octeontx2-pf: Add XDP support to netdev PF

Adds XDP_PASS, XDP_TX, XDP_DROP and XDP_REDIRECT support
for netdev PF.

Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 159 insertions, 15 deletions

diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c
index 3f3ec8ffc4dd..2d2b67319340 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c
@@ -8,6 +8,8 @@
 #include <linux/etherdevice.h>
 #include <net/ip.h>
 #include <net/tso.h>
+#include <linux/bpf.h>
+#include <linux/bpf_trace.h>
 
 #include "otx2_reg.h"
 #include "otx2_common.h"
@@ -17,6 +19,10 @@
 #include "cn10k.h"
 
 #define CQE_ADDR(CQ, idx) ((CQ)->cqe_base + ((CQ)->cqe_size * (idx)))
+static bool otx2_xdp_rcv_pkt_handler(struct otx2_nic *pfvf,
+				     struct bpf_prog *prog,
+				     struct nix_cqe_rx_s *cqe,
+				     struct otx2_cq_queue *cq);
 
 static int otx2_nix_cq_op_status(struct otx2_nic *pfvf,
 				 struct otx2_cq_queue *cq)
@@ -98,6 +104,24 @@ static void otx2_dma_unmap_skb_frags(struct otx2_nic *pfvf, struct sg_list *sg)
 	sg->num_segs = 0;
 }
 
+static void otx2_xdp_snd_pkt_handler(struct otx2_nic *pfvf,
+				     struct otx2_snd_queue *sq,
+				 struct nix_cqe_tx_s *cqe)
+{
+	struct nix_send_comp_s *snd_comp = &cqe->comp;
+	struct sg_list *sg;
+	struct page *page;
+	u64 pa;
+
+	sg = &sq->sg[snd_comp->sqe_id];
+
+	pa = otx2_iova_to_phys(pfvf->iommu_domain, sg->dma_addr[0]);
+	otx2_dma_unmap_page(pfvf, sg->dma_addr[0],
+			    sg->size[0], DMA_TO_DEVICE);
+	page = virt_to_page(phys_to_virt(pa));
+	put_page(page);
+}
+
 static void otx2_snd_pkt_handler(struct otx2_nic *pfvf,
 				 struct otx2_cq_queue *cq,
 				 struct otx2_snd_queue *sq,
@@ -310,6 +334,10 @@ static void otx2_rcv_pkt_handler(struct otx2_nic *pfvf,
 			return;
 	}
 
+	if (pfvf->xdp_prog)
+		if (otx2_xdp_rcv_pkt_handler(pfvf, pfvf->xdp_prog, cqe, cq))
+			return;
+
 	skb = napi_get_frags(napi);
 	if (unlikely(!skb))
 		return;
@@ -373,11 +401,6 @@ process_cqe:
 	otx2_write64(pfvf, NIX_LF_CQ_OP_DOOR,
 		     ((u64)cq->cq_idx << 32) | processed_cqe);
 
-	if (unlikely(!cq->pool_ptrs))
-		return 0;
-	/* Refill pool with new buffers */
-	pfvf->hw_ops->refill_pool_ptrs(pfvf, cq);
-
 	return processed_cqe;
 }
 
@@ -397,7 +420,7 @@ void otx2_refill_pool_ptrs(void *dev, struct otx2_cq_queue *cq)
 static int otx2_tx_napi_handler(struct otx2_nic *pfvf,
 				struct otx2_cq_queue *cq, int budget)
 {
-	int tx_pkts = 0, tx_bytes = 0;
+	int tx_pkts = 0, tx_bytes = 0, qidx;
 	struct nix_cqe_tx_s *cqe;
 	int processed_cqe = 0;
 
@@ -415,9 +438,15 @@ process_cqe:
 				return 0;
 			break;
 		}
-		otx2_snd_pkt_handler(pfvf, cq, &pfvf->qset.sq[cq->cint_idx],
-				     cqe, budget, &tx_pkts, &tx_bytes);
-
+		if (cq->cq_type == CQ_XDP) {
+			qidx = cq->cq_idx - pfvf->hw.rx_queues;
+			otx2_xdp_snd_pkt_handler(pfvf, &pfvf->qset.sq[qidx],
+						 cqe);
+		} else {
+			otx2_snd_pkt_handler(pfvf, cq,
+					     &pfvf->qset.sq[cq->cint_idx],
+					     cqe, budget, &tx_pkts, &tx_bytes);
+		}
 		cqe->hdr.cqe_type = NIX_XQE_TYPE_INVALID;
 		processed_cqe++;
 		cq->pend_cqe--;
@@ -443,6 +472,7 @@ process_cqe:
 
 int otx2_napi_handler(struct napi_struct *napi, int budget)
 {
+	struct otx2_cq_queue *rx_cq = NULL;
 	struct otx2_cq_poll *cq_poll;
 	int workdone = 0, cq_idx, i;
 	struct otx2_cq_queue *cq;
@@ -453,17 +483,13 @@ int otx2_napi_handler(struct napi_struct *napi, int budget)
 	pfvf = (struct otx2_nic *)cq_poll->dev;
 	qset = &pfvf->qset;
 
-	for (i = CQS_PER_CINT - 1; i >= 0; i--) {
+	for (i = 0; i < CQS_PER_CINT; i++) {
 		cq_idx = cq_poll->cq_ids[i];
 		if (unlikely(cq_idx == CINT_INVALID_CQ))
 			continue;
 		cq = &qset->cq[cq_idx];
 		if (cq->cq_type == CQ_RX) {
-			/* If the RQ refill WQ task is running, skip napi
-			 * scheduler for this queue.
-			 */
-			if (cq->refill_task_sched)
-				continue;
+			rx_cq = cq;
 			workdone += otx2_rx_napi_handler(pfvf, napi,
 							 cq, budget);
 		} else {
@@ -471,6 +497,8 @@ int otx2_napi_handler(struct napi_struct *napi, int budget)
 		}
 	}
 
