drivers/net/ethernet/intel/i40e/i40e_xsk.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 2018 Intel Corporation. */
   3
   4 #include <linux/bpf_trace.h>
   5 #include <linux/stringify.h>
   6 #include <net/xdp_sock_drv.h>
   7 #include <net/xdp.h>
   8
   9 #include "i40e.h"
  10 #include "i40e_txrx_common.h"
  11 #include "i40e_xsk.h"
  12
  13 int i40e_alloc_rx_bi_zc(struct i40e_ring *rx_ring)
  14 {
  15         unsigned long sz = sizeof(*rx_ring->rx_bi_zc) * rx_ring->count;
  16
  17         rx_ring->rx_bi_zc = kzalloc(sz, GFP_KERNEL);
  18         return rx_ring->rx_bi_zc ? 0 : -ENOMEM;
  19 }
  20
  21 void i40e_clear_rx_bi_zc(struct i40e_ring *rx_ring)
  22 {
  23         memset(rx_ring->rx_bi_zc, 0,
  24                sizeof(*rx_ring->rx_bi_zc) * rx_ring->count);
  25 }
  26
  27 static struct xdp_buff **i40e_rx_bi(struct i40e_ring *rx_ring, u32 idx)
  28 {
  29         return &rx_ring->rx_bi_zc[idx];
  30 }
  31
  32 /**
  33  * i40e_xsk_pool_enable - Enable/associate an AF_XDP buffer pool to a
  34  * certain ring/qid
  35  * @vsi: Current VSI
  36  * @pool: buffer pool
  37  * @qid: Rx ring to associate buffer pool with
  38  *
  39  * Returns 0 on success, <0 on failure
  40  **/
  41 static int i40e_xsk_pool_enable(struct i40e_vsi *vsi,
  42                                 struct xsk_buff_pool *pool,
  43                                 u16 qid)
  44 {
  45         struct net_device *netdev = vsi->netdev;
  46         bool if_running;
  47         int err;
  48
  49         if (vsi->type != I40E_VSI_MAIN)
  50                 return -EINVAL;
  51
  52         if (qid >= vsi->num_queue_pairs)
  53                 return -EINVAL;
  54
  55         if (qid >= netdev->real_num_rx_queues ||
  56             qid >= netdev->real_num_tx_queues)
  57                 return -EINVAL;
  58
  59         err = xsk_pool_dma_map(pool, &vsi->back->pdev->dev, I40E_RX_DMA_ATTR);
  60         if (err)
  61                 return err;
  62
  63         set_bit(qid, vsi->af_xdp_zc_qps);
  64
  65         if_running = netif_running(vsi->netdev) && i40e_enabled_xdp_vsi(vsi);
  66
  67         if (if_running) {
  68                 err = i40e_queue_pair_disable(vsi, qid);
  69                 if (err)
  70                         return err;
  71
  72                 err = i40e_queue_pair_enable(vsi, qid);
  73                 if (err)
  74                         return err;
  75
  76                 /* Kick start the NAPI context so that receiving will start */
  77                 err = i40e_xsk_wakeup(vsi->netdev, qid, XDP_WAKEUP_RX);
  78                 if (err)
  79                         return err;
  80         }
  81
  82         return 0;
  83 }
  84
  85 /**
  86  * i40e_xsk_pool_disable - Disassociate an AF_XDP buffer pool from a
  87  * certain ring/qid
  88  * @vsi: Current VSI
  89  * @qid: Rx ring to associate buffer pool with
  90  *
  91  * Returns 0 on success, <0 on failure
  92  **/
  93 static int i40e_xsk_pool_disable(struct i40e_vsi *vsi, u16 qid)
  94 {
  95         struct net_device *netdev = vsi->netdev;
  96         struct xsk_buff_pool *pool;
  97         bool if_running;
  98         int err;
  99
 100         pool = xsk_get_pool_from_qid(netdev, qid);
 101         if (!pool)
 102                 return -EINVAL;
 103
 104         if_running = netif_running(vsi->netdev) && i40e_enabled_xdp_vsi(vsi);
 105
 106         if (if_running) {
 107                 err = i40e_queue_pair_disable(vsi, qid);
 108                 if (err)
 109                         return err;
 110         }
 111
 112         clear_bit(qid, vsi->af_xdp_zc_qps);
 113         xsk_pool_dma_unmap(pool, I40E_RX_DMA_ATTR);
 114
 115         if (if_running) {
 116                 err = i40e_queue_pair_enable(vsi, qid);
 117                 if (err)
 118                         return err;
 119         }
 120
 121         return 0;
 122 }
 123
 124 /**
 125  * i40e_xsk_pool_setup - Enable/disassociate an AF_XDP buffer pool to/from
 126  * a ring/qid
 127  * @vsi: Current VSI
 128  * @pool: Buffer pool to enable/associate to a ring, or NULL to disable
 129  * @qid: Rx ring to (dis)associate buffer pool (from)to
 130  *
 131  * This function enables or disables a buffer pool to a certain ring.
