1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2019, Intel Corporation. */ 3 4 #include <linux/bpf_trace.h> 5 #include <net/xdp_sock_drv.h> 6 #include <net/xdp.h> 7 #include "ice.h" 8 #include "ice_base.h" 9 #include "ice_type.h" 10 #include "ice_xsk.h" 11 #include "ice_txrx.h" 12 #include "ice_txrx_lib.h" 13 #include "ice_lib.h" 14 15 /** 16 * ice_qp_reset_stats - Resets all stats for rings of given index 17 * @vsi: VSI that contains rings of interest 18 * @q_idx: ring index in array 19 */ 20 static void ice_qp_reset_stats(struct ice_vsi *vsi, u16 q_idx) 21 { 22 memset(&vsi->rx_rings[q_idx]->rx_stats, 0, 23 sizeof(vsi->rx_rings[q_idx]->rx_stats)); 24 memset(&vsi->tx_rings[q_idx]->stats, 0, 25 sizeof(vsi->tx_rings[q_idx]->stats)); 26 if (ice_is_xdp_ena_vsi(vsi)) 27 memset(&vsi->xdp_rings[q_idx]->stats, 0, 28 sizeof(vsi->xdp_rings[q_idx]->stats)); 29 } 30 31 /** 32 * ice_qp_clean_rings - Cleans all the rings of a given index 33 * @vsi: VSI that contains rings of interest 34 * @q_idx: ring index in array 35 */ 36 static void ice_qp_clean_rings(struct ice_vsi *vsi, u16 q_idx) 37 { 38 ice_clean_tx_ring(vsi->tx_rings[q_idx]); 39 if (ice_is_xdp_ena_vsi(vsi)) 40 ice_clean_tx_ring(vsi->xdp_rings[q_idx]); 41 ice_clean_rx_ring(vsi->rx_rings[q_idx]); 42 } 43 44 /** 45 * ice_qvec_toggle_napi - Enables/disables NAPI for a given q_vector 46 * @vsi: VSI that has netdev 47 * @q_vector: q_vector that has NAPI context 48 * @enable: true for enable, false for disable 49 */ 50 static void 51 ice_qvec_toggle_napi(struct ice_vsi *vsi, struct ice_q_vector *q_vector, 52 bool enable) 53 { 54 if (!vsi->netdev || !q_vector) 55 return; 56 57 if (enable) 58 napi_enable(&q_vector->napi); 59 else 60 napi_disable(&q_vector->napi); 61 } 62 63 /** 64 * ice_qvec_dis_irq - Mask off queue interrupt generation on given ring 65 * @vsi: the VSI that contains queue vector being un-configured 66 * @rx_ring: Rx ring that will have its IRQ disabled 67 * @q_vector: queue vector 68 */ 69 static void 70 ice_qvec_dis_irq(struct ice_vsi *vsi, struct ice_ring *rx_ring, 71 struct ice_q_vector *q_vector) 72 { 73 struct ice_pf *pf = vsi->back; 74 struct ice_hw *hw = &pf->hw; 75 int base = vsi->base_vector; 76 u16 reg; 77 u32 val; 78 79 /* QINT_TQCTL is being cleared in ice_vsi_stop_tx_ring, so handle 80 * here only QINT_RQCTL 81 */ 82 reg = rx_ring->reg_idx; 83 val = rd32(hw, QINT_RQCTL(reg)); 84 val &= ~QINT_RQCTL_CAUSE_ENA_M; 85 wr32(hw, QINT_RQCTL(reg), val); 86 87 if (q_vector) { 88 u16 v_idx = q_vector->v_idx; 89 90 wr32(hw, GLINT_DYN_CTL(q_vector->reg_idx), 0); 91 ice_flush(hw); 92 synchronize_irq(pf->msix_entries[v_idx + base].vector); 93 } 94 } 95 96 /** 97 * ice_qvec_cfg_msix - Enable IRQ for given queue vector 98 * @vsi: the VSI that contains queue vector 99 * @q_vector: queue vector 100 */ 101 static