drivers/net/ethernet/intel/ice/ice_xsk.c

   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                 synchronize_rcu();
  41                 ice_clean_tx_ring(vsi->xdp_rings[q_idx]);
  42         }
  43         ice_clean_rx_ring(vsi->rx_rings[q_idx]);
  44 }
  45
  46 /**
  47  * ice_qvec_toggle_napi - Enables/disables NAPI for a given q_vector
  48  * @vsi: VSI that has netdev
  49  * @q_vector: q_vector that has NAPI context
  50  * @enable: true for enable, false for disable
  51  */
  52 static void
  53 ice_qvec_toggle_napi(struct ice_vsi *vsi, struct ice_q_vector *q_vector,
  54                      bool enable)
  55 {
  56         if (!vsi->netdev || !q_vector)
  57                 return;
  58
  59         if (enable)
  60                 napi_enable(&q_vector->napi);
  61         else
  62                 napi_disable(&q_vector->napi);
  63 }
  64
  65 /**
  66  * ice_qvec_dis_irq - Mask off queue interrupt generation on given ring
  67  * @vsi: the VSI that contains queue vector being un-configured
  68  * @rx_ring: Rx ring that will have its IRQ disabled
  69  * @q_vector: queue vector
  70  */
  71 static void
  72 ice_qvec_dis_irq(struct ice_vsi *vsi, struct ice_ring *rx_ring,
  73                  struct ice_q_vector *q_vector)
  74 {
  75         struct ice_pf *pf = vsi->back;
  76         struct ice_hw *hw = &pf->hw;
  77         int base = vsi->base_vector;
  78         u16 reg;
  79         u32 val;
  80
  81         /* QINT_TQCTL is being cleared in ice_vsi_stop_tx_ring, so handle
  82          * here only QINT_RQCTL
  83          */
  84         reg = rx_ring->reg_idx;
  85         val = rd32(hw, QINT_RQCTL(reg));
  86         val &= ~QINT_RQCTL_CAUSE_ENA_M;
  87         wr32(hw, QINT_RQCTL(reg), val);
  88
  89         if (q_vector) {
  90                 u16 v_idx = q_vector->v_idx;
  91
  92                 wr32(hw, GLINT_DYN_CTL(q_vector->reg_idx), 0);
  93                 ice_flush(hw);
  94                 synchronize_irq(pf->msix_entries[v_idx + base].vector);
  95         }
  96 }
  97
  98 /**
  99  * ice_qvec_cfg_msix - Enable IRQ for given queue vector
 100  * @vsi: the VSI that contains queue vector
 101  * @q_vector: queue vector
 102  */
 103 static void
 104 ice_qvec_cfg_msix(struct ice_vsi *vsi, struct ice_q_vector *q_vector)
 105 {
 106         u16 reg_idx = q_vector->reg_idx;
 107         struct ice_pf *pf = vsi->back;
 108         struct ice_hw *hw = &pf->hw;
 109         struct ice_ring *ring;
 110
 111         ice_cfg_itr(hw, q_vector);
 112
 113         wr32(hw, GLINT_RATE(reg_idx),
 114              ice_intrl_usec_to_reg(q_vector->intrl, hw->intrl_gran));
 115
 116         ice_for_each_ring(ring, q_vector->tx)
 117                 ice_cfg_txq_interrupt(vsi, ring->reg_idx, reg_idx,
 118                                       q_vector->tx.itr_idx);
 119
 120         ice_for_each_ring(ring, q_vector->rx)
 121                 ice_cfg_rxq_interrupt(vsi, ring->reg_idx, reg_idx,
 122                                       q_vector->rx.itr_idx);
 123
 124         ice_flush(hw);
 125 }
 126
 127 /**
 128  * ice_qvec_ena_irq - Enable IRQ for given queue vector
 129  * @vsi: the VSI that contains queue vector
 130  * @q_vector: queue vector
 131  */
 132 static void ice_qvec_ena_irq(struct ice_vsi *vsi, struct ice_q_vector *q_vector)
 133 {
 134         struct ice_pf *pf = vsi->back;
 135         struct ice_hw *hw = &pf->hw;
 136
 137         ice_irq_dynamic_ena(hw, vsi, q_vector);
 138
 139         ice_flush(hw);
 140 }
 141
 142 /**
 143  * ice_qp_dis - Disables a queue pair
 144  * @vsi: VSI of interest
 145  * @q_idx: ring index in array
 146  *
 147  * Returns 0 on success, negative on failure.