 132  *
 133  * Returns 0 on success, <0 on failure
 134  **/
 135 int i40e_xsk_pool_setup(struct i40e_vsi *vsi, struct xsk_buff_pool *pool,
 136                         u16 qid)
 137 {
 138         return pool ? i40e_xsk_pool_enable(vsi, pool, qid) :
 139                 i40e_xsk_pool_disable(vsi, qid);
 140 }
 141
 142 /**
 143  * i40e_run_xdp_zc - Executes an XDP program on an xdp_buff
 144  * @rx_ring: Rx ring
 145  * @xdp: xdp_buff used as input to the XDP program
 146  *
 147  * Returns any of I40E_XDP_{PASS, CONSUMED, TX, REDIR}
 148  **/
 149 static int i40e_run_xdp_zc(struct i40e_ring *rx_ring, struct xdp_buff *xdp)
 150 {
 151         int err, result = I40E_XDP_PASS;
 152         struct i40e_ring *xdp_ring;
 153         struct bpf_prog *xdp_prog;
 154         u32 act;
 155
 156         rcu_read_lock();
 157         /* NB! xdp_prog will always be !NULL, due to the fact that
 158          * this path is enabled by setting an XDP program.
 159          */
 160         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
 161         act = bpf_prog_run_xdp(xdp_prog, xdp);
 162
 163         switch (act) {
 164         case XDP_PASS:
 165                 break;
 166         case XDP_TX:
 167                 xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
 168                 result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
 169                 break;
 170         case XDP_REDIRECT:
 171                 err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
 172                 result = !err ? I40E_XDP_REDIR : I40E_XDP_CONSUMED;
 173                 break;
 174         default:
 175                 bpf_warn_invalid_xdp_action(act);
 176                 fallthrough;
 177         case XDP_ABORTED:
 178                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
 179                 fallthrough; /* handle aborts by dropping packet */
 180         case XDP_DROP:
 181                 result = I40E_XDP_CONSUMED;
 182                 break;
 183         }
 184         rcu_read_unlock();
 185         return result;
 186 }
 187
 188 bool i40e_alloc_rx_buffers_zc(struct i40e_ring *rx_ring, u16 count)
 189 {
 190         u16 ntu = rx_ring->next_to_use;
 191         union i40e_rx_desc *rx_desc;
 192         struct xdp_buff **bi, *xdp;
 193         dma_addr_t dma;
 194         bool ok = true;
 195
 196         rx_desc = I40E_RX_DESC(rx_ring, ntu);
 197         bi = i40e_rx_bi(rx_ring, ntu);
 198         do {
 199                 xdp = xsk_buff_alloc(rx_ring->xsk_pool);
 200                 if (!xdp) {
 201                         ok = false;
 202                         goto no_buffers;
 203                 }
 204                 *bi = xdp;
 205                 dma = xsk_buff_xdp_get_dma(xdp);
 206                 rx_desc->read.pkt_addr = cpu_to_le64(dma);
 207                 rx_desc->read.hdr_addr = 0;
 208
 209                 rx_desc++;
 210                 bi++;
 211                 ntu++;
 212
 213                 if (unlikely(ntu == rx_ring->count)) {
 214                         rx_desc = I40E_RX_DESC(rx_ring, 0);
 215                         bi = i40e_rx_bi(rx_ring, 0);
 216                         ntu = 0;
 217                 }
 218         } while (--count);
 219
 220 no_buffers:
 221         if (rx_ring->next_to_use != ntu) {
 222                 /* clear the status bits for the next_to_use descriptor */
 223                 rx_desc->wb.qword1.status_error_len = 0;
 224                 i40e_release_rx_desc(rx_ring, ntu);
 225         }
 226
 227         return ok;
 228 }
 229
 230 /**
 231  * i40e_construct_skb_zc - Create skbuff from zero-copy Rx buffer
 232  * @rx_ring: Rx ring
 233  * @xdp: xdp_buff
 234  *
 235  * This functions allocates a new skb from a zero-copy Rx buffer.