void 102 ice_qvec_cfg_msix(struct ice_vsi *vsi, struct ice_q_vector *q_vector) 103 { 104 u16 reg_idx = q_vector->reg_idx; 105 struct ice_pf *pf = vsi->back; 106 struct ice_hw *hw = &pf->hw; 107 struct ice_ring *ring; 108 109 ice_cfg_itr(hw, q_vector); 110 111 ice_for_each_ring(ring, q_vector->tx) 112 ice_cfg_txq_interrupt(vsi, ring->reg_idx, reg_idx, 113 q_vector->tx.itr_idx); 114 115 ice_for_each_ring(ring, q_vector->rx) 116 ice_cfg_rxq_interrupt(vsi, ring->reg_idx, reg_idx, 117 q_vector->rx.itr_idx); 118 119 ice_flush(hw); 120 } 121 122 /** 123 * ice_qvec_ena_irq - Enable IRQ for given queue vector 124 * @vsi: the VSI that contains queue vector 125 * @q_vector: queue vector 126 */ 127 static void ice_qvec_ena_irq(struct ice_vsi *vsi, struct ice_q_vector *q_vector) 128 { 129 struct ice_pf *pf = vsi->back; 130 struct ice_hw *hw = &pf->hw; 131 132 ice_irq_dynamic_ena(hw, vsi, q_vector); 133 134 ice_flush(hw); 135 } 136 137 /** 138 * ice_qp_dis - Disables a queue pair 139 * @vsi: VSI of interest 140 * @q_idx: ring index in array 141 * 142 * Returns 0 on success, negative on failure. 143 */ 144 static int ice_qp_dis(struct ice_vsi *vsi, u16 q_idx) 145 { 146 struct ice_txq_meta txq_meta = { }; 147 struct ice_ring *tx_ring, *rx_ring; 148 struct ice_q_vector *q_vector; 149 int timeout = 50; 150 int err; 151 152 if (q_idx >= vsi->num_rxq || q_idx >= vsi->num_txq) 153 return -EINVAL; 154 155 tx_ring = vsi->tx_rings[q_idx]; 156 rx_ring = vsi->rx_rings[q_idx]; 157 q_vector = rx_ring->q_vector; 158 159 while (test_and_set_bit(ICE_CFG_BUSY, vsi->state)) { 160 timeout--; 161 if (!timeout) 162 return -EBUSY; 163 usleep_range(1000, 2000); 164 } 165 netif_tx_stop_queue(netdev_get_tx_queue(vsi->netdev, q_idx)); 166 167 ice_qvec_dis_irq(vsi, rx_ring, q_vector); 168 169 ice_fill_txq_meta(vsi, tx_ring, &txq_meta); 170 err = ice_vsi_stop_tx_ring(vsi, ICE_NO_RESET, 0, tx_ring, &txq_meta); 171 if (err) 172 return err; 173 if (ice_is_xdp_ena_vsi(vsi)) { 174 struct ice_ring *xdp_ring = vsi->xdp_rings[q_idx]; 175 176 memset(&txq_meta, 0, sizeof(txq_meta)); 177 ice_fill_txq_meta(vsi, xdp_ring, &txq_meta); 178 err = ice_vsi_stop_tx_ring(vsi, ICE_NO_RESET, 0, xdp_ring, 179 &txq_meta); 180 if (err) 181 return err; 182 } 183 err = ice_vsi_ctrl_one_rx_ring(vsi, false, q_idx, true); 184 if (err) 185 return err; 186 187 ice_qvec_toggle_napi(vsi, q_vector, false); 188 ice_qp_clean_rings(vsi, q_idx); 189 ice_qp_reset_stats(vsi, q_idx); 190 191 return 0; 192 } 193 194 /** 195 * ice_qp_ena - Enables a queue pair 196 * @vsi: VSI of interest 197 * @q_idx: ring index in array 198 * 199 * Returns 0 on success, negative on failure. 