rx_buf_align_end);
94 unsigned int buf_aligned;
95 unsigned int start_alignment;
96 unsigned int end_padding_alignment;
98 /* Below it is assumed that both alignments are power of 2 */
99 SFC_ASSERT(rte_is_power_of_2(nic_align_start));
100 SFC_ASSERT(rte_is_power_of_2(nic_align_end));
103 * mbuf is always cache line aligned, double-check
104 * that it meets rx buffer start alignment requirements.
107 /* Start from mbuf pool data room size */
108 buf_size = rte_pktmbuf_data_room_size(mb_pool);
110 /* Remove headroom */
111 if (buf_size <= RTE_PKTMBUF_HEADROOM) {
113 "RxQ mbuf pool %s object data room size %u is smaller than headroom %u",
114 mb_pool->name, buf_size, RTE_PKTMBUF_HEADROOM);
117 buf_size -= RTE_PKTMBUF_HEADROOM;
119 /* Calculate guaranteed data start alignment */
120 buf_aligned = sfc_rx_mbuf_data_alignment(mb_pool);
122 /* Reserve space for start alignment */
123 if (buf_aligned < nic_align_start) {
124 start_alignment = nic_align_start - buf_aligned;
125 if (buf_size <= start_alignment) {
127 "RxQ mbuf pool %s object data room size %u is insufficient for headroom %u and buffer start alignment %u required by NIC",
129 rte_pktmbuf_data_room_size(mb_pool),
130 RTE_PKTMBUF_HEADROOM, start_alignment);
133 buf_aligned = nic_align_start;
134 buf_size -= start_alignment;
139 /* Make sure that end padding does not write beyond the buffer */
140 if (buf_aligned < nic_align_end) {
142 * Estimate space which can be lost. If guarnteed buffer
143 * size is odd, lost space is (nic_align_end - 1). More
144 * accurate formula is below.
146 end_padding_alignment = nic_align_end -
147 MIN(buf_aligned, 1u << (rte_bsf32(buf_size) - 1));
148 if (buf_size <= end_padding_alignment) {
150 "RxQ mbuf pool %s object data room size %u is insufficient for headroom %u, buffer start alignment %u and end padding alignment %u required by NIC",
152 rte_pktmbuf_data_room_size(mb_pool),
153 RTE_PKTMBUF_HEADROOM, start_alignment,
154 end_padding_alignment);
157 buf_size -= end_padding_alignment;
160 * Start is aligned the same or better than end,
163 buf_size = P2ALIGN(buf_size, nic_align_end);
170 sfc_rx_qinit(struct sfc_adapter *sa, unsigned int sw_index,
171 uint16_t nb_rx_desc, unsigned int socket_id,
172 const struct rte_eth_rxconf *rx_conf,
173 struct rte_mempool *mb_pool)
175 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
178 struct sfc_rxq_info *rxq_info;
179 unsigned int evq_index;
183 rc = sfc_rx_qcheck_conf(sa, rx_conf);
187 buf_size = sfc_rx_mb_pool_buf_size(sa, mb_pool);
189 sfc_err(sa, "RxQ %u mbuf pool object size is too small",
195 if ((buf_size < sa->port.pdu + encp->enc_rx_prefix_size) &&
196 !sa->eth_dev->data->dev_conf.rxmode.enable_scatter) {
197 sfc_err(sa, "Rx scatter is disabled and RxQ %u mbuf pool "
198 "object size is too small", sw_index);
199 sfc_err(sa, "RxQ %u calculated Rx buffer size is %u vs "
200 "PDU size %u plus Rx prefix %u bytes",
201 sw_index, buf_size, (unsigned int)sa->port.pdu,
202 encp->enc_rx_prefix_size);
207 SFC_ASSERT(sw_index < sa->rxq_count);
208 rxq_info = &sa->rxq_info[sw_index];
210 SFC_ASSERT(nb_rx_desc <= rxq_info->max_entries);
211 rxq_info->entries = nb_rx_desc;
212 rxq_info->type = EFX_RXQ_TYPE_DEFAULT;
214 evq_index = sfc_evq_index_by_rxq_sw_index(sa, sw_index);
216 rc = sfc_ev_qinit(sa, evq_index, rxq_info->entries, socket_id);
220 evq = sa->evq_info[evq_index].evq;
