Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /* Copyright (c) 2019, Intel Corporation. */
0003 
0004 #include <linux/bpf_trace.h>
0005 #include <net/xdp_sock_drv.h>
0006 #include <net/xdp.h>
0007 #include "ice.h"
0008 #include "ice_base.h"
0009 #include "ice_type.h"
0010 #include "ice_xsk.h"
0011 #include "ice_txrx.h"
0012 #include "ice_txrx_lib.h"
0013 #include "ice_lib.h"
0014 
0015 static struct xdp_buff **ice_xdp_buf(struct ice_rx_ring *rx_ring, u32 idx)
0016 {
0017     return &rx_ring->xdp_buf[idx];
0018 }
0019 
0020 /**
0021  * ice_qp_reset_stats - Resets all stats for rings of given index
0022  * @vsi: VSI that contains rings of interest
0023  * @q_idx: ring index in array
0024  */
0025 static void ice_qp_reset_stats(struct ice_vsi *vsi, u16 q_idx)
0026 {
0027     memset(&vsi->rx_rings[q_idx]->rx_stats, 0,
0028            sizeof(vsi->rx_rings[q_idx]->rx_stats));
0029     memset(&vsi->tx_rings[q_idx]->stats, 0,
0030            sizeof(vsi->tx_rings[q_idx]->stats));
0031     if (ice_is_xdp_ena_vsi(vsi))
0032         memset(&vsi->xdp_rings[q_idx]->stats, 0,
0033                sizeof(vsi->xdp_rings[q_idx]->stats));
0034 }
0035 
0036 /**
0037  * ice_qp_clean_rings - Cleans all the rings of a given index
0038  * @vsi: VSI that contains rings of interest
0039  * @q_idx: ring index in array
0040  */
0041 static void ice_qp_clean_rings(struct ice_vsi *vsi, u16 q_idx)
0042 {
0043     ice_clean_tx_ring(vsi->tx_rings[q_idx]);
0044     if (ice_is_xdp_ena_vsi(vsi)) {
0045         synchronize_rcu();
0046         ice_clean_tx_ring(vsi->xdp_rings[q_idx]);
0047     }
0048     ice_clean_rx_ring(vsi->rx_rings[q_idx]);
0049 }
0050 
0051 /**
0052  * ice_qvec_toggle_napi - Enables/disables NAPI for a given q_vector
0053  * @vsi: VSI that has netdev
0054  * @q_vector: q_vector that has NAPI context
0055  * @enable: true for enable, false for disable
0056  */
0057 static void
0058 ice_qvec_toggle_napi(struct ice_vsi *vsi, struct ice_q_vector *q_vector,
0059              bool enable)
0060 {
0061     if (!vsi->netdev || !q_vector)
0062         return;
0063 
0064     if (enable)
0065         napi_enable(&q_vector->napi);
0066     else
0067         napi_disable(&q_vector->napi);
0068 }
0069 
0070 /**
0071  * ice_qvec_dis_irq - Mask off queue interrupt generation on given ring
0072  * @vsi: the VSI that contains queue vector being un-configured
0073  * @rx_ring: Rx ring that will have its IRQ disabled
0074  * @q_vector: queue vector
0075  */
0076 static void
0077 ice_qvec_dis_irq(struct ice_vsi *vsi, struct ice_rx_ring *rx_ring,
0078          struct ice_q_vector *q_vector)
0079 {
0080     struct ice_pf *pf = vsi->back;
0081     struct ice_hw *hw = &pf->hw;
0082     int base = vsi->base_vector;
0083     u16 reg;
0084     u32 val;
0085 
0086     /* QINT_TQCTL is being cleared in ice_vsi_stop_tx_ring, so handle
0087      * here only QINT_RQCTL
0088      */
0089     reg = rx_ring->reg_idx;
0090     val = rd32(hw, QINT_RQCTL(reg));
0091     val &= ~QINT_RQCTL_CAUSE_ENA_M;
0092     wr32(hw, QINT_RQCTL(reg), val);
0093 
0094     if (q_vector) {
0095         u16 v_idx = q_vector->v_idx;
0096 
0097         wr32(hw, GLINT_DYN_CTL(q_vector->reg_idx), 0);
0098         ice_flush(hw);
0099         synchronize_irq(pf->msix_entries[v_idx + base].vector);
0100     }
0101 }
0102 
0103 /**
0104  * ice_qvec_cfg_msix - Enable IRQ for given queue vector
0105  * @vsi: the VSI that contains queue vector
0106  * @q_vector: queue vector
0107  */
0108 static void
0109 ice_qvec_cfg_msix(struct ice_vsi *vsi, struct ice_q_vector *q_vector)
0110 {
0111     u16 reg_idx = q_vector->reg_idx;
0112     struct ice_pf *pf = vsi->back;
0113     struct ice_hw *hw = &pf->hw;
0114     struct ice_tx_ring *tx_ring;
0115     struct ice_rx_ring *rx_ring;
0116 
0117     ice_cfg_itr(hw, q_vector);
0118 
0119     ice_for_each_tx_ring(tx_ring, q_vector->tx)
0120         ice_cfg_txq_interrupt(vsi, tx_ring->reg_idx, reg_idx,
0121                       q_vector->tx.itr_idx);
0122 
0123     ice_for_each_rx_ring(rx_ring, q_vector->rx)
0124         ice_cfg_rxq_interrupt(vsi, rx_ring->reg_idx, reg_idx,
0125                       q_vector->rx.itr_idx);
0126 
0127     ice_flush(hw);
0128 }
0129 
0130 /**
0131  * ice_qvec_ena_irq - Enable IRQ for given queue vector
0132  * @vsi: the VSI that contains queue vector
0133  * @q_vector: queue vector
0134  */
0135 static void ice_qvec_ena_irq(struct ice_vsi *vsi, struct ice_q_vector *q_vector)
0136 {
0137     struct ice_pf *pf = vsi->back;
0138     struct ice_hw *hw = &pf->hw;
0139 
0140     ice_irq_dynamic_ena(hw, vsi, q_vector);
0141 
0142     ice_flush(hw);
0143 }
0144 
0145 /**
0146  * ice_qp_dis - Disables a queue pair
0147  * @vsi: VSI of interest
0148  * @q_idx: ring index in array
0149  *
0150  * Returns 0 on success, negative on failure.