200 */ 201 static int ice_qp_ena(struct ice_vsi *vsi, u16 q_idx) 202 { 203 struct ice_aqc_add_tx_qgrp *qg_buf; 204 struct ice_ring *tx_ring, *rx_ring; 205 struct ice_q_vector *q_vector; 206 u16 size; 207 int err; 208 209 if (q_idx >= vsi->num_rxq || q_idx >= vsi->num_txq) 210 return -EINVAL; 211 212 size = struct_size(qg_buf, txqs, 1); 213 qg_buf = kzalloc(size, GFP_KERNEL); 214 if (!qg_buf) 215 return -ENOMEM; 216 217 qg_buf->num_txqs = 1; 218 219 tx_ring = vsi->tx_rings[q_idx]; 220 rx_ring = vsi->rx_rings[q_idx]; 221 q_vector = rx_ring->q_vector; 222 223 err = ice_vsi_cfg_txq(vsi, tx_ring, qg_buf); 224 if (err) 225 goto free_buf; 226 227 if (ice_is_xdp_ena_vsi(vsi)) { 228 struct ice_ring *xdp_ring = vsi->xdp_rings[q_idx]; 229 230 memset(qg_buf, 0, size); 231 qg_buf->num_txqs = 1; 232 err = ice_vsi_cfg_txq(vsi, xdp_ring, qg_buf); 233 if (err) 234 goto free_buf; 235 ice_set_ring_xdp(xdp_ring); 236 xdp_ring->xsk_pool = ice_xsk_pool(xdp_ring); 237 } 238 239 err = ice_setup_rx_ctx(rx_ring); 240 if (err) 241 goto free_buf; 242 243 ice_qvec_cfg_msix(vsi, q_vector); 244 245 err = ice_vsi_ctrl_one_rx_ring(vsi, true, q_idx, true); 246 if (err) 247 goto free_buf; 248 249 clear_bit(ICE_CFG_BUSY, vsi->state); 250 ice_qvec_toggle_napi(vsi, q_vector, true); 251 ice_qvec_ena_irq(vsi, q_vector); 252 253 netif_tx_start_queue(netdev_get_tx_queue(vsi->netdev, q_idx)); 254 free_buf: 255 kfree(qg_buf); 256 return err; 257 } 258 259 /** 260 * ice_xsk_pool_disable - disable a buffer pool region 261 * @vsi: Current VSI 262 * @qid: queue ID 263 * 264 * Returns 0 on success, negative on failure 265 */ 266 static int ice_xsk_pool_disable(struct ice_vsi *vsi, u16 qid) 267 { 268 struct xsk_buff_pool *pool = xsk_get_pool_from_qid(vsi->netdev, qid); 269 270 if (!pool) 271 return -EINVAL; 272 273 xsk_pool_dma_unmap(pool, ICE_RX_DMA_ATTR); 274 275 return 0; 276 } 277 278 /** 279 * ice_xsk_pool_enable - enable a buffer pool region 280 * @vsi: Current VSI 281 * @pool: pointer to a requested buffer pool region 282 * @qid: queue ID 283 * 284 * Returns 0 on success, negative on failure 285 */ 286 static int 287 ice_xsk_pool_enable(struct ice_vsi *vsi, struct xsk_buff_pool *pool, u16 qid) 288 { 289 int err; 290 291 if (vsi->type != ICE_VSI_PF) 292 return -EINVAL; 293 294 if (qid >= vsi->netdev->real_num_rx_queues || 295 qid >= vsi->netdev->real_num_tx_queues) 296 return -EINVAL; 297 298 err = xsk_pool_dma_map(pool, ice_pf_to_dev(vsi->back), 299 ICE_RX_DMA_ATTR); 300 if (err) 301 return err; 302 303 return 0; 304 } 305 306 /** 307 * ice_xsk_pool_setup - enable/disable a buffer pool region depending on its state 308 * @vsi: Current VSI 309 * @pool: buffer pool to enable/associate to a ring, NULL to disable 310 * @qid: queue ID 311 * 312 * Returns 0 on success, negative on failure 313 */ 314 int ice_xsk_pool_setup(struct ice_vsi *vsi, struct xsk_buff_pool *pool, u16 qid) 315 { 316 bool if_running, pool_present = !!pool; 317 int ret = 0, pool_failure = 0; 318 319 if_running = netif_running(vsi->netdev) && ice_is_xdp_ena_vsi(vsi); 320 321 if (if_running) { 322 ret = ice_qp_dis(vsi, qid); 323 if (ret) { 324 netdev_err(vsi->netdev, "ice_qp_dis error = %d\n", ret); 325 goto xsk_pool_if_up; 326 } 327 } 328 329 pool_failure = pool_present ? ice_xsk_pool_enable(vsi, pool, qid) : 330 ice_xsk_pool_disable(vsi, qid); 331 332 xsk_pool_if_up: 333 if (if_running) { 334 ret = ice_qp_ena(vsi, qid); 335 if (!ret && pool_present) 336 napi_schedule(&vsi->xdp_rings[qid]->q_vector->napi); 337 else if (ret) 338 netdev_err(vsi->netdev, "ice_qp_ena error = %d\n", ret); 339 } 340 341 if (pool_failure) { 342 netdev_err(vsi->netdev, "Could not %sable buffer pool, error = %d\n", 343 pool_present ? "en" : "dis", pool_failure); 344 return pool_failure; 345 } 346 347 return ret; 348 } 349 350 /** 351 * ice_alloc_rx_bufs_zc - allocate a number of Rx buffers 352 * @rx_ring: Rx ring 353 * @count: The number of buffers to allocate 354 * 355 * This function allocates a number of Rx buffers from the fill ring 356 * or the internal recycle mechanism and places them on the Rx ring. 357 * 358 * Returns true if all allocations were successful, false if any fail. 359 */ 360 bool ice_alloc_rx_bufs_zc(struct ice_ring *rx_ring, u16 count) 361 { 362 union ice_32b_rx_flex_desc *rx_desc; 363 u16 ntu = rx_ring->next_to_use; 364 struct ice_rx_buf *rx_buf; 365 bool ok = true; 366 dma_addr_t dma; 367 368 if (!count) 369 return true; 370 371 rx_desc = ICE_RX_DESC(rx_ring, ntu); 372 rx_buf = &rx_ring->rx_buf[ntu]; 373 374 do { 375 rx_buf->xdp = xsk_buff_alloc(rx_ring->xsk_pool); 376 if (!rx_buf->xdp) { 377 ok = false; 378 break; 379 } 380 381 dma = xsk_buff_xdp_get_dma(rx_buf->xdp); 382 rx_desc->read.pkt_addr = cpu_to_le64(dma); 383 rx_desc->wb.status_error0 = 0; 384 385 rx_desc++; 386 rx_buf++; 387 ntu++; 388 389 if (unlikely(ntu == rx_ring->count)) { 390 rx_desc = ICE_RX_DESC(rx_ring, 0); 391 rx_buf = rx_ring->rx_buf; 392 ntu = 0; 393 } 394 } while (--count); 395 396 if (rx_ring->next_to_use != ntu) { 397 /* clear the status bits for the next_to_use descriptor */ 398 rx_desc->wb.status_error0 = 0; 399 ice_release_rx_desc(rx_ring, ntu); 400 } 401 402 return ok; 403 } 404 405 /** 406 * ice_bump_ntc - Bump the next_to_clean counter of an Rx ring 407 * @rx_ring: Rx ring 408 */ 409 static void ice_bump_ntc(struct ice_ring *rx_ring) 410 { 411 int ntc = rx_ring->next_to_clean + 1; 412 413 ntc = (ntc < rx_ring->count) ? ntc : 0; 414 rx_ring->next_to_clean = ntc; 415 prefetch(ICE_RX_DESC(rx_ring, ntc)); 416 } 417 418 /** 419 * ice_construct_skb_zc - Create an sk_buff from zero-copy buffer 420 * @rx_ring: Rx ring 421 * @rx_buf: zero-copy Rx buffer 422 * 423 * This function allocates a new skb from a zero-copy Rx buffer. 424 * 425 * Returns the skb on success, NULL on failure. 