[PATCH 1/1] drivers: fix CN20K mbuf size truncation
Stephen Hemminger
stephen at networkplumber.org
Wed Sep 9 19:16:18 CEST 2026
stop "cargo culting" use of always inline attribute. that attribute should
only be used in special cases where code breaks compilation without it. it
is not a go faster flag
On Wed, Sep 9, 2026, 09:06 Randy L Tice <rtice at cisco.com> wrote:
> CN20K inline security receive paths recover the packet mbuf by
> subtracting sizeof(struct rte_mbuf) from a pointer stored in
> completion metadata.
>
> The size is currently kept in a uint8_t local variable before
> that subtraction. This truncates the value when the mbuf structure
> grows beyond 255 bytes, and can recover the wrong mbuf address.
>
> Use a uint32_t local value for the mbuf byte size so larger mbuf
> layouts are handled correctly.
>
> Fixes: 5856f23129bb ("net/cnxk: support CN20K inline IPsec Rx")
> Fixes: edd0d5f3c299 ("event/cnxk: support CN20K inline IPsec Rx")
> Cc: stable at dpdk.org
>
> Signed-off-by: Randy L Tice <rtice at cisco.com>
> ---
> .mailmap | 1 +
> drivers/event/cnxk/cn20k_worker.h | 4 ++--
> drivers/net/cnxk/cn20k_rx.h | 4 ++--
> 3 files changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/.mailmap b/.mailmap
> index fcb3d1bb3f..2a8b54ea23 100644
> --- a/.mailmap
> +++ b/.mailmap
> @@ -1379,6 +1379,7 @@ Rakesh Kudurumalla <rkudurumalla at marvell.com> <
> rkudurumalla at caviumnetworks.com>
> Ralf Hoffmann <ralf.hoffmann at allegro-packets.com>
> Rami Rosen <ramirose at gmail.com> <rami.rosen at intel.com>
> Rami Rosen <ramirose at gmail.com> <roszenrami at gmail.com>
> +Randy L Tice <rtice at cisco.com>
> Randy Schacher <stuart.schacher at broadcom.com>
> Rani Sharoni <ranish at nvidia.com>
> Ranjit Menon <ranjit.menon at intel.com>
> diff --git a/drivers/event/cnxk/cn20k_worker.h
> b/drivers/event/cnxk/cn20k_worker.h
> index 6442113e09..5723a6eabb 100644
> --- a/drivers/event/cnxk/cn20k_worker.h
> +++ b/drivers/event/cnxk/cn20k_worker.h
> @@ -48,7 +48,7 @@ cn20k_process_vwqe(uintptr_t vwqe, uint16_t port_id,
> const uint32_t flags, struc
> {
> uint64_t mbuf_init = 0x100010000ULL | RTE_PKTMBUF_HEADROOM;
> struct cnxk_timesync_info *tstamp = ws->tstamp[port_id];
> - uint8_t m_sz = sizeof(struct rte_mbuf);
> + const uint32_t m_sz = sizeof(struct rte_mbuf);
> void *lookup_mem = ws->lookup_mem;
> uint64_t meta_aura = 0, laddr = 0;
> uintptr_t lbase = ws->lmt_base;
> @@ -165,7 +165,7 @@ cn20k_process_vwqe(uintptr_t vwqe, uint16_t port_id,
> const uint32_t flags, struc
> static __rte_always_inline void
> cn20k_sso_hws_post_process(struct cn20k_sso_hws *ws, uint64_t *u64, const
> uint32_t flags)
> {
> - uint8_t m_sz = sizeof(struct rte_mbuf);
> + const uint32_t m_sz = sizeof(struct rte_mbuf);
> uintptr_t sa_base = 0;
>
> u64[0] = (u64[0] & (0x3ull << 32)) << 6 | (u64[0] & (0x3FFull <<
> 36)) << 4 |
> diff --git a/drivers/net/cnxk/cn20k_rx.h b/drivers/net/cnxk/cn20k_rx.h
> index f8fa6de2b9..b544868c03 100644
> --- a/drivers/net/cnxk/cn20k_rx.h
> +++ b/drivers/net/cnxk/cn20k_rx.h
> @@ -702,7 +702,7 @@ cn20k_nix_recv_pkts(void *rx_queue, struct rte_mbuf
> **rx_pkts, uint16_t pkts, co
> uint64_t mbuf_init = rxq->mbuf_initializer;
> const void *lookup_mem = rxq->lookup_mem;
> const uint64_t data_off = rxq->data_off;
> - uint8_t m_sz = sizeof(struct rte_mbuf);
> + const uint32_t m_sz = sizeof(struct rte_mbuf);
> const uint64_t wdata = rxq->wdata;
> const uint32_t qmask = rxq->qmask;
> const uintptr_t desc = rxq->desc;
> @@ -815,7 +815,7 @@ cn20k_nix_flush_recv_pkts(void *rx_queue, struct
> rte_mbuf **rx_pkts, uint16_t pk
> uint64_t mbuf_init = rxq->mbuf_initializer;
> const void *lookup_mem = rxq->lookup_mem;
> const uint64_t data_off = rxq->data_off;
> - uint8_t m_sz = sizeof(struct rte_mbuf);
> + const uint32_t m_sz = sizeof(struct rte_mbuf);
> const uint64_t wdata = rxq->wdata;
> const uint32_t qmask = rxq->qmask;
> const uintptr_t desc = rxq->desc;
> --
> 2.35.6
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/stable/attachments/20260909/2a33ca86/attachment.htm>
More information about the stable
mailing list