 236  *
 237  * Returns the skb, or NULL on failure.
 238  **/
 239 static struct sk_buff *i40e_construct_skb_zc(struct i40e_ring *rx_ring,
 240                                              struct xdp_buff *xdp)
 241 {
 242         unsigned int metasize = xdp->data - xdp->data_meta;
 243         unsigned int datasize = xdp->data_end - xdp->data;
 244         struct sk_buff *skb;
 245
 246         /* allocate a skb to store the frags */
 247         skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
 248                                xdp->data_end - xdp->data_hard_start,
 249                                GFP_ATOMIC | __GFP_NOWARN);
 250         if (unlikely(!skb))
 251                 goto out;
 252
 253         skb_reserve(skb, xdp->data - xdp->data_hard_start);
 254         memcpy(__skb_put(skb, datasize), xdp->data, datasize);
 255         if (metasize)
 256                 skb_metadata_set(skb, metasize);
 257
 258 out:
 259         xsk_buff_free(xdp);
 260         return skb;
 261 }
 262
 263 static void i40e_handle_xdp_result_zc(struct i40e_ring *rx_ring,
 264                                       struct xdp_buff *xdp_buff,
 265                                       union i40e_rx_desc *rx_desc,
 266                                       unsigned int *rx_packets,
 267                                       unsigned int *rx_bytes,
 268                                       unsigned int size,
 269                                       unsigned int xdp_res)
 270 {
 271         struct sk_buff *skb;
 272
 273         *rx_packets = 1;
 274         *rx_bytes = size;
 275
 276         if (likely(xdp_res == I40E_XDP_REDIR) || xdp_res == I40E_XDP_TX)
 277                 return;
 278
 279         if (xdp_res == I40E_XDP_CONSUMED) {
 280                 xsk_buff_free(xdp_buff);
 281                 return;
 282         }
 283
 284         if (xdp_res == I40E_XDP_PASS) {
 285                 /* NB! We are not checking for errors using
 286                  * i40e_test_staterr with
 287                  * BIT(I40E_RXD_QW1_ERROR_SHIFT). This is due to that
 288                  * SBP is *not* set in PRT_SBPVSI (default not set).
 289                  */
 290                 skb = i40e_construct_skb_zc(rx_ring, xdp_buff);
 291                 if (!skb) {
 292                         rx_ring->rx_stats.alloc_buff_failed++;
 293                         *rx_packets = 0;
 294                         *rx_bytes = 0;
 295                         return;
 296                 }
 297
 298                 if (eth_skb_pad(skb)) {
 299                         *rx_packets = 0;
 300                         *rx_bytes = 0;
 301                         return;
 302                 }
 303
 304                 *rx_bytes = skb->len;
 305                 i40e_process_skb_fields(rx_ring, rx_desc, skb);
 306                 napi_gro_receive(&rx_ring->q_vector->napi, skb);
 307                 return;
 308         }
 309
 310         /* Should never get here, as all valid cases have been handled already.