426 */ 427 static struct sk_buff * 428 ice_construct_skb_zc(struct ice_ring *rx_ring, struct ice_rx_buf *rx_buf) 429 { 430 unsigned int metasize = rx_buf->xdp->data - rx_buf->xdp->data_meta; 431 unsigned int datasize = rx_buf->xdp->data_end - rx_buf->xdp->data; 432 unsigned int datasize_hard = rx_buf->xdp->data_end - 433 rx_buf->xdp->data_hard_start; 434 struct sk_buff *skb; 435 436 skb = __napi_alloc_skb(&rx_ring->q_vector->napi, datasize_hard, 437 GFP_ATOMIC | __GFP_NOWARN); 438 if (unlikely(!skb)) 439 return NULL; 440 441 skb_reserve(skb, rx_buf->xdp->data - rx_buf->xdp->data_hard_start); 442 memcpy(__skb_put(skb, datasize), rx_buf->xdp->data, datasize); 443 if (metasize) 444 skb_metadata_set(skb, metasize); 445 446 xsk_buff_free(rx_buf->xdp); 447 rx_buf->xdp = NULL; 448 return skb; 449 } 450 451 /** 452 * ice_run_xdp_zc - Executes an XDP program in zero-copy path 453 * @rx_ring: Rx ring 454 * @xdp: xdp_buff used as input to the XDP program 455 * 456 * Returns any of ICE_XDP_{PASS, CONSUMED, TX, REDIR} 457 */ 458 static int 459 ice_run_xdp_zc(struct ice_ring *rx_ring, struct xdp_buff *xdp) 460 { 461 int err, result = ICE_XDP_PASS; 462 struct bpf_prog *xdp_prog; 463 struct ice_ring *xdp_ring; 464 u32 act; 465 466 rcu_read_lock(); 467 /* ZC patch is enabled only when XDP program is set, 468 * so here it can not be NULL 469 */ 470 xdp_prog = READ_ONCE(rx_ring->xdp_prog); 471 472 act = bpf_prog_run_xdp(xdp_prog, xdp); 473 474 if (likely(act == XDP_REDIRECT)) { 475 err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog); 476 result = !err ? ICE_XDP_REDIR : ICE_XDP_CONSUMED; 477 rcu_read_unlock(); 478 return result; 479 } 480 481 switch (act) { 482 case XDP_PASS: 483 break; 484 case XDP_TX: 485 xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->q_index]; 486 result = ice_xmit_xdp_buff(xdp, xdp_ring); 487 break; 488 default: 489 bpf_warn_invalid_xdp_action(act); 490 fallthrough; 491 case XDP_ABORTED: 492 trace_xdp_exception(rx_ring->netdev, xdp_prog, act); 493 fallthrough; 494 case XDP_DROP: 495 result = ICE_XDP_CONSUMED; 496 break; 497 } 498 499 rcu_read_unlock(); 500 return result; 501 } 502 503 /** 504 * ice_clean_rx_irq_zc - consumes packets from the hardware ring 505 * @rx_ring: AF_XDP Rx ring 506 * @budget: NAPI budget 507 * 508 * Returns number of processed packets on success, remaining budget on failure. 509 */ 510 int ice_clean_rx_irq_zc(struct ice_ring *rx_ring, int budget) 511 { 512 unsigned int total_rx_bytes = 0, total_rx_packets = 0; 513 u16 cleaned_count = ICE_DESC_UNUSED(rx_ring); 514 unsigned int xdp_xmit = 0; 515 bool failure = false; 516 517 while (likely(total_rx_packets < (unsigned int)budget)) { 518 union ice_32b_rx_flex_desc *rx_desc; 519 unsigned int size, xdp_res = 0; 520 struct ice_rx_buf *rx_buf; 521 struct sk_buff *skb; 522 u16 stat_err_bits; 523 u16 vlan_tag = 0; 524 u8 