 148  */
 149 static int ice_qp_dis(struct ice_vsi *vsi, u16 q_idx)
 150 {
 151         struct ice_txq_meta txq_meta = { };
 152         struct ice_ring *tx_ring, *rx_ring;
 153         struct ice_q_vector *q_vector;
 154         int timeout = 50;
 155         int err;
 156
 157         if (q_idx >= vsi->num_rxq || q_idx >= vsi->num_txq)
 158                 return -EINVAL;
 159
 160         tx_ring = vsi->tx_rings[q_idx];
 161         rx_ring = vsi->rx_rings[q_idx];
 162         q_vector = rx_ring->q_vector;
 163
 164         while (test_and_set_bit(__ICE_CFG_BUSY, vsi->state)) {
 165                 timeout--;
 166                 if (!timeout)
 167                         return -EBUSY;
 168                 usleep_range(1000, 2000);
 169         }
 170         netif_tx_stop_queue(netdev_get_tx_queue(vsi->netdev, q_idx));
 171
 172         ice_fill_txq_meta(vsi, tx_ring, &txq_meta);
 173         err = ice_vsi_stop_tx_ring(vsi, ICE_NO_RESET, 0, tx_ring, &txq_meta);
 174         if (err)
 175                 return err;
 176         if (ice_is_xdp_ena_vsi(vsi)) {
 177                 struct ice_ring *xdp_ring = vsi->xdp_rings[q_idx];
 178
 179                 memset(&txq_meta, 0, sizeof(txq_meta));
 180                 ice_fill_txq_meta(vsi, xdp_ring, &txq_meta);
 181                 err = ice_vsi_stop_tx_ring(vsi, ICE_NO_RESET, 0, xdp_ring,
 182                                            &txq_meta);
 183                 if (err)
 184                         return err;
 185         }
 186         ice_qvec_dis_irq(vsi, rx_ring, q_vector);
 187
 188         err = ice_vsi_ctrl_one_rx_ring(vsi, false, q_idx, true);
 189         if (err)
 190                 return err;
 191
 192         ice_qvec_toggle_napi(vsi, q_vector, false);
 193         ice_qp_clean_rings(vsi, q_idx);
 194         ice_qp_reset_stats(vsi, q_idx);
 195
 196         return 0;
 197 }
 198
 199 /**
 200  * ice_qp_ena - Enables a queue pair
 201  * @vsi: VSI of interest
 202  * @q_idx: ring index in array
 203  *
 204  * Returns 0 on success, negative on failure.
 205  */
 206 static int ice_qp_ena(struct ice_vsi *vsi, u16 q_idx)
 207 {
 208         struct ice_aqc_add_tx_qgrp *qg_buf;
 209         struct ice_ring *tx_ring, *rx_ring;
 210         struct ice_q_vector *q_vector;
 211         u16 size;
 212         int err;
 213
 214         if (q_idx >= vsi->num_rxq || q_idx >= vsi->num_txq)
 215                 return -EINVAL;
 216
 217         size = struct_size(qg_buf, txqs, 1);
 218         qg_buf = kzalloc(size, GFP_KERNEL);
 219         if (!qg_buf)
 220                 return -ENOMEM;
 221
 222         qg_buf->num_txqs = 1;
 223
 224         tx_ring = vsi->tx_rings[q_idx];
 225         rx_ring = vsi->rx_rings[q_idx];
 226         q_vector = rx_ring->q_vector;
 227
 228         err = ice_vsi_cfg_txq(vsi, tx_ring, qg_buf);
 229         if (err)
 230                 goto free_buf;
 231
 232         if (ice_is_xdp_ena_vsi(vsi)) {
 233                 struct ice_ring *xdp_ring = vsi->xdp_rings[q_idx];
 234
 235                 memset(qg_buf, 0, size);
 236                 qg_buf->num_txqs = 1;
 237                 err = ice_vsi_cfg_txq(vsi, xdp_ring, qg_buf);
 238                 if (err)
 239                         goto free_buf;
 240                 ice_set_ring_xdp(xdp_ring);
 241                 xdp_ring->xsk_pool = ice_xsk_pool(xdp_ring);
 242         }
 243
 244         err = ice_setup_rx_ctx(rx_ring);
 245         if (err)
 246                 goto free_buf;
 247
 248         ice_qvec_cfg_msix(vsi, q_vector);
 249
 250         err = ice_vsi_ctrl_one_rx_ring(vsi, true, q_idx, true);
 251         if (err)
 252                 goto free_buf;
 253
 254         clear_bit(__ICE_CFG_BUSY, vsi->state);
 255         ice_qvec_toggle_napi(vsi, q_vector, true);
 256         ice_qvec_ena_irq(vsi, q_vector);
 257
 258         netif_tx_start_queue(netdev_get_tx_queue(vsi->netdev, q_idx));