0151  */
0152 static int ice_qp_dis(struct ice_vsi *vsi, u16 q_idx)
0153 {
0154     struct ice_txq_meta txq_meta = { };
0155     struct ice_q_vector *q_vector;
0156     struct ice_tx_ring *tx_ring;
0157     struct ice_rx_ring *rx_ring;
0158     int timeout = 50;
0159     int err;
0160 
0161     if (q_idx >= vsi->num_rxq || q_idx >= vsi->num_txq)
0162         return -EINVAL;
0163 
0164     tx_ring = vsi->tx_rings[q_idx];
0165     rx_ring = vsi->rx_rings[q_idx];
0166     q_vector = rx_ring->q_vector;
0167 
0168     while (test_and_set_bit(ICE_CFG_BUSY, vsi->state)) {
0169         timeout--;
0170         if (!timeout)
0171             return -EBUSY;
0172         usleep_range(1000, 2000);
0173     }
0174     netif_tx_stop_queue(netdev_get_tx_queue(vsi->netdev, q_idx));
0175 
0176     ice_qvec_dis_irq(vsi, rx_ring, q_vector);
0177 
0178     ice_fill_txq_meta(vsi, tx_ring, &txq_meta);
0179     err = ice_vsi_stop_tx_ring(vsi, ICE_NO_RESET, 0, tx_ring, &txq_meta);
0180     if (err)
0181         return err;
0182     if (ice_is_xdp_ena_vsi(vsi)) {
0183         struct ice_tx_ring *xdp_ring = vsi->xdp_rings[q_idx];
0184 
0185         memset(&txq_meta, 0, sizeof(txq_meta));
0186         ice_fill_txq_meta(vsi, xdp_ring, &txq_meta);
0187         err = ice_vsi_stop_tx_ring(vsi, ICE_NO_RESET, 0, xdp_ring,
0188                        &txq_meta);
0189         if (err)
0190             return err;
0191     }
0192     err = ice_vsi_ctrl_one_rx_ring(vsi, false, q_idx, true);
0193     if (err)
0194         return err;
0195     ice_clean_rx_ring(rx_ring);
0196 
0197     ice_qvec_toggle_napi(vsi, q_vector, false);
0198     ice_qp_clean_rings(vsi, q_idx);
0199     ice_qp_reset_stats(vsi, q_idx);
0200 
0201     return 0;
0202 }
0203 
0204 /**
0205  * ice_qp_ena - Enables a queue pair
0206  * @vsi: VSI of interest
0207  * @q_idx: ring index in array
0208  *
0209  * Returns 0 on success, negative on failure.
0210  */
0211 static int ice_qp_ena(struct ice_vsi *vsi, u16 q_idx)
0212 {
0213     struct ice_aqc_add_tx_qgrp *qg_buf;
0214     struct ice_q_vector *q_vector;
0215     struct ice_tx_ring *tx_ring;
0216     struct ice_rx_ring *rx_ring;
0217     u16 size;
0218     int err;
0219 
0220     if (q_idx >= vsi->num_rxq || q_idx >= vsi->num_txq)
0221         return -EINVAL;
0222 
0223     size = struct_size(qg_buf, txqs, 1);
0224     qg_buf = kzalloc(size, GFP_KERNEL);
0225     if (!qg_buf)
0226         return -ENOMEM;
0227 
0228     qg_buf->num_txqs = 1;
0229 
0230     tx_ring = vsi->tx_rings[q_idx];
0231     rx_ring = vsi->rx_rings[q_idx];
0232     q_vector = rx_ring->q_vector;
0233 
0234     err = ice_vsi_cfg_txq(vsi, tx_ring, qg_buf);
0235     if (err)
0236         goto free_buf;
0237 
0238     if (ice_is_xdp_ena_vsi(vsi)) {
0239         struct ice_tx_ring *xdp_ring = vsi->xdp_rings[q_idx];
0240 
0241         memset(qg_buf, 0, size);
0242         qg_buf->num_txqs = 1;
0243         err = ice_vsi_cfg_txq(vsi, xdp_ring, qg_buf);
0244         if (err)
0245             goto free_buf;
0246         ice_set_ring_xdp(xdp_ring);
0247         ice_tx_xsk_pool(vsi, q_idx);
0248     }
0249 
0250     err = ice_vsi_cfg_rxq(rx_ring);
0251     if (err)
0252         goto free_buf;
0253 
0254     ice_qvec_cfg_msix(vsi, q_vector);
0255 
0256     err = ice_vsi_ctrl_one_rx_ring(vsi, true, q_idx, true);
0257     if (err)
0258         goto free_buf;
0259 
0260     clear_bit(ICE_CFG_BUSY, vsi->state);
0261     ice_qvec_toggle_napi(vsi, q_vector, true);
0262     ice_qvec_ena_irq(vsi, q_vector);
0263 
0264     netif_tx_start_queue(netdev_get_tx_queue(vsi->netdev, q_idx));
0265 free_buf:
0266     kfree(qg_buf);
0267     return err;
0268 }
0269 
0270 /**
0271  * ice_xsk_pool_disable - disable a buffer pool region
0272  * @vsi: Current VSI
0273  * @qid: queue ID
0274  *
0275  * Returns 0 on success, negative on failure
0276  */
0277 static int ice_xsk_pool_disable(struct ice_vsi *vsi, u16 qid)
0278 {
0279     struct xsk_buff_pool *pool = xsk_get_pool_from_qid(vsi->netdev, qid);
0280 
0281     if (!pool)
0282         return -EINVAL;
0283 
0284     clear_bit(qid, vsi->af_xdp_zc_qps);
0285     xsk_pool_dma_unmap(pool, ICE_RX_DMA_ATTR);
0286 
0287     return 0;
0288 }
0289 
0290 /**
0291  * ice_xsk_pool_enable - enable a buffer pool region
0292  * @vsi: Current VSI
0293  * @pool: pointer to a requested buffer pool region
0294  * @qid: queue ID
0295  *
0296  * Returns 0 on success, negative on failure
0297  */
0298 static int
0299 ice_xsk_pool_enable(struct ice_vsi *vsi, struct xsk_buff_pool *pool, u16 qid)
0300 {
0301     int err;
0302 
0303     if (vsi->type != ICE_VSI_PF)
0304         return -EINVAL;