 311          */
 312         WARN_ON_ONCE(1);
 313 }
 314
 315 /**
 316  * i40e_clean_rx_irq_zc - Consumes Rx packets from the hardware ring
 317  * @rx_ring: Rx ring
 318  * @budget: NAPI budget
 319  *
 320  * Returns amount of work completed
 321  **/
 322 int i40e_clean_rx_irq_zc(struct i40e_ring *rx_ring, int budget)
 323 {
 324         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 325         u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
 326         u16 next_to_clean = rx_ring->next_to_clean;
 327         u16 count_mask = rx_ring->count - 1;
 328         unsigned int xdp_res, xdp_xmit = 0;
 329         bool failure = false;
 330
 331         while (likely(total_rx_packets < (unsigned int)budget)) {
 332                 union i40e_rx_desc *rx_desc;
 333                 unsigned int rx_packets;
 334                 unsigned int rx_bytes;
 335                 struct xdp_buff *bi;
 336                 unsigned int size;
 337                 u64 qword;
 338
 339                 rx_desc = I40E_RX_DESC(rx_ring, next_to_clean);
 340                 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
 341
 342                 /* This memory barrier is needed to keep us from reading
 343                  * any other fields out of the rx_desc until we have
 344                  * verified the descriptor has been written back.
 345                  */
 346                 dma_rmb();
 347
 348                 if (i40e_rx_is_programming_status(qword)) {
 349                         i40e_clean_programming_status(rx_ring,
 350                                                       rx_desc->raw.qword[0],
 351                                                       qword);
 352                         bi = *i40e_rx_bi(rx_ring, next_to_clean);
 353                         xsk_buff_free(bi);
 354                         next_to_clean = (next_to_clean + 1) & count_mask;
 355                         continue;
 356                 }
 357
 358                 size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
 359                        I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
 360                 if (!size)
 361                         break;
 362
 363                 bi = *i40e_rx_bi(rx_ring, next_to_clean);
 364                 bi->data_end = bi->data + size;
 365                 xsk_buff_dma_sync_for_cpu(bi, rx_ring->xsk_pool);
 366
 367                 xdp_res = i40e_run_xdp_zc(rx_ring, bi);
 368                 i40e_handle_xdp_result_zc(rx_ring, bi, rx_desc, &rx_packets,
 369                                           &rx_bytes, size, xdp_res);
 370                 total_rx_packets += rx_packets;
 371                 total_rx_bytes += rx_bytes;
 372                 xdp_xmit |= xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR);
 373                 next_to_clean = (next_to_clean + 1) & count_mask;
 374         }
 375
 376         rx_ring->next_to_clean = next_to_clean;
 377         cleaned_count = (next_to_clean - rx_ring->next_to_use - 1) & count_mask;
 378
 379         if (cleaned_count >= I40E_RX_BUFFER_WRITE)
 380                 failure = !i40e_alloc_rx_buffers_zc(rx_ring, cleaned_count);
 381
 382         i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
 383         i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
 384
 385         if (xsk_uses_need_wakeup(rx_ring->xsk_pool)) {
