1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2018 Intel Corporation. */
4 #include <linux/bpf_trace.h>
5 #include <net/xdp_sock_drv.h>
9 #include "ixgbe_txrx_common.h"
11 struct xsk_buff_pool *ixgbe_xsk_pool(struct ixgbe_adapter *adapter,
12 struct ixgbe_ring *ring)
14 bool xdp_on = READ_ONCE(adapter->xdp_prog);
15 int qid = ring->ring_idx;
17 if (!xdp_on || !test_bit(qid, adapter->af_xdp_zc_qps))
20 return xsk_get_pool_from_qid(adapter->netdev, qid);
23 static int ixgbe_xsk_pool_enable(struct ixgbe_adapter *adapter,
24 struct xsk_buff_pool *pool,
27 struct net_device *netdev = adapter->netdev;
31 if (qid >= adapter->num_rx_queues)
34 if (qid >= netdev->real_num_rx_queues ||
35 qid >= netdev->real_num_tx_queues)
38 err = xsk_pool_dma_map(pool, &adapter->pdev->dev, IXGBE_RX_DMA_ATTR);
42 if_running = netif_running(adapter->netdev) &&
43 ixgbe_enabled_xdp_adapter(adapter);
46 ixgbe_txrx_ring_disable(adapter, qid);
48 set_bit(qid, adapter->af_xdp_zc_qps);
51 ixgbe_txrx_ring_enable(adapter, qid);
53 /* Kick start the NAPI context so that receiving will start */
54 err = ixgbe_xsk_wakeup(adapter->netdev, qid, XDP_WAKEUP_RX);
62 static int ixgbe_xsk_pool_disable(struct ixgbe_adapter *adapter, u16 qid)
64 struct xsk_buff_pool *pool;
67 pool = xsk_get_pool_from_qid(adapter->netdev, qid);
71 if_running = netif_running(adapter->netdev) &&
72 ixgbe_enabled_xdp_adapter(adapter);
75 ixgbe_txrx_ring_disable(adapter, qid);
77 clear_bit(qid, adapter->af_xdp_zc_qps);
78 xsk_pool_dma_unmap(pool, IXGBE_RX_DMA_ATTR);
81 ixgbe_txrx_ring_enable(adapter, qid);
86 int ixgbe_xsk_pool_setup(struct ixgbe_adapter *adapter,
87 struct xsk_buff_pool *pool,
90 return pool ? ixgbe_xsk_pool_enable(adapter, pool, qid) :
91 ixgbe_xsk_pool_disable(adapter, qid);
94 static int ixgbe_run_xdp_zc(struct ixgbe_adapter *adapter,
95 struct ixgbe_ring *rx_ring,
98 int err, result = IXGBE_XDP_PASS;
99 struct bpf_prog *xdp_prog;
100 struct xdp_frame *xdpf;
104 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
105 act = bpf_prog_run_xdp(xdp_prog, xdp);
107 if (likely(act == XDP_REDIRECT)) {
108 err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
109 result = !err ? IXGBE_XDP_REDIR : IXGBE_XDP_CONSUMED;
118 xdpf = xdp_convert_buff_to_frame(xdp);
119 if (unlikely(!xdpf)) {
120 result = IXGBE_XDP_CONSUMED;
123 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
126 bpf_warn_invalid_xdp_action(act);
129 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
130 fallthrough; /* handle aborts by dropping packet */
132 result = IXGBE_XDP_CONSUMED;
139 bool ixgbe_alloc_rx_buffers_zc(struct ixgbe_ring *rx_ring, u16 count)
141 union ixgbe_adv_rx_desc *rx_desc;
142 struct ixgbe_rx_buffer *bi;
143 u16 i = rx_ring->next_to_use;
151 rx_desc = IXGBE_RX_DESC(rx_ring, i);
152 bi = &rx_ring->rx_buffer_info[i];
156 bi->xdp = xsk_buff_alloc(rx_ring->xsk_pool);
162 dma = xsk_buff_xdp_get_dma(bi->xdp);
164 /* Refresh the desc even if buffer_addrs didn't change
165 * because each write-back erases this info.