0305 
0306     if (qid >= vsi->netdev->real_num_rx_queues ||
0307         qid >= vsi->netdev->real_num_tx_queues)
0308         return -EINVAL;
0309 
0310     err = xsk_pool_dma_map(pool, ice_pf_to_dev(vsi->back),
0311                    ICE_RX_DMA_ATTR);
0312     if (err)
0313         return err;
0314 
0315     set_bit(qid, vsi->af_xdp_zc_qps);
0316 
0317     return 0;
0318 }
0319 
0320 /**
0321  * ice_realloc_rx_xdp_bufs - reallocate for either XSK or normal buffer
0322  * @rx_ring: Rx ring
0323  * @pool_present: is pool for XSK present
0324  *
0325  * Try allocating memory and return ENOMEM, if failed to allocate.
0326  * If allocation was successful, substitute buffer with allocated one.
0327  * Returns 0 on success, negative on failure
0328  */
0329 static int
0330 ice_realloc_rx_xdp_bufs(struct ice_rx_ring *rx_ring, bool pool_present)
0331 {
0332     size_t elem_size = pool_present ? sizeof(*rx_ring->xdp_buf) :
0333                       sizeof(*rx_ring->rx_buf);
0334     void *sw_ring = kcalloc(rx_ring->count, elem_size, GFP_KERNEL);
0335 
0336     if (!sw_ring)
0337         return -ENOMEM;
0338 
0339     if (pool_present) {
0340         kfree(rx_ring->rx_buf);
0341         rx_ring->rx_buf = NULL;
0342         rx_ring->xdp_buf = sw_ring;
0343     } else {
0344         kfree(rx_ring->xdp_buf);
0345         rx_ring->xdp_buf = NULL;
0346         rx_ring->rx_buf = sw_ring;
0347     }
0348 
0349     return 0;
0350 }
0351 
0352 /**
0353  * ice_realloc_zc_buf - reallocate XDP ZC queue pairs
0354  * @vsi: Current VSI
0355  * @zc: is zero copy set
0356  *
0357  * Reallocate buffer for rx_rings that might be used by XSK.
0358  * XDP requires more memory, than rx_buf provides.
0359  * Returns 0 on success, negative on failure
0360  */
0361 int ice_realloc_zc_buf(struct ice_vsi *vsi, bool zc)
0362 {
0363     struct ice_rx_ring *rx_ring;
0364     unsigned long q;
0365 
0366     for_each_set_bit(q, vsi->af_xdp_zc_qps,
0367              max_t(int, vsi->alloc_txq, vsi->alloc_rxq)) {
0368         rx_ring = vsi->rx_rings[q];
0369         if (ice_realloc_rx_xdp_bufs(rx_ring, zc))
0370             return -ENOMEM;
0371     }
0372 
0373     return 0;
0374 }
0375 
0376 /**
0377  * ice_xsk_pool_setup - enable/disable a buffer pool region depending on its state
0378  * @vsi: Current VSI
0379  * @pool: buffer pool to enable/associate to a ring, NULL to disable
0380  * @qid: queue ID
0381  *
0382  * Returns 0 on success, negative on failure
0383  */
0384 int ice_xsk_pool_setup(struct ice_vsi *vsi, struct xsk_buff_pool *pool, u16 qid)
0385 {
0386     bool if_running, pool_present = !!pool;
0387     int ret = 0, pool_failure = 0;
0388 
0389     if (qid >= vsi->num_rxq || qid >= vsi->num_txq) {
0390         netdev_err(vsi->netdev, "Please use queue id in scope of combined queues count\n");
0391         pool_failure = -EINVAL;
0392         goto failure;
0393     }
0394 
0395     if_running = netif_running(vsi->netdev) && ice_is_xdp_ena_vsi(vsi);
0396 
0397     if (if_running) {
0398         struct ice_rx_ring *rx_ring = vsi->rx_rings[qid];
0399 
0400         ret = ice_qp_dis(vsi, qid);
0401         if (ret) {
0402             netdev_err(vsi->netdev, "ice_qp_dis error = %d\n", ret);
0403             goto xsk_pool_if_up;
0404         }
0405 
0406         ret = ice_realloc_rx_xdp_bufs(rx_ring, pool_present);
0407         if (ret)
0408             goto xsk_pool_if_up;
0409     }
0410 
0411     pool_failure = pool_present ? ice_xsk_pool_enable(vsi, pool, qid) :
0412                       ice_xsk_pool_disable(vsi, qid);
0413 
0414 xsk_pool_if_up:
0415     if (if_running) {
0416         ret = ice_qp_ena(vsi, qid);
0417         if (!ret && pool_present)
0418             napi_schedule(&vsi->rx_rings[qid]->xdp_ring->q_vector->napi);
0419         else if (ret)
0420             netdev_err(vsi->netdev, "ice_qp_ena error = %d\n", ret);
0421     }
0422 
0423 failure:
0424     if (pool_failure) {
0425         netdev_err(vsi->netdev, "Could not %sable buffer pool, error = %d\n",
0426                pool_present ? "en" : "dis", pool_failure);
0427         return pool_failure;
0428     }
0429 
0430     return ret;
0431 }
0432 
0433 /**
0434  * ice_fill_rx_descs - pick buffers from XSK buffer pool and use it
0435  * @pool: XSK Buffer pool to pull the buffers from
0436  * @xdp: SW ring of xdp_buff that will hold the buffers
0437  * @rx_desc: Pointer to Rx descriptors that will be filled
0438  * @count: The number of buffers to allocate
0439  *
0440  * This function allocates a number of Rx buffers from the fill ring
0441  * or the internal recycle mechanism and places them on the Rx ring.