223 rxq = rte_zmalloc_socket("sfc-rxq", sizeof(*rxq), RTE_CACHE_LINE_SIZE,
228 rc = sfc_dma_alloc(sa, "rxq", sw_index, EFX_RXQ_SIZE(rxq_info->entries),
229 socket_id, &rxq->mem);
234 rxq->sw_desc = rte_calloc_socket("sfc-rxq-sw_desc", rxq_info->entries,
235 sizeof(*rxq->sw_desc),
236 RTE_CACHE_LINE_SIZE, socket_id);
237 if (rxq->sw_desc == NULL)
238 goto fail_desc_alloc;
242 rxq->ptr_mask = rxq_info->entries - 1;
243 rxq->refill_mb_pool = mb_pool;
244 rxq->buf_size = buf_size;
245 rxq->hw_index = sw_index;
247 rxq->state = SFC_RXQ_INITIALIZED;
254 sfc_dma_free(sa, &rxq->mem);
260 sfc_ev_qfini(sa, evq_index);
263 rxq_info->entries = 0;
266 sfc_log_init(sa, "failed %d", rc);
271 sfc_rx_qfini(struct sfc_adapter *sa, unsigned int sw_index)
273 struct sfc_rxq_info *rxq_info;
276 SFC_ASSERT(sw_index < sa->rxq_count);
278 rxq_info = &sa->rxq_info[sw_index];
281 SFC_ASSERT(rxq->state == SFC_RXQ_INITIALIZED);
283 rxq_info->rxq = NULL;
284 rxq_info->entries = 0;
286 rte_free(rxq->sw_desc);
287 sfc_dma_free(sa, &rxq->mem);
292 sfc_rx_qinit_info(struct sfc_adapter *sa, unsigned int sw_index)
294 struct sfc_rxq_info *rxq_info = &sa->rxq_info[sw_index];
295 unsigned int max_entries;
297 max_entries = EFX_RXQ_MAXNDESCS;
298 SFC_ASSERT(rte_is_power_of_2(max_entries));
300 rxq_info->max_entries = max_entries;
306 sfc_rx_check_mode(struct sfc_adapter *sa, struct rte_eth_rxmode *rxmode)
310 switch (rxmode->mq_mode) {
312 /* No special checks are required */
315 sfc_err(sa, "Rx multi-queue mode %u not supported",
320 if (rxmode->header_split) {
321 sfc_err(sa, "Header split on Rx not supported");
325 if (rxmode->hw_vlan_filter) {
326 sfc_err(sa, "HW VLAN filtering not supported");
330 if (rxmode->hw_vlan_strip) {
331 sfc_err(sa, "HW VLAN stripping not supported");
335 if (rxmode->hw_vlan_extend) {
337 "Q-in-Q HW VLAN stripping not supported");
341 if (!rxmode->hw_strip_crc) {
343 "FCS stripping control not supported - always stripped");
344 rxmode->hw_strip_crc = 1;
347 if (rxmode->enable_scatter) {
348 sfc_err(sa, "Scatter on Rx not supported");
352 if (rxmode->enable_lro) {
353 sfc_err(sa, "LRO not supported");
361 * Initialize Rx subsystem.
363 * Called at device configuration stage when number of receive queues is
364 * specified together with other device level receive configuration.
366 * It should be used to allocate NUMA-unaware resources.
369 sfc_rx_init(struct sfc_adapter *sa)
371 struct rte_eth_conf *dev_conf = &sa->eth_dev->data->dev_conf;
372 unsigned int sw_index;
375 rc = sfc_rx_check_mode(sa, &dev_conf->rxmode);
377 goto fail_check_mode;
379 sa->rxq_count = sa->eth_dev->data->nb_rx_queues;
382 sa->rxq_info = rte_calloc_socket("sfc-rxqs", sa->rxq_count,
383 sizeof(struct sfc_rxq_info), 0,
385 if (sa->rxq_info == NULL)
386 goto fail_rxqs_alloc;
388 for (sw_index = 0; sw_index < sa->rxq_count; ++sw_index) {
389 rc = sfc_rx_qinit_info(sa, sw_index);
391 goto fail_rx_qinit_info;
397 rte_free(sa->rxq_info);
403 sfc_log_init(sa, "failed %d", rc);
408 * Shutdown Rx subsystem.
410 * Called at device close stage, for example, before device
411 * reconfiguration or shutdown.
414 sfc_rx_fini(struct sfc_adapter *sa)
416 unsigned int sw_index;
418 sw_index = sa->rxq_count;
419 while (sw_index-- > 0) {
420 if (sa->rxq_info[sw_index].rxq != NULL)
421 sfc_rx_qfini(sa, sw_index);
424 rte_free(sa->rxq_info);