1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2019, Intel Corporation. */ 3 4 #include <linux/filter.h> 5 6 #include "ice_txrx_lib.h" 7 #include "ice_eswitch.h" 8 #include "ice_lib.h" 9 10 /** 11 * ice_release_rx_desc - Store the new tail and head values 12 * @rx_ring: ring to bump 13 * @val: new head index 14 */ 15 void ice_release_rx_desc(struct ice_rx_ring *rx_ring, u16 val) 16 { 17 u16 prev_ntu = rx_ring->next_to_use & ~0x7; 18 19 rx_ring->next_to_use = val; 20 21 /* update next to alloc since we have filled the ring */ 22 rx_ring->next_to_alloc = val; 23 24 /* QRX_TAIL will be updated with any tail value, but hardware ignores 25 * the lower 3 bits. This makes it so we only bump tail on meaningful 26 * boundaries. Also, this allows us to bump tail on intervals of 8 up to 27 * the budget depending on the current traffic load. 28 */ 29 val &= ~0x7; 30 if (prev_ntu != val) { 31 /* Force memory writes to complete before letting h/w 32 * know there are new descriptors to fetch. (Only 33 * applicable for weak-ordered memory model archs, 34 * such as IA-64). 35 */ 36 wmb(); 37 writel(val, rx_ring->tail); 38 } 39 } 40 41 /** 42 * ice_ptype_to_htype - get a hash type 43 * @ptype: the ptype value from the descriptor 44 * 45 * Returns appropriate hash type (such as PKT_HASH_TYPE_L2/L3/L4) to be used by 46 * skb_set_hash based on PTYPE as parsed by HW Rx pipeline and is part of 47 * Rx desc. 48 */ 49 static enum pkt_hash_types ice_ptype_to_htype(u16 ptype) 50 { 51 struct ice_rx_ptype_decoded decoded = ice_decode_rx_desc_ptype(ptype); 52 53 if (!decoded.known) 54 return PKT_HASH_TYPE_NONE; 55 if (decoded.payload_layer == ICE_RX_PTYPE_PAYLOAD_LAYER_PAY4) 56 return PKT_HASH_TYPE_L4; 57 if (decoded.payload_layer == ICE_RX_PTYPE_PAYLOAD_LAYER_PAY3) 58 return PKT_HASH_TYPE_L3; 59 if (decoded.outer_ip == ICE_RX_PTYPE_OUTER_L2) 60 return PKT_HASH_TYPE_L2; 61 62 return PKT_HASH_TYPE_NONE; 63 } 64 65 /** 66 * ice_rx_hash - set the hash value in the skb 67 * @rx_ring: descriptor ring 68 * @rx_desc: specific descriptor 69 * @skb: pointer to current skb 70 * @rx_ptype: the ptype value from the descriptor 71 */ 72 static void 73 ice_rx_hash(struct ice_rx_ring *rx_ring, union ice_32b_rx_flex_desc *rx_desc, 74 struct sk_buff *skb, u16 rx_ptype) 75 { 76 struct ice_32b_rx_flex_desc_nic *nic_mdid; 77 u32 hash; 78 79 if (!(rx_ring->netdev->features & NETIF_F_RXHASH)) 80 return; 81 82 if (rx_desc->wb.rxdid != ICE_RXDID_FLEX_NIC) 83 return; 84 85 nic_mdid = (struct ice_32b_rx_flex_desc_nic *)rx_desc; 86 hash = le32_to_cpu(nic_mdid->rss_hash); 87 skb_set_hash(skb, hash, ice_ptype_to_htype(rx_ptype)); 88 } 89 90 /** 91 * ice_rx_csum - Indicate in skb if checksum is good 92 * @ring: the ring we care about 93 * @skb: skb currently being received and modified 94 * @rx_desc: the receive descriptor 95 * @ptype: the packet type decoded by hardware 96 * 97 * skb->protocol must be set before this function is