0442  *
0443  * Note that ring wrap should be handled by caller of this function.
0444  *
0445  * Returns the amount of allocated Rx descriptors
0446  */
0447 static u16 ice_fill_rx_descs(struct xsk_buff_pool *pool, struct xdp_buff **xdp,
0448                  union ice_32b_rx_flex_desc *rx_desc, u16 count)
0449 {
0450     dma_addr_t dma;
0451     u16 buffs;
0452     int i;
0453 
0454     buffs = xsk_buff_alloc_batch(pool, xdp, count);
0455     for (i = 0; i < buffs; i++) {
0456         dma = xsk_buff_xdp_get_dma(*xdp);
0457         rx_desc->read.pkt_addr = cpu_to_le64(dma);
0458         rx_desc->wb.status_error0 = 0;
0459 
0460         rx_desc++;
0461         xdp++;
0462     }
0463 
0464     return buffs;
0465 }
0466 
0467 /**
0468  * __ice_alloc_rx_bufs_zc - allocate a number of Rx buffers
0469  * @rx_ring: Rx ring
0470  * @count: The number of buffers to allocate
0471  *
0472  * Place the @count of descriptors onto Rx ring. Handle the ring wrap
0473  * for case where space from next_to_use up to the end of ring is less
0474  * than @count. Finally do a tail bump.
0475  *
0476  * Returns true if all allocations were successful, false if any fail.
0477  */
0478 static bool __ice_alloc_rx_bufs_zc(struct ice_rx_ring *rx_ring, u16 count)
0479 {
0480     u32 nb_buffs_extra = 0, nb_buffs = 0;
0481     union ice_32b_rx_flex_desc *rx_desc;
0482     u16 ntu = rx_ring->next_to_use;
0483     u16 total_count = count;
0484     struct xdp_buff **xdp;
0485 
0486     rx_desc = ICE_RX_DESC(rx_ring, ntu);
0487     xdp = ice_xdp_buf(rx_ring, ntu);
0488 
0489     if (ntu + count >= rx_ring->count) {
0490         nb_buffs_extra = ice_fill_rx_descs(rx_ring->xsk_pool, xdp,
0491                            rx_desc,
0492                            rx_ring->count - ntu);
0493         if (nb_buffs_extra != rx_ring->count - ntu) {
0494             ntu += nb_buffs_extra;
0495             goto exit;
0496         }
0497         rx_desc = ICE_RX_DESC(rx_ring, 0);
0498         xdp = ice_xdp_buf(rx_ring, 0);
0499         ntu = 0;
0500         count -= nb_buffs_extra;
0501         ice_release_rx_desc(rx_ring, 0);
0502     }
0503 
0504     nb_buffs = ice_fill_rx_descs(rx_ring->xsk_pool, xdp, rx_desc, count);
0505 
0506     ntu += nb_buffs;
0507     if (ntu == rx_ring->count)
0508         ntu = 0;
0509 
0510 exit:
0511     if (rx_ring->next_to_use != ntu)
0512         ice_release_rx_desc(rx_ring, ntu);
0513 
0514     return total_count == (nb_buffs_extra + nb_buffs);
0515 }
0516 
0517 /**
0518  * ice_alloc_rx_bufs_zc - allocate a number of Rx buffers
0519  * @rx_ring: Rx ring
0520  * @count: The number of buffers to allocate
0521  *
0522  * Wrapper for internal allocation routine; figure out how many tail
0523  * bumps should take place based on the given threshold
0524  *
0525  * Returns true if all calls to internal alloc routine succeeded
0526  */
0527 bool ice_alloc_rx_bufs_zc(struct ice_rx_ring *rx_ring, u16 count)
0528 {
0529     u16 rx_thresh = ICE_RING_QUARTER(rx_ring);
0530     u16 leftover, i, tail_bumps;
0531 
0532     tail_bumps = count / rx_thresh;
0533     leftover = count - (tail_bumps * rx_thresh);
0534 
0535     for (i = 0; i < tail_bumps; i++)
0536         if (!__ice_alloc_rx_bufs_zc(rx_ring, rx_thresh))
0537             return false;
0538     return __ice_alloc_rx_bufs_zc(rx_ring, leftover);
0539 }
0540 
0541 /**
0542  * ice_bump_ntc - Bump the next_to_clean counter of an Rx ring
0543  * @rx_ring: Rx ring
0544  */
0545 static void ice_bump_ntc(struct ice_rx_ring *rx_ring)
0546 {
0547     int ntc = rx_ring->next_to_clean + 1;
0548 
0549     ntc = (ntc < rx_ring->count) ? ntc : 0;
0550     rx_ring->next_to_clean = ntc;
0551     prefetch(ICE_RX_DESC(rx_ring, ntc));
0552 }
0553 
0554 /**
0555  * ice_construct_skb_zc - Create an sk_buff from zero-copy buffer
0556  * @rx_ring: Rx ring
0557  * @xdp: Pointer to XDP buffer
0558  *
0559  * This function allocates a new skb from a zero-copy Rx buffer.
