ceph/src/seastar/dpdk/drivers/net/sfc/sfc_tso.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  * Copyright (c) 2016-2017 Solarflare Communications Inc.
   5  * All rights reserved.
   6  *
   7  * This software was jointly developed between OKTET Labs (under contract
   8  * for Solarflare) and Solarflare Communications, Inc.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright notice,
  14  *    this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright notice,
  16  *    this list of conditions and the following disclaimer in the documentation
  17  *    and/or other materials provided with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
  21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  22  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
  23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
  24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
  26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
  28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
  29  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  30  */
  31
  32 #include <rte_ip.h>
  33 #include <rte_tcp.h>
  34
  35 #include "sfc.h"
  36 #include "sfc_debug.h"
  37 #include "sfc_tx.h"
  38 #include "sfc_ev.h"
  39
  40 /** Standard TSO header length */
  41 #define SFC_TSOH_STD_LEN        256
  42
  43 /** The number of TSO option descriptors that precede the packet descriptors */
  44 #define SFC_TSO_OPDESCS_IDX_SHIFT       2
  45
  46 int
  47 sfc_efx_tso_alloc_tsoh_objs(struct sfc_efx_tx_sw_desc *sw_ring,
  48                             unsigned int txq_entries, unsigned int socket_id)
  49 {
  50         unsigned int i;
  51
  52         for (i = 0; i < txq_entries; ++i) {
  53                 sw_ring[i].tsoh = rte_malloc_socket("sfc-efx-txq-tsoh-obj",
  54                                                     SFC_TSOH_STD_LEN,
  55                                                     RTE_CACHE_LINE_SIZE,
  56                                                     socket_id);
  57                 if (sw_ring[i].tsoh == NULL)
  58                         goto fail_alloc_tsoh_objs;
  59         }
  60
  61         return 0;
  62
  63 fail_alloc_tsoh_objs:
  64         while (i > 0)
  65                 rte_free(sw_ring[--i].tsoh);
  66
  67         return ENOMEM;
  68 }
  69
  70 void
  71 sfc_efx_tso_free_tsoh_objs(struct sfc_efx_tx_sw_desc *sw_ring,
  72                            unsigned int txq_entries)
  73 {
  74         unsigned int i;
  75
  76         for (i = 0; i < txq_entries; ++i) {
  77                 rte_free(sw_ring[i].tsoh);
  78                 sw_ring[i].tsoh = NULL;
  79         }
  80 }
  81
  82 static void
  83 sfc_efx_tso_prepare_header(struct sfc_efx_txq *txq, struct rte_mbuf **in_seg,
  84                            size_t *in_off, unsigned int idx, size_t bytes_left)
  85 {
  86         struct rte_mbuf *m = *in_seg;
  87         size_t bytes_to_copy = 0;
  88         uint8_t *tsoh = txq->sw_ring[idx & txq->ptr_mask].tsoh;
  89
  90         do {
  91                 bytes_to_copy = MIN(bytes_left, m->data_len);
  92
  93                 rte_memcpy(tsoh, rte_pktmbuf_mtod(m, uint8_t *),
  94                            bytes_to_copy);
  95
  96                 bytes_left -= bytes_to_copy;
  97                 tsoh += bytes_to_copy;
  98
  99                 if (bytes_left > 0) {
 100                         m = m->next;
 101                         SFC_ASSERT(m != NULL);
 102                 }
 103         } while (bytes_left > 0);
 104
 105         if (bytes_to_copy == m->data_len) {
 106                 *in_seg = m->next;
 107                 *in_off = 0;
 108         } else {
 109                 *in_seg = m;
 110                 *in_off = bytes_to_copy;
 111         }
 112 }
 113
 114 int
 115 sfc_efx_tso_do(struct sfc_efx_txq *txq, unsigned int idx,
 116                struct rte_mbuf **in_seg, size_t *in_off, efx_desc_t **pend,
 117                unsigned int *pkt_descs, size_t *pkt_len)
 118 {
 119         uint8_t *tsoh;
 120         const struct tcp_hdr *th;
 121         efsys_dma_addr_t header_paddr;
 122         uint16_t packet_id;
 123         uint32_t sent_seq;
 124         struct rte_mbuf *m = *in_seg;
 125         size_t nh_off = m->l2_len; /* IP header offset */
 126         size_t tcph_off = m->l2_len + m->l3_len; /* TCP header offset */
 127         size_t header_len = m->l2_len + m->l3_len + m->l4_len;
 128         const efx_nic_cfg_t *encp = efx_nic_cfg_get(txq->evq->sa->nic);
 129
 130         idx += SFC_TSO_OPDESCS_IDX_SHIFT;
 131
 132         /* Packets which have too big headers should be discarded */
 133         if (unlikely(header_len > SFC_TSOH_STD_LEN))
 134                 return EMSGSIZE;
 135
 136         /*
 137          * The TCP header must start at most 208 bytes into the frame.