called 98 */ 99 static void 100 ice_rx_csum(struct ice_rx_ring *ring, struct sk_buff *skb, 101 union ice_32b_rx_flex_desc *rx_desc, u16 ptype) 102 { 103 struct ice_rx_ptype_decoded decoded; 104 u16 rx_status0, rx_status1; 105 bool ipv4, ipv6; 106 107 rx_status0 = le16_to_cpu(rx_desc->wb.status_error0); 108 rx_status1 = le16_to_cpu(rx_desc->wb.status_error1); 109 110 decoded = ice_decode_rx_desc_ptype(ptype); 111 112 /* Start with CHECKSUM_NONE and by default csum_level = 0 */ 113 skb->ip_summed = CHECKSUM_NONE; 114 skb_checksum_none_assert(skb); 115 116 /* check if Rx checksum is enabled */ 117 if (!(ring->netdev->features & NETIF_F_RXCSUM)) 118 return; 119 120 /* check if HW has decoded the packet and checksum */ 121 if (!(rx_status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_L3L4P_S))) 122 return; 123 124 if (!(decoded.known && decoded.outer_ip)) 125 return; 126 127 ipv4 = (decoded.outer_ip == ICE_RX_PTYPE_OUTER_IP) && 128 (decoded.outer_ip_ver == ICE_RX_PTYPE_OUTER_IPV4); 129 ipv6 = (decoded.outer_ip == ICE_RX_PTYPE_OUTER_IP) && 130 (decoded.outer_ip_ver == ICE_RX_PTYPE_OUTER_IPV6); 131 132 if (ipv4 && (rx_status0 & (BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_IPE_S) | 133 BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_EIPE_S)))) 134 goto checksum_fail; 135 136 if (ipv6 && (rx_status0 & (BIT(ICE_RX_FLEX_DESC_STATUS0_IPV6EXADD_S)))) 137 goto checksum_fail; 138 139 /* check for L4 errors and handle packets that were not able to be 140 * checksummed due to arrival speed 141 */ 142 if (rx_status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_L4E_S)) 143 goto checksum_fail; 144 145 /* check for outer UDP checksum error in tunneled packets */ 146 if ((rx_status1 & BIT(ICE_RX_FLEX_DESC_STATUS1_NAT_S)) && 147 (rx_status0 & BIT(ICE_RX_FLEX_DESC_STATUS0_XSUM_EUDPE_S))) 148 goto checksum_fail; 149 150 /* If there is an outer header present that might contain a checksum 151 * we need to bump the checksum level by 1 to reflect the fact that 152 * we are indicating we validated the inner checksum. 153 */ 154 if (decoded.tunnel_type >= ICE_RX_PTYPE_TUNNEL_IP_GRENAT) 155 skb->csum_level = 1; 156 157 /* Only report checksum unnecessary for TCP, UDP, or SCTP */ 158 switch (decoded.inner_prot) { 159 case ICE_RX_PTYPE_INNER_PROT_TCP: 160 case ICE_RX_PTYPE_INNER_PROT_UDP: 161 case ICE_RX_PTYPE_INNER_PROT_SCTP: 162 skb->ip_summed = CHECKSUM_UNNECESSARY; 163 break; 164 default: 165 break; 166 } 167 return; 168 169 checksum_fail: 170 ring->vsi->back->hw_csum_rx_error++; 171 } 172 173 /** 174 * ice_process_skb_fields - Populate skb header fields from Rx descriptor 175 * @rx_ring: Rx descriptor ring packet is being transacted on 176 * @rx_desc: pointer to the EOP Rx descriptor 177 * @skb: pointer to current skb being populated 178 * @ptype: the packet type decoded by hardware 179 * 180 * This function checks the ring, descriptor, and packet information in 181 * order to