0560  *
0561  * Returns the skb on success, NULL on failure.
0562  */
0563 static struct sk_buff *
0564 ice_construct_skb_zc(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp)
0565 {
0566     unsigned int totalsize = xdp->data_end - xdp->data_meta;
0567     unsigned int metasize = xdp->data - xdp->data_meta;
0568     struct sk_buff *skb;
0569 
0570     net_prefetch(xdp->data_meta);
0571 
0572     skb = __napi_alloc_skb(&rx_ring->q_vector->napi, totalsize,
0573                    GFP_ATOMIC | __GFP_NOWARN);
0574     if (unlikely(!skb))
0575         return NULL;
0576 
0577     memcpy(__skb_put(skb, totalsize), xdp->data_meta,
0578            ALIGN(totalsize, sizeof(long)));
0579 
0580     if (metasize) {
0581         skb_metadata_set(skb, metasize);
0582         __skb_pull(skb, metasize);
0583     }
0584 
0585     xsk_buff_free(xdp);
0586     return skb;
0587 }
0588 
0589 /**
0590  * ice_run_xdp_zc - Executes an XDP program in zero-copy path
0591  * @rx_ring: Rx ring
0592  * @xdp: xdp_buff used as input to the XDP program
0593  * @xdp_prog: XDP program to run
0594  * @xdp_ring: ring to be used for XDP_TX action
0595  *
0596  * Returns any of ICE_XDP_{PASS, CONSUMED, TX, REDIR}
0597  */
0598 static int
0599 ice_run_xdp_zc(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp,
0600            struct bpf_prog *xdp_prog, struct ice_tx_ring *xdp_ring)
0601 {
0602     int err, result = ICE_XDP_PASS;
0603     u32 act;
0604 
0605     act = bpf_prog_run_xdp(xdp_prog, xdp);
0606 
0607     if (likely(act == XDP_REDIRECT)) {
0608         err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
0609         if (!err)
0610             return ICE_XDP_REDIR;
0611         if (xsk_uses_need_wakeup(rx_ring->xsk_pool) && err == -ENOBUFS)
0612             result = ICE_XDP_EXIT;
0613         else
0614             result = ICE_XDP_CONSUMED;
0615         goto out_failure;
0616     }
0617 
0618     switch (act) {
0619     case XDP_PASS:
0620         break;
0621     case XDP_TX:
0622         result = ice_xmit_xdp_buff(xdp, xdp_ring);
0623         if (result == ICE_XDP_CONSUMED)
0624             goto out_failure;
0625         break;
0626     case XDP_DROP:
0627         result = ICE_XDP_CONSUMED;
0628         break;
0629     default:
0630         bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
0631         fallthrough;
0632     case XDP_ABORTED:
0633         result = ICE_XDP_CONSUMED;
0634 out_failure:
0635         trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
0636         break;
0637     }
0638 
0639     return result;
0640 }
0641 
0642 /**
0643  * ice_clean_rx_irq_zc - consumes packets from the hardware ring
0644  * @rx_ring: AF_XDP Rx ring
0645  * @budget: NAPI budget
0646  *
0647  * Returns number of processed packets on success, remaining budget on failure.
0648  */
0649 int ice_clean_rx_irq_zc(struct ice_rx_ring *rx_ring, int budget)
0650 {
0651     unsigned int total_rx_bytes = 0, total_rx_packets = 0;
0652     struct ice_tx_ring *xdp_ring;
0653     unsigned int xdp_xmit = 0;
0654     struct bpf_prog *xdp_prog;
0655     bool failure = false;
0656     int entries_to_alloc;
0657 
0658     /* ZC patch is enabled only when XDP program is set,
0659      * so here it can not be NULL
0660      */
0661     xdp_prog = READ_ONCE(rx_ring->xdp_prog);
0662     xdp_ring = rx_ring->xdp_ring;
0663 
0664     while (likely(total_rx_packets < (unsigned int)budget)) {
0665         union ice_32b_rx_flex_desc *rx_desc;
0666         unsigned int size, xdp_res = 0;
0667         struct xdp_buff *xdp;
0668         struct sk_buff *skb;
0669         u16 stat_err_bits;
0670         u16 vlan_tag = 0;
0671         u16 rx_ptype;
0672 
0673         rx_desc = ICE_RX_DESC(rx_ring, rx_ring->next_to_clean);
0674 
0675         stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S);
0676         if (!ice_test_staterr(rx_desc->wb.status_error0, stat_err_bits))
0677             break;
0678 
0679         /* This memory barrier is needed to keep us from reading
0680          * any other fields out of the rx_desc until we have
0681          * verified the descriptor has been written back.