+	if (rx_cq && rx_cq->pool_ptrs)
+		pfvf->hw_ops->refill_pool_ptrs(pfvf, rx_cq);
 	/* Clear the IRQ */
 	otx2_write64(pfvf, NIX_LF_CINTX_INT(cq_poll->cint_idx), BIT_ULL(0));
 
@@ -977,6 +1005,9 @@ void otx2_cleanup_rx_cqes(struct otx2_nic *pfvf, struct otx2_cq_queue *cq)
 	int processed_cqe = 0;
 	u64 iova, pa;
 
+	if (pfvf->xdp_prog)
+		xdp_rxq_info_unreg(&cq->xdp_rxq);
+
 	if (otx2_nix_cq_op_status(pfvf, cq) || !cq->pend_cqe)
 		return;
 
@@ -1056,3 +1087,116 @@ int otx2_rxtx_enable(struct otx2_nic *pfvf, bool enable)
 	mutex_unlock(&pfvf->mbox.lock);
 	return err;
 }
+
+static void otx2_xdp_sqe_add_sg(struct otx2_snd_queue *sq, u64 dma_addr,
+				int len, int *offset)
+{
+	struct nix_sqe_sg_s *sg = NULL;
+	u64 *iova = NULL;
+
+	sg = (struct nix_sqe_sg_s *)(sq->sqe_base + *offset);
+	sg->ld_type = NIX_SEND_LDTYPE_LDD;
+	sg->subdc = NIX_SUBDC_SG;
+	sg->segs = 1;
+	sg->seg1_size = len;
+	iova = (void *)sg + sizeof(*sg);
+	*iova = dma_addr;
+	*offset += sizeof(*sg) + sizeof(u64);
+
+	sq->sg[sq->head].dma_addr[0] = dma_addr;
+	sq->sg[sq->head].size[0] = len;
+	sq->sg[sq->head].num_segs = 1;
+}
+
+bool otx2_xdp_sq_append_pkt(struct otx2_nic *pfvf, u64 iova, int len, u16 qidx)
+{
+	struct nix_sqe_hdr_s *sqe_hdr;
+	struct otx2_snd_queue *sq;
+	int offset, free_sqe;
+
+	sq = &pfvf->qset.sq[qidx];
+	free_sqe = (sq->num_sqbs - *sq->aura_fc_addr) * sq->sqe_per_sqb;
+	if (free_sqe < sq->sqe_thresh)
+		return false;
+
+	memset(sq->sqe_base + 8, 0, sq->sqe_size - 8);
+
+	sqe_hdr = (struct nix_sqe_hdr_s *)(sq->sqe_base);
+
+	if (!sqe_hdr->total) {
+		sqe_hdr->aura = sq->aura_id;
+		sqe_hdr->df = 1;
+		sqe_hdr->sq = qidx;
+		sqe_hdr->pnc = 1;
+	}
+	sqe_hdr->total = len;
+	sqe_hdr->sqe_id = sq->head;
+
+	offset = sizeof(*sqe_hdr);
+
+	otx2_xdp_sqe_add_sg(sq, iova, len, &offset);
+	sqe_hdr->sizem1 = (offset / 16) - 1;
+	pfvf->hw_ops->sqe_flush(pfvf, sq, offset, qidx);
+
+	return true;
+}
+
+static bool otx2_xdp_rcv_pkt_handler(struct otx2_nic *pfvf,
+				     struct bpf_prog *prog,
+				     struct nix_cqe_rx_s *cqe,
+				     struct otx2_cq_queue *cq)
+{
+	unsigned char *hard_start, *data;
+	int qidx = cq->cq_idx;
+	struct xdp_buff xdp;
+	struct page *page;
+	u64 iova, pa;
+	u32 act;
+	int err;
+
+	iova = cqe->sg.seg_addr - OTX2_HEAD_ROOM;
+	pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
+	page = virt_to_page(phys_to_virt(pa));
+
+	xdp_init_buff(&xdp, pfvf->rbsize, &cq->xdp_rxq);
+
+	data = (unsigned char *)phys_to_virt(pa);
+	hard_start = page_address(page);
+	xdp_prepare_buff(&xdp, hard_start, data - hard_start,
+			 cqe->sg.seg_size, false);
+
+	act = bpf_prog_run_xdp(prog, &xdp);
+
+	switch (act) {
+	case XDP_PASS:
+		break;
+	case XDP_TX:
+		qidx += pfvf->hw.tx_queues;
+		cq->pool_ptrs++;
+		return otx2_xdp_sq_append_pkt(pfvf, iova,
+					      cqe->sg.seg_size, qidx);
+	case XDP_REDIRECT:
+		cq->pool_ptrs++;
+		err = xdp_do_redirect(pfvf->netdev, &xdp, prog);
+
+		otx2_dma_unmap_page(pfvf, iova, pfvf->rbsize,
+				    DMA_FROM_DEVICE);
+		if (!err)
+			return true;
+		put_page(page);
+		break;
+	default:
+		bpf_warn_invalid_xdp_action(act);
+		break;
+	case XDP_ABORTED:
+		trace_xdp_exception(pfvf->netdev, prog, act);
+		break;
+	case XDP_DROP:
+		otx2_dma_unmap_page(pfvf, iova, pfvf->rbsize,
+				    DMA_FROM_DEVICE);
+		put_page(page);
+		cq->pool_ptrs++;
+		return true;
+	}
+	return false;
+}