Re: [net-next PATCH v6 3/6] octeontx2-pf: AF_XDP zero copy receive support
From: Kees Bakker <hidden>
Date: 2025-02-19 21:37:19
Also in:
bpf, lkml
Op 13-02-2025 om 06:31 schreef Suman Ghosh:
quoted hunk ↗ jump to hunk
This patch adds support to AF_XDP zero copy for CN10K. This patch specifically adds receive side support. In this approach once a xdp program with zero copy support on a specific rx queue is enabled, then that receive quse is disabled/detached from the existing kernel queue and re-assigned to the umem memory. Signed-off-by: Suman Ghosh <redacted> --- .../ethernet/marvell/octeontx2/nic/Makefile | 2 +- .../ethernet/marvell/octeontx2/nic/cn10k.c | 7 +- .../marvell/octeontx2/nic/otx2_common.c | 114 ++++++++--- .../marvell/octeontx2/nic/otx2_common.h | 6 +- .../ethernet/marvell/octeontx2/nic/otx2_pf.c | 25 ++- .../marvell/octeontx2/nic/otx2_txrx.c | 73 +++++-- .../marvell/octeontx2/nic/otx2_txrx.h | 6 + .../ethernet/marvell/octeontx2/nic/otx2_vf.c | 12 +- .../ethernet/marvell/octeontx2/nic/otx2_xsk.c | 182 ++++++++++++++++++ .../ethernet/marvell/octeontx2/nic/otx2_xsk.h | 21 ++ .../ethernet/marvell/octeontx2/nic/qos_sq.c | 2 +- 11 files changed, 389 insertions(+), 61 deletions(-) create mode 100644 drivers/net/ethernet/marvell/octeontx2/nic/otx2_xsk.c create mode 100644 drivers/net/ethernet/marvell/octeontx2/nic/otx2_xsk.h [...]diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_xsk.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_xsk.c new file mode 100644 index 000000000000..894c1e0aea6f --- /dev/null +++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_xsk.c[...] +static void otx2_clean_up_rq(struct otx2_nic *pfvf, int qidx) +{ + struct otx2_qset *qset = &pfvf->qset; + struct otx2_cq_queue *cq; + struct otx2_pool *pool; + u64 iova; + + /* If the DOWN flag is set SQs are already freed */ + if (pfvf->flags & OTX2_FLAG_INTF_DOWN) + return; + + cq = &qset->cq[qidx]; + if (cq) + otx2_cleanup_rx_cqes(pfvf, cq, qidx);
The if check makes no sense, cq is always != NULL
+
[...]
+int otx2_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
+{
+ struct otx2_nic *pf = netdev_priv(dev);
+ struct otx2_cq_poll *cq_poll = NULL;
+ struct otx2_qset *qset = &pf->qset;
+
+ if (pf->flags & OTX2_FLAG_INTF_DOWN)
+ return -ENETDOWN;
+
+ if (queue_id >= pf->hw.rx_queues)
+ return -EINVAL;
+
+ cq_poll = &qset->napi[queue_id];
+ if (!cq_poll)
+ return -EINVAL;cq_poll can never be NULL.
[...]