167 rx_desc->read.pkt_addr = cpu_to_le64(dma);
173 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
174 bi = rx_ring->rx_buffer_info;
178 /* clear the length for the next_to_use descriptor */
179 rx_desc->wb.upper.length = 0;
186 if (rx_ring->next_to_use != i) {
187 rx_ring->next_to_use = i;
189 /* Force memory writes to complete before letting h/w
190 * know there are new descriptors to fetch. (Only
191 * applicable for weak-ordered memory model archs,
195 writel(i, rx_ring->tail);
201 static struct sk_buff *ixgbe_construct_skb_zc(struct ixgbe_ring *rx_ring,
202 struct ixgbe_rx_buffer *bi)
204 unsigned int metasize = bi->xdp->data - bi->xdp->data_meta;
205 unsigned int datasize = bi->xdp->data_end - bi->xdp->data;
208 /* allocate a skb to store the frags */
209 skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
210 bi->xdp->data_end - bi->xdp->data_hard_start,
211 GFP_ATOMIC | __GFP_NOWARN);
215 skb_reserve(skb, bi->xdp->data - bi->xdp->data_hard_start);
216 memcpy(__skb_put(skb, datasize), bi->xdp->data, datasize);
218 skb_metadata_set(skb, metasize);
220 xsk_buff_free(bi->xdp);
225 static void ixgbe_inc_ntc(struct ixgbe_ring *rx_ring)
227 u32 ntc = rx_ring->next_to_clean + 1;
229 ntc = (ntc < rx_ring->count) ? ntc : 0;
230 rx_ring->next_to_clean = ntc;
231 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
234 int ixgbe_clean_rx_irq_zc(struct ixgbe_q_vector *q_vector,
235 struct ixgbe_ring *rx_ring,
238 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
239 struct ixgbe_adapter *adapter = q_vector->adapter;
240 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
241 unsigned int xdp_res, xdp_xmit = 0;
242 bool failure = false;
245 while (likely(total_rx_packets < budget)) {
246 union ixgbe_adv_rx_desc *rx_desc;
247 struct ixgbe_rx_buffer *bi;
250 /* return some buffers to hardware, one at a time is too slow */
251 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
253 !ixgbe_alloc_rx_buffers_zc(rx_ring,
258 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
259 size = le16_to_cpu(rx_desc->wb.upper.length);
263 /* This memory barrier is needed to keep us from reading
264 * any other fields out of the rx_desc until we know the
265 * descriptor has been written back
269 bi = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
271 if (unlikely(!ixgbe_test_staterr(rx_desc,
272 IXGBE_RXD_STAT_EOP))) {
273 struct ixgbe_rx_buffer *next_bi;
275 xsk_buff_free(bi->xdp);
277 ixgbe_inc_ntc(rx_ring);
279 &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
280 next_bi->discard = true;
284 if (unlikely(bi->discard)) {
285 xsk_buff_free(bi->xdp);
288 ixgbe_inc_ntc(rx_ring);
292 bi->xdp->data_end = bi->xdp->data + size;
293 xsk_buff_dma_sync_for_cpu(bi->xdp, rx_ring->xsk_pool);
294 xdp_res = ixgbe_run_xdp_zc(adapter, rx_ring, bi->xdp);
297 if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR))
300 xsk_buff_free(bi->xdp);
304 total_rx_bytes += size;
307 ixgbe_inc_ntc(rx_ring);
312 skb = ixgbe_construct_skb_zc(rx_ring, bi);
314 rx_ring->rx_stats.alloc_rx_buff_failed++;
319 ixgbe_inc_ntc(rx_ring);
321 if (eth_skb_pad(skb))
324 total_rx_bytes += skb->len;
327 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
328 ixgbe_rx_skb(q_vector, skb);
331 if (xdp_xmit & IXGBE_XDP_REDIR)
334 if (xdp_xmit & IXGBE_XDP_TX) {
335 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
337 /* Force memory writes to complete before letting h/w
338 * know there are new descriptors to fetch.
341 writel(ring->next_to_use, ring->tail);
344 u64_stats_update_begin(&rx_ring->syncp);
345 rx_ring->stats.packets += total_rx_packets;
346 rx_ring->stats.bytes += total_rx_bytes;
347 u64_stats_update_end(&rx_ring->syncp);
348 q_vector->rx.total_packets += total_rx_packets;
349 q_vector->rx.total_bytes += total_rx_bytes;
351 if (xsk_uses_need_wakeup(rx_ring->xsk_pool)) {
352 if (failure || rx_ring->next_to_clean == rx_ring->next_to_use)
353 xsk_set_rx_need_wakeup(rx_ring->xsk_pool);
355 xsk_clear_rx_need_wakeup(rx_ring->xsk_pool);
357 return (int)total_rx_packets;