 259 free_buf:
 260         kfree(qg_buf);
 261         return err;
 262 }
 263
 264 /**
 265  * ice_xsk_alloc_pools - allocate a buffer pool for an XDP socket
 266  * @vsi: VSI to allocate the buffer pool on
 267  *
 268  * Returns 0 on success, negative on error
 269  */
 270 static int ice_xsk_alloc_pools(struct ice_vsi *vsi)
 271 {
 272         if (vsi->xsk_pools)
 273                 return 0;
 274
 275         vsi->xsk_pools = kcalloc(vsi->num_xsk_pools, sizeof(*vsi->xsk_pools),
 276                                  GFP_KERNEL);
 277
 278         if (!vsi->xsk_pools) {
 279                 vsi->num_xsk_pools = 0;
 280                 return -ENOMEM;
 281         }
 282
 283         return 0;
 284 }
 285
 286 /**
 287  * ice_xsk_remove_pool - Remove an buffer pool for a certain ring/qid
 288  * @vsi: VSI from which the VSI will be removed
 289  * @qid: Ring/qid associated with the buffer pool
 290  */
 291 static void ice_xsk_remove_pool(struct ice_vsi *vsi, u16 qid)
 292 {
 293         vsi->xsk_pools[qid] = NULL;
 294         vsi->num_xsk_pools_used--;
 295
 296         if (vsi->num_xsk_pools_used == 0) {
 297                 kfree(vsi->xsk_pools);
 298                 vsi->xsk_pools = NULL;
 299                 vsi->num_xsk_pools = 0;
 300         }
 301 }
 302
 303 /**
 304  * ice_xsk_pool_disable - disable a buffer pool region
 305  * @vsi: Current VSI
 306  * @qid: queue ID
 307  *
 308  * Returns 0 on success, negative on failure
 309  */
 310 static int ice_xsk_pool_disable(struct ice_vsi *vsi, u16 qid)
 311 {
 312         if (!vsi->xsk_pools || qid >= vsi->num_xsk_pools ||
 313             !vsi->xsk_pools[qid])
 314                 return -EINVAL;
 315
 316         xsk_pool_dma_unmap(vsi->xsk_pools[qid], ICE_RX_DMA_ATTR);
 317         ice_xsk_remove_pool(vsi, qid);
 318
 319         return 0;
 320 }
 321
 322 /**
 323  * ice_xsk_pool_enable - enable a buffer pool region
 324  * @vsi: Current VSI
 325  * @pool: pointer to a requested buffer pool region
 326  * @qid: queue ID
 327  *
 328  * Returns 0 on success, negative on failure
 329  */
 330 static int
 331 ice_xsk_pool_enable(struct ice_vsi *vsi, struct xsk_buff_pool *pool, u16 qid)
 332 {
 333         int err;
 334
 335         if (vsi->type != ICE_VSI_PF)
 336                 return -EINVAL;
 337
 338         if (!vsi->num_xsk_pools)
 339                 vsi->num_xsk_pools = min_t(u16, vsi->num_rxq, vsi->num_txq);
 340         if (qid >= vsi->num_xsk_pools)
 341                 return -EINVAL;
 342
 343         err = ice_xsk_alloc_pools(vsi);
 344         if (err)
 345                 return err;
 346
 347         if (vsi->xsk_pools && vsi->xsk_pools[qid])
 348                 return -EBUSY;
 349
 350         vsi->xsk_pools[qid] = pool;
 351         vsi->num_xsk_pools_used++;
 352
 353         err = xsk_pool_dma_map(vsi->xsk_pools[qid], ice_pf_to_dev(vsi->back),
 354                                ICE_RX_DMA_ATTR);
 355         if (err)
 356                 return err;
 357
 358         return 0;
 359 }
 360
 361 /**
 362  * ice_xsk_pool_setup - enable/disable a buffer pool region depending on its state
 363  * @vsi: Current VSI
 364  * @pool: buffer pool to enable/associate to a ring, NULL to disable
 365  * @qid: queue ID
 366  *
 367  * Returns 0 on success, negative on failure
 368  */
 369 int ice_xsk_pool_setup(struct ice_vsi *vsi, struct xsk_buff_pool *pool, u16 qid)
 370 {
 371         bool if_running, pool_present = !!pool;
 372         int ret = 0, pool_failure = 0;
 373
 374         if (qid >= vsi->num_rxq || qid >= vsi->num_txq) {
 375                 netdev_err(vsi->netdev, "Please use queue id in scope of combined queues count\n");
 376                 pool_failure = -EINVAL;
 377                 goto failure;
 378         }
 379
 380         if (!is_power_of_2(vsi->rx_rings[qid]->count) ||
 381             !is_power_of_2(vsi->tx_rings[qid]->count)) {