rx_ptype; 525 526 rx_desc = ICE_RX_DESC(rx_ring, rx_ring->next_to_clean); 527 528 stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S); 529 if (!ice_test_staterr(rx_desc, stat_err_bits)) 530 break; 531 532 /* This memory barrier is needed to keep us from reading 533 * any other fields out of the rx_desc until we have 534 * verified the descriptor has been written back. 535 */ 536 dma_rmb(); 537 538 size = le16_to_cpu(rx_desc->wb.pkt_len) & 539 ICE_RX_FLX_DESC_PKT_LEN_M; 540 if (!size) 541 break; 542 543 rx_buf = &rx_ring->rx_buf[rx_ring->next_to_clean]; 544 rx_buf->xdp->data_end = rx_buf->xdp->data + size; 545 xsk_buff_dma_sync_for_cpu(rx_buf->xdp, rx_ring->xsk_pool); 546 547 xdp_res = ice_run_xdp_zc(rx_ring, rx_buf->xdp); 548 if (xdp_res) { 549 if (xdp_res & (ICE_XDP_TX | ICE_XDP_REDIR)) 550 xdp_xmit |= xdp_res; 551 else 552 xsk_buff_free(rx_buf->xdp); 553 554 rx_buf->xdp = NULL; 555 total_rx_bytes += size; 556 total_rx_packets++; 557 cleaned_count++; 558 559 ice_bump_ntc(rx_ring); 560 continue; 561 } 562 563 /* XDP_PASS path */ 564 skb = ice_construct_skb_zc(rx_ring, rx_buf); 565 if (!skb) { 566 rx_ring->rx_stats.alloc_buf_failed++; 567 break; 568 } 569 570 cleaned_count++; 571 ice_bump_ntc(rx_ring); 572 573 if (eth_skb_pad(skb)) { 574 skb = NULL; 575 continue; 576 } 577 578 total_rx_bytes += skb->len; 579 total_rx_packets++; 580 581 stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_L2TAG1P_S); 582 if (ice_test_staterr(rx_desc, stat_err_bits)) 583 vlan_tag = le16_to_cpu(rx_desc->wb.l2tag1); 584 585 rx_ptype = le16_to_cpu(rx_desc->wb.ptype_flex_flags0) & 586 ICE_RX_FLEX_DESC_PTYPE_M; 587 588 ice_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype); 589 ice_receive_skb(rx_ring, skb, vlan_tag); 590 } 591 592 if (cleaned_count >= ICE_RX_BUF_WRITE) 593 failure = !ice_alloc_rx_bufs_zc(rx_ring, cleaned_count); 594 595 ice_finalize_xdp_rx(rx_ring, xdp_xmit); 596 ice_update_rx_ring_stats(rx_ring, total_rx_packets, total_rx_bytes); 597 598 if (xsk_uses_need_wakeup(rx_ring->xsk_pool)) { 599 if (failure || rx_ring->next_to_clean == rx_ring->next_to_use) 600 xsk_set_rx_need_wakeup(rx_ring->xsk_pool); 601 else 602 xsk_clear_rx_need_wakeup(rx_ring->xsk_pool); 603 604 return (int)total_rx_packets; 605 } 606 607 return failure ? budget : (int)total_rx_packets; 608 } 609 610 /** 611 * ice_xmit_zc - Completes AF_XDP entries, and cleans XDP entries 612 * @xdp_ring: XDP Tx ring 613 * @budget: max number of frames to xmit 614 * 615 * Returns true if cleanup/transmission is done. 616 */ 617 static bool ice_xmit_zc(struct ice_ring *xdp_ring, int budget) 618 { 619 struct ice_tx_desc *tx_desc = NULL; 620 bool work_done = true; 621 struct xdp_desc desc; 622 dma_addr_t dma; 623 624 while (likely(budget-- > 0)) { 625 struct ice_tx_buf *tx_buf; 626 627 if (unlikely(!ICE_DESC_UNUSED(xdp_ring))) { 