0682          */
0683         dma_rmb();
0684 
0685         if (unlikely(rx_ring->next_to_clean == rx_ring->next_to_use))
0686             break;
0687 
0688         xdp = *ice_xdp_buf(rx_ring, rx_ring->next_to_clean);
0689 
0690         size = le16_to_cpu(rx_desc->wb.pkt_len) &
0691                    ICE_RX_FLX_DESC_PKT_LEN_M;
0692         if (!size) {
0693             xdp->data = NULL;
0694             xdp->data_end = NULL;
0695             xdp->data_hard_start = NULL;
0696             xdp->data_meta = NULL;
0697             goto construct_skb;
0698         }
0699 
0700         xsk_buff_set_size(xdp, size);
0701         xsk_buff_dma_sync_for_cpu(xdp, rx_ring->xsk_pool);
0702 
0703         xdp_res = ice_run_xdp_zc(rx_ring, xdp, xdp_prog, xdp_ring);
0704         if (likely(xdp_res & (ICE_XDP_TX | ICE_XDP_REDIR))) {
0705             xdp_xmit |= xdp_res;
0706         } else if (xdp_res == ICE_XDP_EXIT) {
0707             failure = true;
0708             break;
0709         } else if (xdp_res == ICE_XDP_CONSUMED) {
0710             xsk_buff_free(xdp);
0711         } else if (xdp_res == ICE_XDP_PASS) {
0712             goto construct_skb;
0713         }
0714 
0715         total_rx_bytes += size;
0716         total_rx_packets++;
0717 
0718         ice_bump_ntc(rx_ring);
0719         continue;
0720 
0721 construct_skb:
0722         /* XDP_PASS path */
0723         skb = ice_construct_skb_zc(rx_ring, xdp);
0724         if (!skb) {
0725             rx_ring->rx_stats.alloc_buf_failed++;
0726             break;
0727         }
0728 
0729         ice_bump_ntc(rx_ring);
0730 
0731         if (eth_skb_pad(skb)) {
0732             skb = NULL;
0733             continue;
0734         }
0735 
0736         total_rx_bytes += skb->len;
0737         total_rx_packets++;
0738 
0739         vlan_tag = ice_get_vlan_tag_from_rx_desc(rx_desc);
0740 
0741         rx_ptype = le16_to_cpu(rx_desc->wb.ptype_flex_flags0) &
0742                        ICE_RX_FLEX_DESC_PTYPE_M;
0743 
0744         ice_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
0745         ice_receive_skb(rx_ring, skb, vlan_tag);
0746     }
0747 
0748     entries_to_alloc = ICE_DESC_UNUSED(rx_ring);
0749     if (entries_to_alloc > ICE_RING_QUARTER(rx_ring))
0750         failure |= !ice_alloc_rx_bufs_zc(rx_ring, entries_to_alloc);
0751 
0752     ice_finalize_xdp_rx(xdp_ring, xdp_xmit);
0753     ice_update_rx_ring_stats(rx_ring, total_rx_packets, total_rx_bytes);
0754 
0755     if (xsk_uses_need_wakeup(rx_ring->xsk_pool)) {
0756         if (failure || rx_ring->next_to_clean == rx_ring->next_to_use)
0757             xsk_set_rx_need_wakeup(rx_ring->xsk_pool);
0758         else
0759             xsk_clear_rx_need_wakeup(rx_ring->xsk_pool);
0760 
0761         return (int)total_rx_packets;
0762     }
0763 
0764     return failure ? budget : (int)total_rx_packets;
0765 }
0766 
0767 /**
0768  * ice_clean_xdp_tx_buf - Free and unmap XDP Tx buffer
0769  * @xdp_ring: XDP Tx ring
0770  * @tx_buf: Tx buffer to clean
0771  */
0772 static void
0773 ice_clean_xdp_tx_buf(struct ice_tx_ring *xdp_ring, struct ice_tx_buf *tx_buf)
0774 {
0775     xdp_return_frame((struct xdp_frame *)tx_buf->raw_buf);
0776     xdp_ring->xdp_tx_active--;
0777     dma_unmap_single(xdp_ring->dev, dma_unmap_addr(tx_buf, dma),
0778              dma_unmap_len(tx_buf, len), DMA_TO_DEVICE);
0779     dma_unmap_len_set(tx_buf, len, 0);
0780 }
0781 
0782 /**
0783  * ice_clean_xdp_irq_zc - produce AF_XDP descriptors to CQ
0784  * @xdp_ring: XDP Tx ring
0785  */
0786 static void ice_clean_xdp_irq_zc(struct ice_tx_ring *xdp_ring)
0787 {
0788     u16 ntc = xdp_ring->next_to_clean;
0789     struct ice_tx_desc *tx_desc;
0790     u16 cnt = xdp_ring->count;
0791     struct ice_tx_buf *tx_buf;
0792     u16 xsk_frames = 0;
0793     u16 last_rs;
0794     int i;
0795 
0796     last_rs = xdp_ring->next_to_use ? xdp_ring->next_to_use - 1 : cnt - 1;
0797     tx_desc = ICE_TX_DESC(xdp_ring, last_rs);
0798     if ((tx_desc->cmd_type_offset_bsz &
0799         cpu_to_le64(ICE_TX_DESC_DTYPE_DESC_DONE))) {
0800         if (last_rs >= ntc)
0801             xsk_frames = last_rs - ntc + 1;
0802         else
0803             xsk_frames = last_rs + cnt - ntc + 1;
0804     }
0805 
0806     if (!xsk_frames)
0807         return;
0808 
0809     if (likely(!xdp_ring->xdp_tx_active))
0810         goto skip;
0811 
0812     ntc = xdp_ring->next_to_clean;
0813     for (i = 0; i < xsk_frames; i++) {
0814         tx_buf = &xdp_ring->tx_buf[ntc];
0815 
0816         if (tx_buf->raw_buf) {
