Re: [PATCH net-next v2 12/14] octeontx2-pf: ipsec: Process CPT metapackets

On 2025-06-20 at 16:36:57, Simon Horman (horms@kernel.org) wrote:

On Wed, Jun 18, 2025 at 05:00:06PM +0530, Tanmay Jagdale wrote:

quoted

CPT hardware forwards decrypted IPsec packets to NIX via the X2P bus
as metapackets which are of 256 bytes in length. Each metapacket
contains CPT_PARSE_HDR_S and initial bytes of the decrypted packet
that helps NIX RX in classifying and submitting to CPU. Additionally,
CPT also sets BIT(11) of the channel number to indicate that it's a
2nd pass packet from CPT.

Since the metapackets are not complete packets, they don't have to go
through L3/L4 layer length and checksum verification so these are
disabled via the NIX_LF_INLINE_RQ_CFG mailbox during IPsec initialization.

The CPT_PARSE_HDR_S contains a WQE pointer to the complete decrypted
packet. Add code in the rx NAPI handler to parse the header and extract
WQE pointer. Later, use this WQE pointer to construct the skb, set the
XFRM packet mode flags to indicate successful decryption before submitting
it to the network stack.

Signed-off-by: Tanmay Jagdale <redacted>
---
Changes in V2:
- Removed unnecessary casts
- Don't convert complete cpt_parse_hdr from BE to LE and just
  convert required fields
- Fixed logic to avoid repeated calculation for start and end in sg

V1 Link: https://lore.kernel.org/netdev/20250502132005.611698-15-tanmay@marvell.com/ (local)

 .../marvell/octeontx2/nic/cn10k_ipsec.c       | 52 +++++++++++++++++++
 .../marvell/octeontx2/nic/cn10k_ipsec.h       | 48 +++++++++++++++++
 .../marvell/octeontx2/nic/otx2_common.h       |  2 +
 .../marvell/octeontx2/nic/otx2_struct.h       | 16 ++++++
 .../marvell/octeontx2/nic/otx2_txrx.c         | 29 +++++++++--
 5 files changed, 144 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k_ipsec.c b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k_ipsec.c
index 5cb6bc835e56..a95878378334 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k_ipsec.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k_ipsec.c

@@ -346,6 +346,58 @@ static int cn10k_outb_cpt_init(struct net_device *netdev)
 	return ret;
 }
 
+struct nix_wqe_rx_s *cn10k_ipsec_process_cpt_metapkt(struct otx2_nic *pfvf,
+						     struct nix_cqe_rx_s *cqe,
+						     struct sk_buff *skb,
+						     int qidx)
+{
+	struct nix_rx_sg_s *sg = &cqe->sg;
+	struct nix_wqe_rx_s *wqe = NULL;
+	u64 *seg_addr = &sg->seg_addr;
+	struct cpt_parse_hdr_s *cptp;
+	struct xfrm_offload *xo;
+	struct xfrm_state *xs;
+	struct sec_path *sp;
+	void *va;
+
+	/* CPT_PARSE_HDR_S is present in the beginning of the buffer */
+	va = phys_to_virt(otx2_iova_to_phys(pfvf->iommu_domain, *seg_addr));
+
+	cptp = (struct cpt_parse_hdr_s *)va;
+
+	/* Convert the wqe_ptr from CPT_PARSE_HDR_S to a CPU usable pointer */
+	wqe = phys_to_virt(otx2_iova_to_phys(pfvf->iommu_domain,
+					     be64_to_cpu(cptp->wqe_ptr)));

Hi Tanmay,

Hi Simon,

be64_to_cpu expects a __be64 argument, but the type of cptp->wqe_ptr is u64.
Or, IOW, be64_to_cpu expects to be based a big endian value but
the type of it's argument is host byte order.

Okay. I will fix the structure definition to use __be types.

quoted

+
+	/* Get the XFRM state pointer stored in SA context */
+	xs = pfvf->ipsec.inb_sa->base +
+	     (be32_to_cpu(cptp->cookie) * pfvf->ipsec.sa_tbl_entry_sz) + 1024;

Likewise with cookie here.

ACK.

:Flagged by Sparse.

...

With Regards,
Tanmay