628 xdp_ring->tx_stats.tx_busy++; 629 work_done = false; 630 break; 631 } 632 633 tx_buf = &xdp_ring->tx_buf[xdp_ring->next_to_use]; 634 635 if (!xsk_tx_peek_desc(xdp_ring->xsk_pool, &desc)) 636 break; 637 638 dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc.addr); 639 xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, 640 desc.len); 641 642 tx_buf->bytecount = desc.len; 643 644 tx_desc = ICE_TX_DESC(xdp_ring, xdp_ring->next_to_use); 645 tx_desc->buf_addr = cpu_to_le64(dma); 646 tx_desc->cmd_type_offset_bsz = 647 ice_build_ctob(ICE_TXD_LAST_DESC_CMD, 0, desc.len, 0); 648 649 xdp_ring->next_to_use++; 650 if (xdp_ring->next_to_use == xdp_ring->count) 651 xdp_ring->next_to_use = 0; 652 } 653 654 if (tx_desc) { 655 ice_xdp_ring_update_tail(xdp_ring); 656 xsk_tx_release(xdp_ring->xsk_pool); 657 } 658 659 return budget > 0 && work_done; 660 } 661 662 /** 663 * ice_clean_xdp_tx_buf - Free and unmap XDP Tx buffer 664 * @xdp_ring: XDP Tx ring 665 * @tx_buf: Tx buffer to clean 666 */ 667 static void 668 ice_clean_xdp_tx_buf(struct ice_ring *xdp_ring, struct ice_tx_buf *tx_buf) 669 { 670 xdp_return_frame((struct xdp_frame *)tx_buf->raw_buf); 671 dma_unmap_single(xdp_ring->dev, dma_unmap_addr(tx_buf, dma), 672 dma_unmap_len(tx_buf, len), DMA_TO_DEVICE); 673 dma_unmap_len_set(tx_buf, len, 0); 674 } 675 676 /** 677 * ice_clean_tx_irq_zc - Completes AF_XDP entries, and cleans XDP entries 678 * @xdp_ring: XDP Tx ring 679 * @budget: NAPI budget 680 * 681 * Returns true if cleanup/tranmission is done. 682 */ 683 bool ice_clean_tx_irq_zc(struct ice_ring *xdp_ring, int budget) 684 { 685 int total_packets = 0, total_bytes = 0; 686 s16 ntc = xdp_ring->next_to_clean; 687 struct ice_tx_desc *tx_desc; 688 struct ice_tx_buf *tx_buf; 689 u32 xsk_frames = 0; 690 bool xmit_done; 691 692 tx_desc = ICE_TX_DESC(xdp_ring, ntc); 693 tx_buf = &xdp_ring->tx_buf[ntc]; 694 ntc -= xdp_ring->count; 695 696 do { 697 if (!(tx_desc->cmd_type_offset_bsz & 698 cpu_to_le64(ICE_TX_DESC_DTYPE_DESC_DONE))) 699 break; 700 701 total_bytes += tx_buf->bytecount; 702 total_packets++; 703 704 if (tx_buf->raw_buf) { 705 ice_clean_xdp_tx_buf(xdp_ring, tx_buf); 706 tx_buf->raw_buf = NULL; 707 } else { 708 xsk_frames++; 709 } 710 711 tx_desc->cmd_type_offset_bsz = 0; 712 tx_buf++; 713 tx_desc++; 714 ntc++; 715 716 if (unlikely(!ntc)) { 717 ntc -= xdp_ring->count; 718 tx_buf = xdp_ring->tx_buf; 719 tx_desc = ICE_TX_DESC(xdp_ring, 0); 720 } 721 722 prefetch(tx_desc); 723 724 } while (likely(--budget)); 725 726 ntc += xdp_ring->count; 727 xdp_ring->next_to_clean = ntc; 728 729 if (xsk_frames) 730 xsk_tx_completed(xdp_ring->xsk_pool, xsk_frames); 731 732 if (xsk_uses_need_wakeup(xdp_ring->xsk_pool)) 733 xsk_set_tx_need_wakeup(xdp_ring->xsk_pool); 734 735 ice_update_tx_ring_stats(xdp_ring, total_packets, total_bytes); 736 xmit_done = ice_xmit_zc(xdp_ring, ICE_DFLT_IRQ_WORK); 737 738 return budget > 0 && xmit_done; 739 } 740 741 /** 742 * ice_xsk_wakeup - Implements ndo_xsk_wakeup 743 * @netdev: net_device 744 * @queue_id: queue to wake up 745 * @flags: ignored in our case, since we have Rx and Tx in the same NAPI 746 * 747 * Returns negative on error, zero otherwise. 