359 return failure ? budget : (int)total_rx_packets;
362 void ixgbe_xsk_clean_rx_ring(struct ixgbe_ring *rx_ring)
364 struct ixgbe_rx_buffer *bi;
367 for (i = 0; i < rx_ring->count; i++) {
368 bi = &rx_ring->rx_buffer_info[i];
373 xsk_buff_free(bi->xdp);
378 static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
380 struct xsk_buff_pool *pool = xdp_ring->xsk_pool;
381 union ixgbe_adv_tx_desc *tx_desc = NULL;
382 struct ixgbe_tx_buffer *tx_bi;
383 bool work_done = true;
384 struct xdp_desc desc;
388 while (budget-- > 0) {
389 if (unlikely(!ixgbe_desc_unused(xdp_ring)) ||
390 !netif_carrier_ok(xdp_ring->netdev)) {
395 if (!xsk_tx_peek_desc(pool, &desc))
398 dma = xsk_buff_raw_get_dma(pool, desc.addr);
399 xsk_buff_raw_dma_sync_for_device(pool, dma, desc.len);
401 tx_bi = &xdp_ring->tx_buffer_info[xdp_ring->next_to_use];
402 tx_bi->bytecount = desc.len;
406 tx_desc = IXGBE_TX_DESC(xdp_ring, xdp_ring->next_to_use);
407 tx_desc->read.buffer_addr = cpu_to_le64(dma);
409 /* put descriptor type bits */
410 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
411 IXGBE_ADVTXD_DCMD_DEXT |
412 IXGBE_ADVTXD_DCMD_IFCS;
413 cmd_type |= desc.len | IXGBE_TXD_CMD;
414 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
415 tx_desc->read.olinfo_status =
416 cpu_to_le32(desc.len << IXGBE_ADVTXD_PAYLEN_SHIFT);
418 xdp_ring->next_to_use++;
419 if (xdp_ring->next_to_use == xdp_ring->count)
420 xdp_ring->next_to_use = 0;
424 ixgbe_xdp_ring_update_tail(xdp_ring);
425 xsk_tx_release(pool);
428 return !!budget && work_done;
431 static void ixgbe_clean_xdp_tx_buffer(struct ixgbe_ring *tx_ring,
432 struct ixgbe_tx_buffer *tx_bi)
434 xdp_return_frame(tx_bi->xdpf);
435 dma_unmap_single(tx_ring->dev,
436 dma_unmap_addr(tx_bi, dma),
437 dma_unmap_len(tx_bi, len), DMA_TO_DEVICE);
438 dma_unmap_len_set(tx_bi, len, 0);
441 bool ixgbe_clean_xdp_tx_irq(struct ixgbe_q_vector *q_vector,
442 struct ixgbe_ring *tx_ring, int napi_budget)
444 u16 ntc = tx_ring->next_to_clean, ntu = tx_ring->next_to_use;
445 unsigned int total_packets = 0, total_bytes = 0;
446 struct xsk_buff_pool *pool = tx_ring->xsk_pool;
447 union ixgbe_adv_tx_desc *tx_desc;
448 struct ixgbe_tx_buffer *tx_bi;
451 tx_bi = &tx_ring->tx_buffer_info[ntc];
452 tx_desc = IXGBE_TX_DESC(tx_ring, ntc);
455 if (!(tx_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
458 total_bytes += tx_bi->bytecount;
459 total_packets += tx_bi->gso_segs;
462 ixgbe_clean_xdp_tx_buffer(tx_ring, tx_bi);
471 if (unlikely(ntc == tx_ring->count)) {
473 tx_bi = tx_ring->tx_buffer_info;
474 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
477 /* issue prefetch for next Tx descriptor */
481 tx_ring->next_to_clean = ntc;
483 u64_stats_update_begin(&tx_ring->syncp);
484 tx_ring->stats.bytes += total_bytes;
485 tx_ring->stats.packets += total_packets;
486 u64_stats_update_end(&tx_ring->syncp);
487 q_vector->tx.total_bytes += total_bytes;
488 q_vector->tx.total_packets += total_packets;
491 xsk_tx_completed(pool, xsk_frames);
493 if (xsk_uses_need_wakeup(pool))
494 xsk_set_tx_need_wakeup(pool);
496 return ixgbe_xmit_zc(tx_ring, q_vector->tx.work_limit);
499 int ixgbe_xsk_wakeup(struct net_device *dev, u32 qid, u32 flags)
501 struct ixgbe_adapter *adapter = netdev_priv(dev);
502 struct ixgbe_ring *ring;
504 if (test_bit(__IXGBE_DOWN, &adapter->state))
507 if (!READ_ONCE(adapter->xdp_prog))
510 if (qid >= adapter->num_xdp_queues)
513 ring = adapter->xdp_ring[qid];
515 if (test_bit(__IXGBE_TX_DISABLED, &ring->state))
521 if (!napi_if_scheduled_mark_missed(&ring->q_vector->napi)) {
522 u64 eics = BIT_ULL(ring->q_vector->v_idx);
524 ixgbe_irq_rearm_queues(adapter, eics);
530 void ixgbe_xsk_clean_tx_ring(struct ixgbe_ring *tx_ring)
532 u16 ntc = tx_ring->next_to_clean, ntu = tx_ring->next_to_use;
533 struct xsk_buff_pool *pool = tx_ring->xsk_pool;
534 struct ixgbe_tx_buffer *tx_bi;
538 tx_bi = &tx_ring->tx_buffer_info[ntc];
541 ixgbe_clean_xdp_tx_buffer(tx_ring, tx_bi);
548 if (ntc == tx_ring->count)
553 xsk_tx_completed(pool, xsk_frames);