 382                 netdev_err(vsi->netdev, "Please align ring sizes to power of 2\n");
 383                 pool_failure = -EINVAL;
 384                 goto failure;
 385         }
 386
 387         if_running = netif_running(vsi->netdev) && ice_is_xdp_ena_vsi(vsi);
 388
 389         if (if_running) {
 390                 ret = ice_qp_dis(vsi, qid);
 391                 if (ret) {
 392                         netdev_err(vsi->netdev, "ice_qp_dis error = %d\n", ret);
 393                         goto xsk_pool_if_up;
 394                 }
 395         }
 396
 397         pool_failure = pool_present ? ice_xsk_pool_enable(vsi, pool, qid) :
 398                                       ice_xsk_pool_disable(vsi, qid);
 399
 400 xsk_pool_if_up:
 401         if (if_running) {
 402                 ret = ice_qp_ena(vsi, qid);
 403                 if (!ret && pool_present)
 404                         napi_schedule(&vsi->xdp_rings[qid]->q_vector->napi);
 405                 else if (ret)
 406                         netdev_err(vsi->netdev, "ice_qp_ena error = %d\n", ret);
 407         }
 408
 409 failure:
 410         if (pool_failure) {
 411                 netdev_err(vsi->netdev, "Could not %sable buffer pool, error = %d\n",
 412                            pool_present ? "en" : "dis", pool_failure);
 413                 return pool_failure;
 414         }
 415
 416         return ret;
 417 }
 418
 419 /**
 420  * ice_alloc_rx_bufs_zc - allocate a number of Rx buffers
 421  * @rx_ring: Rx ring
 422  * @count: The number of buffers to allocate
 423  *
 424  * This function allocates a number of Rx buffers from the fill ring
 425  * or the internal recycle mechanism and places them on the Rx ring.
 426  *
 427  * Returns false if all allocations were successful, true if any fail.
 428  */
 429 bool ice_alloc_rx_bufs_zc(struct ice_ring *rx_ring, u16 count)
 430 {
 431         union ice_32b_rx_flex_desc *rx_desc;
 432         u16 ntu = rx_ring->next_to_use;
 433         struct ice_rx_buf *rx_buf;
 434         bool ret = false;
 435         dma_addr_t dma;
 436
 437         if (!count)
 438                 return false;
 439
 440         rx_desc = ICE_RX_DESC(rx_ring, ntu);
 441         rx_buf = &rx_ring->rx_buf[ntu];
 442
 443         do {
 444                 rx_buf->xdp = xsk_buff_alloc(rx_ring->xsk_pool);
 445                 if (!rx_buf->xdp) {
 446                         ret = true;
 447                         break;
 448                 }
 449
 450                 dma = xsk_buff_xdp_get_dma(rx_buf->xdp);
 451                 rx_desc->read.pkt_addr = cpu_to_le64(dma);
 452                 rx_desc->wb.status_error0 = 0;
 453
 454                 rx_desc++;
 455                 rx_buf++;
 456                 ntu++;
 457
 458                 if (unlikely(ntu == rx_ring->count)) {
 459                         rx_desc = ICE_RX_DESC(rx_ring, 0);
 460                         rx_buf = rx_ring->rx_buf;
 461                         ntu = 0;
 462                 }
 463         } while (--count);
 464
 465         if (rx_ring->next_to_use != ntu) {
 466                 /* clear the status bits for the next_to_use descriptor */
 467                 rx_desc->wb.status_error0 = 0;
 468                 ice_release_rx_desc(rx_ring, ntu);
 469         }
 470
 471         return ret;
 472 }
 473
 474 /**
 475  * ice_bump_ntc - Bump the next_to_clean counter of an Rx ring
 476  * @rx_ring: Rx ring
 477  */
 478 static void ice_bump_ntc(struct ice_ring *rx_ring)
 479 {
 480         int ntc = rx_ring->next_to_clean + 1;
 481
 482         ntc = (ntc < rx_ring->count) ? ntc : 0;
 483         rx_ring->next_to_clean = ntc;
 484         prefetch(ICE_RX_DESC(rx_ring, ntc));
 485 }
 486
 487 /**
 488  * ice_construct_skb_zc - Create an sk_buff from zero-copy buffer
 489  * @rx_ring: Rx ring
 490  * @rx_buf: zero-copy Rx buffer
 491  *
 492  * This function allocates a new skb from a zero-copy Rx buffer.