748 */ 749 int 750 ice_xsk_wakeup(struct net_device *netdev, u32 queue_id, 751 u32 __always_unused flags) 752 { 753 struct ice_netdev_priv *np = netdev_priv(netdev); 754 struct ice_q_vector *q_vector; 755 struct ice_vsi *vsi = np->vsi; 756 struct ice_ring *ring; 757 758 if (test_bit(ICE_DOWN, vsi->state)) 759 return -ENETDOWN; 760 761 if (!ice_is_xdp_ena_vsi(vsi)) 762 return -ENXIO; 763 764 if (queue_id >= vsi->num_txq) 765 return -ENXIO; 766 767 if (!vsi->xdp_rings[queue_id]->xsk_pool) 768 return -ENXIO; 769 770 ring = vsi->xdp_rings[queue_id]; 771 772 /* The idea here is that if NAPI is running, mark a miss, so 773 * it will run again. If not, trigger an interrupt and 774 * schedule the NAPI from interrupt context. If NAPI would be 775 * scheduled here, the interrupt affinity would not be 776 * honored. 777 */ 778 q_vector = ring->q_vector; 779 if (!napi_if_scheduled_mark_missed(&q_vector->napi)) 780 ice_trigger_sw_intr(&vsi->back->hw, q_vector); 781 782 return 0; 783 } 784 785 /** 786 * ice_xsk_any_rx_ring_ena - Checks if Rx rings have AF_XDP buff pool attached 787 * @vsi: VSI to be checked 788 * 789 * Returns true if any of the Rx rings has an AF_XDP buff pool attached 790 */ 791 bool ice_xsk_any_rx_ring_ena(struct ice_vsi *vsi) 792 { 793 int i; 794 795 ice_for_each_rxq(vsi, i) { 796 if (xsk_get_pool_from_qid(vsi->netdev, i)) 797 return true; 798 } 799 800 return false; 801 } 802 803 /** 804 * ice_xsk_clean_rx_ring - clean buffer pool queues connected to a given Rx ring 805 * @rx_ring: ring to be cleaned 806 */ 807 void ice_xsk_clean_rx_ring(struct ice_ring *rx_ring) 808 { 809 u16 i; 810 811 for (i = 0; i < rx_ring->count; i++) { 812 struct ice_rx_buf *rx_buf = &rx_ring->rx_buf[i]; 813 814 if (!rx_buf->xdp) 815 continue; 816 817 rx_buf->xdp = NULL; 818 } 819 } 820 821 /** 822 * ice_xsk_clean_xdp_ring - Clean the XDP Tx ring and its buffer pool queues 823 * @xdp_ring: XDP_Tx ring 824 */ 825 void ice_xsk_clean_xdp_ring(struct ice_ring *xdp_ring) 826 { 827 u16 ntc = xdp_ring->next_to_clean, ntu = xdp_ring->next_to_use; 828 u32 xsk_frames = 0; 829 830 while (ntc != ntu) { 831 struct ice_tx_buf *tx_buf = &xdp_ring->tx_buf[ntc]; 832 833 if (tx_buf->raw_buf) 834 ice_clean_xdp_tx_buf(xdp_ring, tx_buf); 835 else 836 xsk_frames++; 837 838 tx_buf->raw_buf = NULL; 839 840 ntc++; 841 if (ntc >= xdp_ring->count) 842 ntc = 0; 843 } 844 845 if (xsk_frames) 846 xsk_tx_completed(xdp_ring->xsk_pool, xsk_frames); 847 } 848