 386                 if (failure || next_to_clean == rx_ring->next_to_use)
 387                         xsk_set_rx_need_wakeup(rx_ring->xsk_pool);
 388                 else
 389                         xsk_clear_rx_need_wakeup(rx_ring->xsk_pool);
 390
 391                 return (int)total_rx_packets;
 392         }
 393         return failure ? budget : (int)total_rx_packets;
 394 }
 395
 396 static void i40e_xmit_pkt(struct i40e_ring *xdp_ring, struct xdp_desc *desc,
 397                           unsigned int *total_bytes)
 398 {
 399         struct i40e_tx_desc *tx_desc;
 400         dma_addr_t dma;
 401
 402         dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc->addr);
 403         xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, desc->len);
 404
 405         tx_desc = I40E_TX_DESC(xdp_ring, xdp_ring->next_to_use++);
 406         tx_desc->buffer_addr = cpu_to_le64(dma);
 407         tx_desc->cmd_type_offset_bsz = build_ctob(I40E_TX_DESC_CMD_ICRC | I40E_TX_DESC_CMD_EOP,
 408                                                   0, desc->len, 0);
 409
 410         *total_bytes += desc->len;
 411 }
 412
 413 static void i40e_xmit_pkt_batch(struct i40e_ring *xdp_ring, struct xdp_desc *desc,
 414                                 unsigned int *total_bytes)
 415 {
 416         u16 ntu = xdp_ring->next_to_use;
 417         struct i40e_tx_desc *tx_desc;
 418         dma_addr_t dma;
 419         u32 i;
 420
 421         loop_unrolled_for(i = 0; i < PKTS_PER_BATCH; i++) {
 422                 dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc[i].addr);
 423                 xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, desc[i].len);
 424
 425                 tx_desc = I40E_TX_DESC(xdp_ring, ntu++);
 426                 tx_desc->buffer_addr = cpu_to_le64(dma);
 427                 tx_desc->cmd_type_offset_bsz = build_ctob(I40E_TX_DESC_CMD_ICRC |
 428                                                           I40E_TX_DESC_CMD_EOP,
 429                                                           0, desc[i].len, 0);
 430
 431                 *total_bytes += desc[i].len;
 432         }
 433
 434         xdp_ring->next_to_use = ntu;
 435 }
 436
 437 static void i40e_fill_tx_hw_ring(struct i40e_ring *xdp_ring, struct xdp_desc *descs, u32 nb_pkts,
 438                                  unsigned int *total_bytes)
 439 {
 440         u32 batched, leftover, i;
 441
 442         batched = nb_pkts & ~(PKTS_PER_BATCH - 1);
 443         leftover = nb_pkts & (PKTS_PER_BATCH - 1);
 444         for (i = 0; i < batched; i += PKTS_PER_BATCH)
 445                 i40e_xmit_pkt_batch(xdp_ring, &descs[i], total_bytes);
 446         for (i = batched; i < batched + leftover; i++)
 447                 i40e_xmit_pkt(xdp_ring, &descs[i], total_bytes);
 448 }
 449
 450 static void i40e_set_rs_bit(struct i40e_ring *xdp_ring)
 451 {
 452         u16 ntu = xdp_ring->next_to_use ? xdp_ring->next_to_use - 1 : xdp_ring->count - 1;
 453         struct i40e_tx_desc *tx_desc;
 454
 455         tx_desc = I40E_TX_DESC(xdp_ring, ntu);
 456         tx_desc->cmd_type_offset_bsz |= cpu_to_le64(I40E_TX_DESC_CMD_RS << I40E_TXD_QW1_CMD_SHIFT);
 457 }
 458
 459 /**
 460  * i40e_xmit_zc - Performs zero-copy Tx AF_XDP
 461  * @xdp_ring: XDP Tx ring
 462  * @budget: NAPI budget
 463  *
 464  * Returns true if the work is finished.