 493  *
 494  * Returns the skb on success, NULL on failure.
 495  */
 496 static struct sk_buff *
 497 ice_construct_skb_zc(struct ice_ring *rx_ring, struct ice_rx_buf *rx_buf)
 498 {
 499         unsigned int metasize = rx_buf->xdp->data - rx_buf->xdp->data_meta;
 500         unsigned int datasize = rx_buf->xdp->data_end - rx_buf->xdp->data;
 501         unsigned int datasize_hard = rx_buf->xdp->data_end -
 502                                      rx_buf->xdp->data_hard_start;
 503         struct sk_buff *skb;
 504
 505         skb = __napi_alloc_skb(&rx_ring->q_vector->napi, datasize_hard,
 506                                GFP_ATOMIC | __GFP_NOWARN);
 507         if (unlikely(!skb))
 508                 return NULL;
 509
 510         skb_reserve(skb, rx_buf->xdp->data - rx_buf->xdp->data_hard_start);
 511         memcpy(__skb_put(skb, datasize), rx_buf->xdp->data, datasize);
 512         if (metasize)
 513                 skb_metadata_set(skb, metasize);
 514
 515         xsk_buff_free(rx_buf->xdp);
 516         rx_buf->xdp = NULL;
 517         return skb;
 518 }
 519
 520 /**
 521  * ice_run_xdp_zc - Executes an XDP program in zero-copy path
 522  * @rx_ring: Rx ring
 523  * @xdp: xdp_buff used as input to the XDP program
 524  *
 525  * Returns any of ICE_XDP_{PASS, CONSUMED, TX, REDIR}
 526  */
 527 static int
 528 ice_run_xdp_zc(struct ice_ring *rx_ring, struct xdp_buff *xdp)
 529 {
 530         int err, result = ICE_XDP_PASS;
 531         struct bpf_prog *xdp_prog;
 532         struct ice_ring *xdp_ring;
 533         u32 act;
 534
 535         rcu_read_lock();
 536         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
 537         if (!xdp_prog) {
 538                 rcu_read_unlock();
 539                 return ICE_XDP_PASS;
 540         }
 541
 542         act = bpf_prog_run_xdp(xdp_prog, xdp);
 543
 544         if (likely(act == XDP_REDIRECT)) {
 545                 err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
 546                 if (err)
 547                         goto out_failure;
 548                 rcu_read_unlock();
 549                 return ICE_XDP_REDIR;
 550         }
 551
 552         switch (act) {
 553         case XDP_PASS:
 554                 break;
 555         case XDP_TX:
 556                 xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->q_index];
 557                 result = ice_xmit_xdp_buff(xdp, xdp_ring);
 558                 if (result == ICE_XDP_CONSUMED)
 559                         goto out_failure;
 560                 break;
 561         default:
 562                 bpf_warn_invalid_xdp_action(act);
 563                 fallthrough;
 564         case XDP_ABORTED:
 565 out_failure:
 566                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
 567                 fallthrough;
 568         case XDP_DROP:
 569                 result = ICE_XDP_CONSUMED;
 570                 break;
 571         }
 572
 573         rcu_read_unlock();
 574         return result;
 575 }
 576
 577 /**
 578  * ice_clean_rx_irq_zc - consumes packets from the hardware ring
 579  * @rx_ring: AF_XDP Rx ring
 580  * @budget: NAPI budget
 581  *
 582  * Returns number of processed packets on success, remaining budget on failure.