0817             ice_clean_xdp_tx_buf(xdp_ring, tx_buf);
0818             tx_buf->raw_buf = NULL;
0819         } else {
0820             xsk_frames++;
0821         }
0822 
0823         ntc++;
0824         if (ntc >= xdp_ring->count)
0825             ntc = 0;
0826     }
0827 skip:
0828     tx_desc->cmd_type_offset_bsz = 0;
0829     xdp_ring->next_to_clean += xsk_frames;
0830     if (xdp_ring->next_to_clean >= cnt)
0831         xdp_ring->next_to_clean -= cnt;
0832     if (xsk_frames)
0833         xsk_tx_completed(xdp_ring->xsk_pool, xsk_frames);
0834 }
0835 
0836 /**
0837  * ice_xmit_pkt - produce a single HW Tx descriptor out of AF_XDP descriptor
0838  * @xdp_ring: XDP ring to produce the HW Tx descriptor on
0839  * @desc: AF_XDP descriptor to pull the DMA address and length from
0840  * @total_bytes: bytes accumulator that will be used for stats update
0841  */
0842 static void ice_xmit_pkt(struct ice_tx_ring *xdp_ring, struct xdp_desc *desc,
0843              unsigned int *total_bytes)
0844 {
0845     struct ice_tx_desc *tx_desc;
0846     dma_addr_t dma;
0847 
0848     dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc->addr);
0849     xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, desc->len);
0850 
0851     tx_desc = ICE_TX_DESC(xdp_ring, xdp_ring->next_to_use++);
0852     tx_desc->buf_addr = cpu_to_le64(dma);
0853     tx_desc->cmd_type_offset_bsz = ice_build_ctob(ICE_TX_DESC_CMD_EOP,
0854                               0, desc->len, 0);
0855 
0856     *total_bytes += desc->len;
0857 }
0858 
0859 /**
0860  * ice_xmit_pkt_batch - produce a batch of HW Tx descriptors out of AF_XDP descriptors
0861  * @xdp_ring: XDP ring to produce the HW Tx descriptors on
0862  * @descs: AF_XDP descriptors to pull the DMA addresses and lengths from
0863  * @total_bytes: bytes accumulator that will be used for stats update
0864  */
0865 static void ice_xmit_pkt_batch(struct ice_tx_ring *xdp_ring, struct xdp_desc *descs,
0866                    unsigned int *total_bytes)
0867 {
0868     u16 ntu = xdp_ring->next_to_use;
0869     struct ice_tx_desc *tx_desc;
0870     u32 i;
0871 
0872     loop_unrolled_for(i = 0; i < PKTS_PER_BATCH; i++) {
0873         dma_addr_t dma;
0874 
0875         dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, descs[i].addr);
0876         xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, descs[i].len);
0877 
0878         tx_desc = ICE_TX_DESC(xdp_ring, ntu++);
0879         tx_desc->buf_addr = cpu_to_le64(dma);
0880         tx_desc->cmd_type_offset_bsz = ice_build_ctob(ICE_TX_DESC_CMD_EOP,
0881                                   0, descs[i].len, 0);
0882 
0883         *total_bytes += descs[i].len;
0884     }
0885 
0886     xdp_ring->next_to_use = ntu;
0887 }
0888 
0889 /**
0890  * ice_fill_tx_hw_ring - produce the number of Tx descriptors onto ring
0891  * @xdp_ring: XDP ring to produce the HW Tx descriptors on
0892  * @descs: AF_XDP descriptors to pull the DMA addresses and lengths from
0893  * @nb_pkts: count of packets to be send
0894  * @total_bytes: bytes accumulator that will be used for stats update
0895  */
0896 static void ice_fill_tx_hw_ring(struct ice_tx_ring *xdp_ring, struct xdp_desc *descs,
0897                 u32 nb_pkts, unsigned int *total_bytes)
0898 {
0899     u32 batched, leftover, i;
0900 
0901     batched = ALIGN_DOWN(nb_pkts, PKTS_PER_BATCH);
0902     leftover = nb_pkts & (PKTS_PER_BATCH - 1);
0903     for (i = 0; i < batched; i += PKTS_PER_BATCH)
0904         ice_xmit_pkt_batch(xdp_ring, &descs[i], total_bytes);
0905     for (; i < batched + leftover; i++)
0906         ice_xmit_pkt(xdp_ring, &descs[i], total_bytes);
0907 }
0908 
0909 /**
0910  * ice_set_rs_bit - set RS bit on last produced descriptor (one behind current NTU)
0911  * @xdp_ring: XDP ring to produce the HW Tx descriptors on
0912  */
0913 static void ice_set_rs_bit(struct ice_tx_ring *xdp_ring)
0914 {
0915     u16 ntu = xdp_ring->next_to_use ? xdp_ring->next_to_use - 1 : xdp_ring->count - 1;
0916     struct ice_tx_desc *tx_desc;
0917 
0918     tx_desc = ICE_TX_DESC(xdp_ring, ntu);
0919     tx_desc->cmd_type_offset_bsz |=
0920         cpu_to_le64(ICE_TX_DESC_CMD_RS << ICE_TXD_QW1_CMD_S);
0921 }
0922 
0923 /**
0924  * ice_xmit_zc - take entries from XSK Tx ring and place them onto HW Tx ring
0925  * @xdp_ring: XDP ring to produce the HW Tx descriptors on
0926  *
0927  * Returns true if there is no more work that needs to be done, false otherwise
0928  */
0929 bool ice_xmit_zc(struct ice_tx_ring *xdp_ring)