 138          * If it starts later than this then the NIC won't realise
 139          * it's a TCP packet and TSO edits won't be applied
 140          */
 141         if (unlikely(tcph_off > encp->enc_tx_tso_tcp_header_offset_limit))
 142                 return EMSGSIZE;
 143
 144         header_paddr = rte_pktmbuf_mtophys(m);
 145
 146         /*
 147          * Sometimes headers may be split across multiple mbufs. In such cases
 148          * we need to glue those pieces and store them in some temporary place.
 149          * Also, packet headers must be contiguous in memory, so that
 150          * they can be referred to with a single DMA descriptor. EF10 has no
 151          * limitations on address boundaries crossing by DMA descriptor data.
 152          */
 153         if (m->data_len < header_len) {
 154                 sfc_efx_tso_prepare_header(txq, in_seg, in_off, idx,
 155                                            header_len);
 156                 tsoh = txq->sw_ring[idx & txq->ptr_mask].tsoh;
 157
 158                 header_paddr = rte_malloc_virt2phy((void *)tsoh);
 159         } else {
 160                 if (m->data_len == header_len) {
 161                         *in_off = 0;
 162                         *in_seg = m->next;
 163                 } else {
 164                         *in_off = header_len;
 165                 }
 166
 167                 tsoh = rte_pktmbuf_mtod(m, uint8_t *);
 168         }
 169
 170         /* Handle IP header */
 171         if (m->ol_flags & PKT_TX_IPV4) {
 172                 const struct ipv4_hdr *iphe4;
 173
 174                 iphe4 = (const struct ipv4_hdr *)(tsoh + nh_off);
 175                 rte_memcpy(&packet_id, &iphe4->packet_id, sizeof(uint16_t));
 176                 packet_id = rte_be_to_cpu_16(packet_id);
 177         } else if (m->ol_flags & PKT_TX_IPV6) {
 178                 packet_id = 0;
 179         } else {
 180                 return EINVAL;
 181         }
 182
 183         /* Handle TCP header */
 184         th = (const struct tcp_hdr *)(tsoh + tcph_off);
 185
 186         rte_memcpy(&sent_seq, &th->sent_seq, sizeof(uint32_t));
 187         sent_seq = rte_be_to_cpu_32(sent_seq);
 188
 189         efx_tx_qdesc_tso2_create(txq->common, packet_id, sent_seq, m->tso_segsz,
 190                                  *pend, EFX_TX_FATSOV2_OPT_NDESCS);
 191
 192         *pend += EFX_TX_FATSOV2_OPT_NDESCS;
 193         *pkt_descs += EFX_TX_FATSOV2_OPT_NDESCS;
 194
 195         efx_tx_qdesc_dma_create(txq->common, header_paddr, header_len,
 196                                 B_FALSE, (*pend)++);
 197         (*pkt_descs)++;
 198         *pkt_len -= header_len;
 199
 200         return 0;
 201 }