 583  */
 584 int ice_clean_rx_irq_zc(struct ice_ring *rx_ring, int budget)
 585 {
 586         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 587         u16 cleaned_count = ICE_DESC_UNUSED(rx_ring);
 588         unsigned int xdp_xmit = 0;
 589         bool failure = false;
 590
 591         while (likely(total_rx_packets < (unsigned int)budget)) {
 592                 union ice_32b_rx_flex_desc *rx_desc;
 593                 unsigned int size, xdp_res = 0;
 594                 struct ice_rx_buf *rx_buf;
 595                 struct sk_buff *skb;
 596                 u16 stat_err_bits;
 597                 u16 vlan_tag = 0;
 598                 u8 rx_ptype;
 599
 600                 if (cleaned_count >= ICE_RX_BUF_WRITE) {
 601                         failure |= ice_alloc_rx_bufs_zc(rx_ring,
 602                                                         cleaned_count);
 603                         cleaned_count = 0;
 604                 }
 605
 606                 rx_desc = ICE_RX_DESC(rx_ring, rx_ring->next_to_clean);
 607
 608                 stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S);
 609                 if (!ice_test_staterr(rx_desc, stat_err_bits))
 610                         break;
 611
 612                 /* This memory barrier is needed to keep us from reading
 613                  * any other fields out of the rx_desc until we have
 614                  * verified the descriptor has been written back.
 615                  */
 616                 dma_rmb();
 617
 618                 size = le16_to_cpu(rx_desc->wb.pkt_len) &
 619                                    ICE_RX_FLX_DESC_PKT_LEN_M;
 620                 if (!size)
 621                         break;
 622
 623                 rx_buf = &rx_ring->rx_buf[rx_ring->next_to_clean];
 624                 rx_buf->xdp->data_end = rx_buf->xdp->data + size;
 625                 xsk_buff_dma_sync_for_cpu(rx_buf->xdp, rx_ring->xsk_pool);
 626
 627                 xdp_res = ice_run_xdp_zc(rx_ring, rx_buf->xdp);
 628                 if (xdp_res) {
 629                         if (xdp_res & (ICE_XDP_TX | ICE_XDP_REDIR))
 630                                 xdp_xmit |= xdp_res;
 631                         else
 632                                 xsk_buff_free(rx_buf->xdp);
 633
 634                         rx_buf->xdp = NULL;
 635                         total_rx_bytes += size;
 636                         total_rx_packets++;
 637                         cleaned_count++;
 638
 639                         ice_bump_ntc(rx_ring);
 640                         continue;
 641                 }
 642
 643                 /* XDP_PASS path */
 644                 skb = ice_construct_skb_zc(rx_ring, rx_buf);
 645                 if (!skb) {
 646                         rx_ring->rx_stats.alloc_buf_failed++;
 647                         break;
 648                 }
 649
 650                 cleaned_count++;
 651                 ice_bump_ntc(rx_ring);
 652
 653                 if (eth_skb_pad(skb)) {
 654                         skb = NULL;
 655                         continue;
 656                 }
 657
 658                 total_rx_bytes += skb->len;
 659                 total_rx_packets++;
 660
 661                 stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_L2TAG1P_S);
 662                 if (ice_test_staterr(rx_desc, stat_err_bits))
 663                         vlan_tag = le16_to_cpu(rx_desc->wb.l2tag1);
 664
 665                 rx_ptype = le16_to_cpu(rx_desc->wb.ptype_flex_flags0) &
 666                                        ICE_RX_FLEX_DESC_PTYPE_M;
 667
 668                 ice_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
 669                 ice_receive_skb(rx_ring, skb, vlan_tag);
 670         }
 671
 672         ice_finalize_xdp_rx(rx_ring, xdp_xmit);
 673         ice_update_rx_ring_stats(rx_ring, total_rx_packets, total_rx_bytes);
 674
 675         if (xsk_uses_need_wakeup(rx_ring->xsk_pool)) {
 676                 if (failure || rx_ring->next_to_clean == rx_ring->next_to_use)
 677                         xsk_set_rx_need_wakeup(rx_ring->xsk_pool);
 678                 else
 679                         xsk_clear_rx_need_wakeup(rx_ring->xsk_pool);
 680
 681                 return (int)total_rx_packets;
 682         }
 683
 684         return failure ? budget : (int)total_rx_packets;
 685 }
 686
 687 /**
 688  * ice_xmit_zc - Completes AF_XDP entries, and cleans XDP entries
 689  * @xdp_ring: XDP Tx ring
 690  * @budget: max number of frames to xmit
 691  *
 692  * Returns true if cleanup/transmission is done.