 465  **/
 466 static bool i40e_xmit_zc(struct i40e_ring *xdp_ring, unsigned int budget)
 467 {
 468         struct xdp_desc *descs = xdp_ring->xsk_descs;
 469         u32 nb_pkts, nb_processed = 0;
 470         unsigned int total_bytes = 0;
 471
 472         nb_pkts = xsk_tx_peek_release_desc_batch(xdp_ring->xsk_pool, descs, budget);
 473         if (!nb_pkts)
 474                 return false;
 475
 476         if (xdp_ring->next_to_use + nb_pkts >= xdp_ring->count) {
 477                 nb_processed = xdp_ring->count - xdp_ring->next_to_use;
 478                 i40e_fill_tx_hw_ring(xdp_ring, descs, nb_processed, &total_bytes);
 479                 xdp_ring->next_to_use = 0;
 480         }
 481
 482         i40e_fill_tx_hw_ring(xdp_ring, &descs[nb_processed], nb_pkts - nb_processed,
 483                              &total_bytes);
 484
 485         /* Request an interrupt for the last frame and bump tail ptr. */
 486         i40e_set_rs_bit(xdp_ring);
 487         i40e_xdp_ring_update_tail(xdp_ring);
 488
 489         i40e_update_tx_stats(xdp_ring, nb_pkts, total_bytes);
 490
 491         return true;
 492 }
 493
 494 /**
 495  * i40e_clean_xdp_tx_buffer - Frees and unmaps an XDP Tx entry
 496  * @tx_ring: XDP Tx ring
 497  * @tx_bi: Tx buffer info to clean
 498  **/
 499 static void i40e_clean_xdp_tx_buffer(struct i40e_ring *tx_ring,
 500                                      struct i40e_tx_buffer *tx_bi)
 501 {
 502         xdp_return_frame(tx_bi->xdpf);
 503         tx_ring->xdp_tx_active--;
 504         dma_unmap_single(tx_ring->dev,
 505                          dma_unmap_addr(tx_bi, dma),
 506                          dma_unmap_len(tx_bi, len), DMA_TO_DEVICE);
 507         dma_unmap_len_set(tx_bi, len, 0);
 508 }
 509
 510 /**
 511  * i40e_clean_xdp_tx_irq - Completes AF_XDP entries, and cleans XDP entries
 512  * @vsi: Current VSI
 513  * @tx_ring: XDP Tx ring
 514  *
 515  * Returns true if cleanup/tranmission is done.
 516  **/
 517 bool i40e_clean_xdp_tx_irq(struct i40e_vsi *vsi, struct i40e_ring *tx_ring)
 518 {
 519         struct xsk_buff_pool *bp = tx_ring->xsk_pool;
 520         u32 i, completed_frames, xsk_frames = 0;
 521         u32 head_idx = i40e_get_head(tx_ring);
 522         struct i40e_tx_buffer *tx_bi;
 523         unsigned int ntc;
 524
 525         if (head_idx < tx_ring->next_to_clean)
 526                 head_idx += tx_ring->count;
 527         completed_frames = head_idx - tx_ring->next_to_clean;
 528
 529         if (completed_frames == 0)
 530                 goto out_xmit;
 531
 532         if (likely(!tx_ring->xdp_tx_active)) {
 533                 xsk_frames = completed_frames;
 534                 goto skip;
 535         }
 536
 537         ntc = tx_ring->next_to_clean;
 538
 539         for (i = 0; i < completed_frames; i++) {
 540                 tx_bi = &tx_ring->tx_bi[ntc];
 541
 542                 if (tx_bi->xdpf) {
 543                         i40e_clean_xdp_tx_buffer(tx_ring, tx_bi);
 544                         tx_bi->xdpf = NULL;
 545                 } else {
 546                         xsk_frames++;
 547                 }
 548
 549                 if (++ntc >= tx_ring->count)
 550                         ntc = 0;
 551         }
 552
 553 skip:
 554         tx_ring->next_to_clean += completed_frames;
 555         if (unlikely(tx_ring->next_to_clean >= tx_ring->count))
 556                 tx_ring->next_to_clean -= tx_ring->count;
 557
 558         if (xsk_frames)
 559                 xsk_tx_completed(bp, xsk_frames);
 560
 561         i40e_arm_wb(tx_ring, vsi, completed_frames);
 562
 563 out_xmit:
 564         if (xsk_uses_need_wakeup(tx_ring->xsk_pool))
 565                 xsk_set_tx_need_wakeup(tx_ring->xsk_pool);
 566
 567         return i40e_xmit_zc(tx_ring, I40E_DESC_UNUSED(tx_ring));
 568 }
 569
 570 /**
 571  * i40e_xsk_wakeup - Implements the ndo_xsk_wakeup
 572  * @dev: the netdevice
 573  * @queue_id: queue id to wake up
 574  * @flags: ignored in our case since we have Rx and Tx in the same NAPI.