populate the hash, checksum, VLAN, protocol, and 182 * other fields within the skb. 183 */ 184 void 185 ice_process_skb_fields(struct ice_rx_ring *rx_ring, 186 union ice_32b_rx_flex_desc *rx_desc, 187 struct sk_buff *skb, u16 ptype) 188 { 189 ice_rx_hash(rx_ring, rx_desc, skb, ptype); 190 191 /* modifies the skb - consumes the enet header */ 192 skb->protocol = eth_type_trans(skb, rx_ring->netdev); 193 194 ice_rx_csum(rx_ring, skb, rx_desc, ptype); 195 196 if (rx_ring->ptp_rx) 197 ice_ptp_rx_hwtstamp(rx_ring, rx_desc, skb); 198 } 199 200 /** 201 * ice_receive_skb - Send a completed packet up the stack 202 * @rx_ring: Rx ring in play 203 * @skb: packet to send up 204 * @vlan_tag: VLAN tag for packet 205 * 206 * This function sends the completed packet (via. skb) up the stack using 207 * gro receive functions (with/without VLAN tag) 208 */ 209 void 210 ice_receive_skb(struct ice_rx_ring *rx_ring, struct sk_buff *skb, u16 vlan_tag) 211 { 212 netdev_features_t features = rx_ring->netdev->features; 213 bool non_zero_vlan = !!(vlan_tag & VLAN_VID_MASK); 214 215 if ((features & NETIF_F_HW_VLAN_CTAG_RX) && non_zero_vlan) 216 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag); 217 else if ((features & NETIF_F_HW_VLAN_STAG_RX) && non_zero_vlan) 218 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021AD), vlan_tag); 219 220 napi_gro_receive(&rx_ring->q_vector->napi, skb); 221 } 222 223 /** 224 * ice_clean_xdp_irq - Reclaim resources after transmit completes on XDP ring 225 * @xdp_ring: XDP ring to clean 226 */ 227 static void ice_clean_xdp_irq(struct ice_tx_ring *xdp_ring) 228 { 229 unsigned int total_bytes = 0, total_pkts = 0; 230 u16 tx_thresh = ICE_RING_QUARTER(xdp_ring); 231 u16 ntc = xdp_ring->next_to_clean; 232 struct ice_tx_desc *next_dd_desc; 233 u16 next_dd = xdp_ring->next_dd; 234 struct ice_tx_buf *tx_buf; 235 int i; 236 237 next_dd_desc = ICE_TX_DESC(xdp_ring, next_dd); 238 if (!(next_dd_desc->cmd_type_offset_bsz & 239 cpu_to_le64(ICE_TX_DESC_DTYPE_DESC_DONE))) 240 return; 241 242 for (i = 0; i < tx_thresh; i++) { 243 tx_buf = &xdp_ring->tx_buf[ntc]; 244 245 total_bytes += tx_buf->bytecount; 246 /* normally tx_buf->gso_segs was taken but at this point 247 * it's always 1 for us 248 */ 249 total_pkts++; 250 251 page_frag_free(tx_buf->raw_buf); 252 dma_unmap_single(xdp_ring->dev, dma_unmap_addr(tx_buf, dma), 253 dma_unmap_len(tx_buf, len), DMA_TO_DEVICE); 254 dma_unmap_len_set(tx_buf, len, 0); 255 tx_buf->raw_buf = NULL; 256 257 ntc++; 258 if (ntc >= xdp_ring->count) 259 ntc = 0; 260 } 261 262 next_dd_desc->cmd_type_offset_bsz = 0; 263 xdp_ring->next_dd = xdp_ring->next_dd + tx_thresh; 264 if (xdp_ring->next_dd > xdp_ring->count) 265 xdp_ring->next_dd = tx_thresh - 1; 266 xdp_ring->next_to_clean = ntc; 267 ice_update_tx_ring_stats(xdp_ring, total_pkts, total_bytes); 268 } 269 270 /** 271 * ice_xmit_xdp_ring - submit single packet to XDP ring for transmission 272 * @data: packet data pointer 273 * @size: packet data size 274 * @xdp_ring: XDP ring for transmission 275 */ 276 int ice_xmit_xdp_ring(void *data, u16 size, struct ice_tx_ring *xdp_ring) 277 { 278 u16 tx_thresh = ICE_RING_QUARTER(xdp_ring); 279 u16 i = xdp_ring->next_to_use; 280 struct ice_tx_desc *tx_desc; 281 struct ice_tx_buf *tx_buf; 282 dma_addr_t dma; 283 284 if (ICE_DESC_UNUSED(xdp_ring) < tx_thresh) 285 ice_clean_xdp_irq(xdp_ring); 286 287 if (!unlikely(ICE_DESC_UNUSED(xdp_ring))) { 288 xdp_ring->ring_stats->tx_stats.tx_busy++; 289 return ICE_XDP_CONSUMED; 290 } 291 292 dma = dma_map_single(xdp_ring->dev, data, size, DMA_TO_DEVICE); 293 if (dma_mapping_error(xdp_ring->dev, dma)) 294 return ICE_XDP_CONSUMED; 295 296 tx_buf = &xdp_ring->tx_buf[i]; 297 tx_buf->bytecount = size; 298 tx_buf->gso_segs = 1; 299 tx_buf->raw_buf = data; 300 301 /* record length, and DMA address */ 302 dma_unmap_len_set(tx_buf, len, size); 303 dma_unmap_addr_set(tx_buf, dma, dma); 304 305 tx_desc = ICE_TX_DESC(xdp_ring, i); 306 tx_desc->buf_addr = cpu_to_le64(dma); 307 tx_desc->cmd_type_offset_bsz = ice_build_ctob(ICE_TX_DESC_CMD_EOP, 0, 308 size, 0); 309 310 xdp_ring->xdp_tx_active++; 311 i++; 312 if (i == xdp_ring->count) { 313 i = 0; 314 tx_desc = ICE_TX_DESC(xdp_ring, xdp_ring->next_rs); 315 tx_desc->cmd_type_offset_bsz |= 316 cpu_to_le64(ICE_TX_DESC_CMD_RS << ICE_TXD_QW1_CMD_S); 317 xdp_ring->next_rs = tx_thresh - 1; 318 } 319 xdp_ring->next_to_use = i; 320 321 if (i > xdp_ring->next_rs) { 322 tx_desc = ICE_TX_DESC(xdp_ring, xdp_ring->next_rs); 323 tx_desc->cmd_type_offset_bsz |= 324 cpu_to_le64(ICE_TX_DESC_CMD_RS << ICE_TXD_QW1_CMD_S); 325 xdp_ring->next_rs += tx_thresh; 326 } 327 328 return ICE_XDP_TX; 329 } 330 331 /** 332 * ice_xmit_xdp_buff - convert an XDP buffer to an XDP frame and send it 333 * @xdp: XDP buffer 334 * @xdp_ring: XDP Tx ring 335 * 336 * Returns negative on failure, 0 on success. 337 */ 338 int ice_xmit_xdp_buff(struct xdp_buff *xdp, struct ice_tx_ring *xdp_ring) 339 { 340 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp); 341 342 if (unlikely(!xdpf)) 343 return ICE_XDP_CONSUMED; 344 345 return ice_xmit_xdp_ring(xdpf->data, xdpf->len, xdp_ring); 346 } 347 348 /** 349 * ice_finalize_xdp_rx - Bump XDP Tx tail and/or flush redirect map 350 * @xdp_ring: XDP ring 351 * @xdp_res: Result of the receive batch 352 * 353 * This function bumps XDP Tx tail and/or flush redirect map, and 354 * should be called when a batch of packets has been processed in the 355 * napi loop. 356 */ 357 void ice_finalize_xdp_rx(struct ice_tx_ring *xdp_ring, unsigned int xdp_res) 358 { 359 if (xdp_res & ICE_XDP_REDIR) 360 xdp_do_flush_map(); 361 362 if (xdp_res & ICE_XDP_TX) { 363 if (static_branch_unlikely(&ice_xdp_locking_key)) 364 spin_lock(&xdp_ring->tx_lock); 365 ice_xdp_ring_update_tail(xdp_ring); 366 if (static_branch_unlikely(&ice_xdp_locking_key)) 367 spin_unlock(&xdp_ring->tx_lock); 368 } 369 } 370