 693  */
 694 static bool ice_xmit_zc(struct ice_ring *xdp_ring, int budget)
 695 {
 696         struct ice_tx_desc *tx_desc = NULL;
 697         bool work_done = true;
 698         struct xdp_desc desc;
 699         dma_addr_t dma;
 700
 701         while (likely(budget-- > 0)) {
 702                 struct ice_tx_buf *tx_buf;
 703
 704                 if (unlikely(!ICE_DESC_UNUSED(xdp_ring))) {
 705                         xdp_ring->tx_stats.tx_busy++;
 706                         work_done = false;
 707                         break;
 708                 }
 709
 710                 tx_buf = &xdp_ring->tx_buf[xdp_ring->next_to_use];
 711
 712                 if (!xsk_tx_peek_desc(xdp_ring->xsk_pool, &desc))
 713                         break;
 714
 715                 dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc.addr);
 716                 xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma,
 717                                                  desc.len);
 718
 719                 tx_buf->bytecount = desc.len;
 720
 721                 tx_desc = ICE_TX_DESC(xdp_ring, xdp_ring->next_to_use);
 722                 tx_desc->buf_addr = cpu_to_le64(dma);
 723                 tx_desc->cmd_type_offset_bsz =
 724                         ice_build_ctob(ICE_TXD_LAST_DESC_CMD, 0, desc.len, 0);
 725
 726                 xdp_ring->next_to_use++;
 727                 if (xdp_ring->next_to_use == xdp_ring->count)
 728                         xdp_ring->next_to_use = 0;
 729         }
 730
 731         if (tx_desc) {
 732                 ice_xdp_ring_update_tail(xdp_ring);
 733                 xsk_tx_release(xdp_ring->xsk_pool);
 734         }
 735
 736         return budget > 0 && work_done;
 737 }
 738
 739 /**
 740  * ice_clean_xdp_tx_buf - Free and unmap XDP Tx buffer
 741  * @xdp_ring: XDP Tx ring
 742  * @tx_buf: Tx buffer to clean
 743  */
 744 static void
 745 ice_clean_xdp_tx_buf(struct ice_ring *xdp_ring, struct ice_tx_buf *tx_buf)
 746 {
 747         xdp_return_frame((struct xdp_frame *)tx_buf->raw_buf);
 748         dma_unmap_single(xdp_ring->dev, dma_unmap_addr(tx_buf, dma),
 749                          dma_unmap_len(tx_buf, len), DMA_TO_DEVICE);
 750         dma_unmap_len_set(tx_buf, len, 0);
 751 }
 752
 753 /**
 754  * ice_clean_tx_irq_zc - Completes AF_XDP entries, and cleans XDP entries
 755  * @xdp_ring: XDP Tx ring
 756  * @budget: NAPI budget
 757  *
 758  * Returns true if cleanup/tranmission is done.
 759  */
 760 bool ice_clean_tx_irq_zc(struct ice_ring *xdp_ring, int budget)
 761 {
 762         int total_packets = 0, total_bytes = 0;
 763         s16 ntc = xdp_ring->next_to_clean;
 764         struct ice_tx_desc *tx_desc;
 765         struct ice_tx_buf *tx_buf;
 766         u32 xsk_frames = 0;
 767         bool xmit_done;
 768
 769         tx_desc = ICE_TX_DESC(xdp_ring, ntc);
 770         tx_buf = &xdp_ring->tx_buf[ntc];
 771         ntc -= xdp_ring->count;
 772
 773         do {
 774                 if (!(tx_desc->cmd_type_offset_bsz &
 775                       cpu_to_le64(ICE_TX_DESC_DTYPE_DESC_DONE)))
 776                         break;
 777
 778                 total_bytes += tx_buf->bytecount;
 779                 total_packets++;
 780
 781                 if (tx_buf->raw_buf) {
 782                         ice_clean_xdp_tx_buf(xdp_ring, tx_buf);
 783                         tx_buf->raw_buf = NULL;
 784                 } else {
 785                         xsk_frames++;
 786                 }
 787
 788                 tx_desc->cmd_type_offset_bsz = 0;
 789                 tx_buf++;
 790                 tx_desc++;
 791                 ntc++;
 792
 793                 if (unlikely(!ntc)) {
 794                         ntc -= xdp_ring->count;
 795                         tx_buf = xdp_ring->tx_buf;
 796                         tx_desc = ICE_TX_DESC(xdp_ring, 0);
 797                 }
 798
 799                 prefetch(tx_desc);
 800
 801         } while (likely(--budget));
 802
 803         ntc += xdp_ring->count;
 804         xdp_ring->next_to_clean = ntc;
 805
 806         if (xsk_frames)
 807                 xsk_tx_completed(xdp_ring->xsk_pool, xsk_frames);
 808
 809         if (xsk_uses_need_wakeup(xdp_ring->xsk_pool))
 810                 xsk_set_tx_need_wakeup(xdp_ring->xsk_pool);
 811
 812         ice_update_tx_ring_stats(xdp_ring, total_packets, total_bytes);
 813         xmit_done = ice_xmit_zc(xdp_ring, ICE_DFLT_IRQ_WORK);
 814
 815         return budget > 0 && xmit_done;
 816 }
 817
 818 /**
 819  * ice_xsk_wakeup - Implements ndo_xsk_wakeup
 820  * @netdev: net_device
 821  * @queue_id: queue to wake up
 822  * @flags: ignored in our case, since we have Rx and Tx in the same NAPI
 823  *
 824  * Returns negative on error, zero otherwise.