0930 {
0931     struct xdp_desc *descs = xdp_ring->xsk_pool->tx_descs;
0932     u32 nb_pkts, nb_processed = 0;
0933     unsigned int total_bytes = 0;
0934     int budget;
0935 
0936     ice_clean_xdp_irq_zc(xdp_ring);
0937 
0938     budget = ICE_DESC_UNUSED(xdp_ring);
0939     budget = min_t(u16, budget, ICE_RING_QUARTER(xdp_ring));
0940 
0941     nb_pkts = xsk_tx_peek_release_desc_batch(xdp_ring->xsk_pool, budget);
0942     if (!nb_pkts)
0943         return true;
0944 
0945     if (xdp_ring->next_to_use + nb_pkts >= xdp_ring->count) {
0946         nb_processed = xdp_ring->count - xdp_ring->next_to_use;
0947         ice_fill_tx_hw_ring(xdp_ring, descs, nb_processed, &total_bytes);
0948         xdp_ring->next_to_use = 0;
0949     }
0950 
0951     ice_fill_tx_hw_ring(xdp_ring, &descs[nb_processed], nb_pkts - nb_processed,
0952                 &total_bytes);
0953 
0954     ice_set_rs_bit(xdp_ring);
0955     ice_xdp_ring_update_tail(xdp_ring);
0956     ice_update_tx_ring_stats(xdp_ring, nb_pkts, total_bytes);
0957 
0958     if (xsk_uses_need_wakeup(xdp_ring->xsk_pool))
0959         xsk_set_tx_need_wakeup(xdp_ring->xsk_pool);
0960 
0961     return nb_pkts < budget;
0962 }
0963 
0964 /**
0965  * ice_xsk_wakeup - Implements ndo_xsk_wakeup
0966  * @netdev: net_device
0967  * @queue_id: queue to wake up
0968  * @flags: ignored in our case, since we have Rx and Tx in the same NAPI
0969  *
0970  * Returns negative on error, zero otherwise.
0971  */
0972 int
0973 ice_xsk_wakeup(struct net_device *netdev, u32 queue_id,
0974            u32 __always_unused flags)
0975 {
0976     struct ice_netdev_priv *np = netdev_priv(netdev);
0977     struct ice_q_vector *q_vector;
0978     struct ice_vsi *vsi = np->vsi;
0979     struct ice_tx_ring *ring;
0980 
0981     if (test_bit(ICE_VSI_DOWN, vsi->state))
0982         return -ENETDOWN;
0983 
0984     if (!ice_is_xdp_ena_vsi(vsi))
0985         return -EINVAL;
0986 
0987     if (queue_id >= vsi->num_txq || queue_id >= vsi->num_rxq)
0988         return -EINVAL;
0989 
0990     ring = vsi->rx_rings[queue_id]->xdp_ring;
0991 
0992     if (!ring->xsk_pool)
0993         return -EINVAL;
0994 
0995     /* The idea here is that if NAPI is running, mark a miss, so
0996      * it will run again. If not, trigger an interrupt and
0997      * schedule the NAPI from interrupt context. If NAPI would be
0998      * scheduled here, the interrupt affinity would not be
0999      * honored.
1000      */
1001     q_vector = ring->q_vector;
1002     if (!napi_if_scheduled_mark_missed(&q_vector->napi))
1003         ice_trigger_sw_intr(&vsi->back->hw, q_vector);
1004 
1005     return 0;
1006 }
1007 
1008 /**
1009  * ice_xsk_any_rx_ring_ena - Checks if Rx rings have AF_XDP buff pool attached
1010  * @vsi: VSI to be checked
1011  *
1012  * Returns true if any of the Rx rings has an AF_XDP buff pool attached
1013  */
1014 bool ice_xsk_any_rx_ring_ena(struct ice_vsi *vsi)
1015 {
1016     int i;
1017 
1018     ice_for_each_rxq(vsi, i) {
1019         if (xsk_get_pool_from_qid(vsi->netdev, i))
1020             return true;
1021     }
1022 
1023     return false;
1024 }
1025 
1026 /**
1027  * ice_xsk_clean_rx_ring - clean buffer pool queues connected to a given Rx ring
1028  * @rx_ring: ring to be cleaned
1029  */
1030 void ice_xsk_clean_rx_ring(struct ice_rx_ring *rx_ring)
1031 {
1032     u16 ntc = rx_ring->next_to_clean;
1033     u16 ntu = rx_ring->next_to_use;
1034 
1035     while (ntc != ntu) {
1036         struct xdp_buff *xdp = *ice_xdp_buf(rx_ring, ntc);
1037 
1038         xsk_buff_free(xdp);
1039         ntc++;
1040         if (ntc >= rx_ring->count)
1041             ntc = 0;
1042     }
1043 }
1044 
1045 /**
1046  * ice_xsk_clean_xdp_ring - Clean the XDP Tx ring and its buffer pool queues
1047  * @xdp_ring: XDP_Tx ring
1048  */
1049 void ice_xsk_clean_xdp_ring(struct ice_tx_ring *xdp_ring)
1050 {
1051     u16 ntc = xdp_ring->next_to_clean, ntu = xdp_ring->next_to_use;
1052     u32 xsk_frames = 0;
1053 
1054     while (ntc != ntu) {
1055         struct ice_tx_buf *tx_buf = &xdp_ring->tx_buf[ntc];
1056 
1057         if (tx_buf->raw_buf)
1058             ice_clean_xdp_tx_buf(xdp_ring, tx_buf);
1059         else
1060             xsk_frames++;
1061 
1062         tx_buf->raw_buf = NULL;
1063 
1064         ntc++;
1065         if (ntc >= xdp_ring->count)
1066             ntc = 0;
1067     }
1068 
1069     if (xsk_frames)
1070         xsk_tx_completed(xdp_ring->xsk_pool, xsk_frames);
1071 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!