 575  *
 576  * Returns <0 for errors, 0 otherwise.
 577  **/
 578 int i40e_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
 579 {
 580         struct i40e_netdev_priv *np = netdev_priv(dev);
 581         struct i40e_vsi *vsi = np->vsi;
 582         struct i40e_pf *pf = vsi->back;
 583         struct i40e_ring *ring;
 584
 585         if (test_bit(__I40E_CONFIG_BUSY, pf->state))
 586                 return -EAGAIN;
 587
 588         if (test_bit(__I40E_VSI_DOWN, vsi->state))
 589                 return -ENETDOWN;
 590
 591         if (!i40e_enabled_xdp_vsi(vsi))
 592                 return -ENXIO;
 593
 594         if (queue_id >= vsi->num_queue_pairs)
 595                 return -ENXIO;
 596
 597         if (!vsi->xdp_rings[queue_id]->xsk_pool)
 598                 return -ENXIO;
 599
 600         ring = vsi->xdp_rings[queue_id];
 601
 602         /* The idea here is that if NAPI is running, mark a miss, so
 603          * it will run again. If not, trigger an interrupt and
 604          * schedule the NAPI from interrupt context. If NAPI would be
 605          * scheduled here, the interrupt affinity would not be
 606          * honored.
 607          */
 608         if (!napi_if_scheduled_mark_missed(&ring->q_vector->napi))
 609                 i40e_force_wb(vsi, ring->q_vector);
 610
 611         return 0;
 612 }
 613
 614 void i40e_xsk_clean_rx_ring(struct i40e_ring *rx_ring)
 615 {
 616         u16 count_mask = rx_ring->count - 1;
 617         u16 ntc = rx_ring->next_to_clean;
 618         u16 ntu = rx_ring->next_to_use;
 619
 620         for ( ; ntc != ntu; ntc = (ntc + 1)  & count_mask) {
 621                 struct xdp_buff *rx_bi = *i40e_rx_bi(rx_ring, ntc);
 622
 623                 xsk_buff_free(rx_bi);
 624         }
 625 }
 626
 627 /**
 628  * i40e_xsk_clean_xdp_ring - Clean the XDP Tx ring on shutdown
 629  * @tx_ring: XDP Tx ring
 630  **/
 631 void i40e_xsk_clean_tx_ring(struct i40e_ring *tx_ring)
 632 {
 633         u16 ntc = tx_ring->next_to_clean, ntu = tx_ring->next_to_use;
 634         struct xsk_buff_pool *bp = tx_ring->xsk_pool;
 635         struct i40e_tx_buffer *tx_bi;
 636         u32 xsk_frames = 0;
 637
 638         while (ntc != ntu) {
 639                 tx_bi = &tx_ring->tx_bi[ntc];
 640
 641                 if (tx_bi->xdpf)
 642                         i40e_clean_xdp_tx_buffer(tx_ring, tx_bi);
 643                 else
 644                         xsk_frames++;
 645
 646                 tx_bi->xdpf = NULL;
 647
 648                 ntc++;
 649                 if (ntc >= tx_ring->count)
 650                         ntc = 0;
 651         }
 652
 653         if (xsk_frames)
 654                 xsk_tx_completed(bp, xsk_frames);
 655 }
 656
 657 /**
 658  * i40e_xsk_any_rx_ring_enabled - Checks if Rx rings have an AF_XDP
 659  * buffer pool attached
 660  * @vsi: vsi
 661  *
 662  * Returns true if any of the Rx rings has an AF_XDP buffer pool attached
 663  **/
 664 bool i40e_xsk_any_rx_ring_enabled(struct i40e_vsi *vsi)
 665 {
 666         struct net_device *netdev = vsi->netdev;
 667         int i;
 668
 669         for (i = 0; i < vsi->num_queue_pairs; i++) {
 670                 if (xsk_get_pool_from_qid(netdev, i))
 671                         return true;
 672         }
 673
 674         return false;
 675 }