 825  */
 826 int
 827 ice_xsk_wakeup(struct net_device *netdev, u32 queue_id,
 828                u32 __always_unused flags)
 829 {
 830         struct ice_netdev_priv *np = netdev_priv(netdev);
 831         struct ice_q_vector *q_vector;
 832         struct ice_vsi *vsi = np->vsi;
 833         struct ice_ring *ring;
 834
 835         if (test_bit(__ICE_DOWN, vsi->state))
 836                 return -ENETDOWN;
 837
 838         if (!ice_is_xdp_ena_vsi(vsi))
 839                 return -ENXIO;
 840
 841         if (queue_id >= vsi->num_txq)
 842                 return -ENXIO;
 843
 844         if (!vsi->xdp_rings[queue_id]->xsk_pool)
 845                 return -ENXIO;
 846
 847         ring = vsi->xdp_rings[queue_id];
 848
 849         /* The idea here is that if NAPI is running, mark a miss, so
 850          * it will run again. If not, trigger an interrupt and
 851          * schedule the NAPI from interrupt context. If NAPI would be
 852          * scheduled here, the interrupt affinity would not be
 853          * honored.
 854          */
 855         q_vector = ring->q_vector;
 856         if (!napi_if_scheduled_mark_missed(&q_vector->napi))
 857                 ice_trigger_sw_intr(&vsi->back->hw, q_vector);
 858
 859         return 0;
 860 }
 861
 862 /**
 863  * ice_xsk_any_rx_ring_ena - Checks if Rx rings have AF_XDP buff pool attached
 864  * @vsi: VSI to be checked
 865  *
 866  * Returns true if any of the Rx rings has an AF_XDP buff pool attached
 867  */
 868 bool ice_xsk_any_rx_ring_ena(struct ice_vsi *vsi)
 869 {
 870         int i;
 871
 872         if (!vsi->xsk_pools)
 873                 return false;
 874
 875         for (i = 0; i < vsi->num_xsk_pools; i++) {
 876                 if (vsi->xsk_pools[i])
 877                         return true;
 878         }
 879
 880         return false;
 881 }
 882
 883 /**
 884  * ice_xsk_clean_rx_ring - clean buffer pool queues connected to a given Rx ring
 885  * @rx_ring: ring to be cleaned
 886  */
 887 void ice_xsk_clean_rx_ring(struct ice_ring *rx_ring)
 888 {
 889         u16 i;
 890
 891         for (i = 0; i < rx_ring->count; i++) {
 892                 struct ice_rx_buf *rx_buf = &rx_ring->rx_buf[i];
 893
 894                 if (!rx_buf->xdp)
 895                         continue;
 896
 897                 rx_buf->xdp = NULL;
 898         }
 899 }
 900
 901 /**
 902  * ice_xsk_clean_xdp_ring - Clean the XDP Tx ring and its buffer pool queues
 903  * @xdp_ring: XDP_Tx ring
 904  */
 905 void ice_xsk_clean_xdp_ring(struct ice_ring *xdp_ring)
 906 {
 907         u16 ntc = xdp_ring->next_to_clean, ntu = xdp_ring->next_to_use;
 908         u32 xsk_frames = 0;
 909
 910         while (ntc != ntu) {
 911                 struct ice_tx_buf *tx_buf = &xdp_ring->tx_buf[ntc];
 912
 913                 if (tx_buf->raw_buf)
 914                         ice_clean_xdp_tx_buf(xdp_ring, tx_buf);
 915                 else
 916                         xsk_frames++;
 917
 918                 tx_buf->raw_buf = NULL;
 919
 920                 ntc++;
 921                 if (ntc >= xdp_ring->count)
 922                         ntc = 0;
 923         }
 924
 925         if (xsk_frames)
 926                 xsk_tx_completed(xdp_ring->xsk_pool, xsk_frames);
 927 }