[PATCH v2] net/memif: Fix memif-rx buffer linking

[PATCH v2] net/memif: Fix memif-rx buffer linking

[Mike Bly]

[-- Attachment #1: Type: text/plain, Size: 904 bytes --]

Moving where dst_off is set in memif-rx function. It should be
initialized for each new dpdk-mbuf, not each memif-buffer.
Bugzilla ID: 1609
Fixes: aa17df860891 ("net/memif: add a Rx fast path")
Cc: stable@dpdk.org
Cc: Jakub Grajciar <jgrajcia@cisco.com>

Signed-off-by: Mike Bly <bly454@gmail.com>
---
 drivers/net/memif/rte_eth_memif.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/net/memif/rte_eth_memif.c
b/drivers/net/memif/rte_eth_memif.c
index c0bf1e3bbf..bef1606550 100644
--- a/drivers/net/memif/rte_eth_memif.c
+++ b/drivers/net/memif/rte_eth_memif.c
@@ -414,13 +414,12 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs,
uint16_t nb_pkts)
  goto no_free_bufs;
  mbuf = mbuf_head;
  mbuf->port = mq->in_port;
-
+ dst_off = 0;
 next_slot2:
  s0 = cur_slot & mask;
  d0 = &ring->desc[s0];

  src_len = d0->length;
- dst_off = 0;
  src_off = 0;

  do {
-- 
2.43.5

[-- Attachment #2: Type: text/html, Size: 1215 bytes --]

Re: [PATCH v2] net/memif: Fix memif-rx buffer linking

[Stephen Hemminger]

On Tue, 7 Jan 2025 10:22:10 -0800
Mike Bly <bly454@gmail.com> wrote:

> Cc: stable@dpdk.org, jgrajcia@cisco.com
> Subject: [PATCH v2] net/memif: Fix memif-rx buffer linking
> Date: Tue, 7 Jan 2025 10:22:10 -0800
> 
> Moving where dst_off is set in memif-rx function. It should be
> initialized for each new dpdk-mbuf, not each memif-buffer.
> Bugzilla ID: 1609
> Fixes: aa17df860891 ("net/memif: add a Rx fast path")
> Cc: stable@dpdk.org
> Cc: Jakub Grajciar <jgrajcia@cisco.com>
> 
> Signed-off-by: Mike Bly <bly454@gmail.com>
> ---
>  drivers/net/memif/rte_eth_memif.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/drivers/net/memif/rte_eth_memif.c
> b/drivers/net/memif/rte_eth_memif.c
> index c0bf1e3bbf..bef1606550 100644
> --- a/drivers/net/memif/rte_eth_memif.c
> +++ b/drivers/net/memif/rte_eth_memif.c
> @@ -414,13 +414,12 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
>   goto no_free_bufs;
>   mbuf = mbuf_head;
>   mbuf->port = mq->in_port;
> -
> + dst_off = 0;
>  next_slot2:
>   s0 = cur_slot & mask;
>   d0 = &ring->desc[s0];

Looks like mailer whitespace corruption again.

Re: [PATCH v2] net/memif: Fix memif-rx buffer linking

[Sriram Yagnaraman]

Hi Mike,

Do you know the status of this patch? It looks like it hasn't received any
review since it was posted in January 2025.

We've independently hit the same bug (dst_off reset per-descriptor
instead of per-packet in the RX slow path) and your fix solves our
issue.

If this patch is stale and unlikely to be picked up, I can send a
new patchset against the latest tree that also includes a 
boundary check for MEMIF_DESC_FLAG_NEXT when n_slots is
exhausted.

Let me know if you'd prefer to respin your series or if I should send
a new one.

Thanks,
Sriram

Re: [PATCH v2] net/memif: Fix memif-rx buffer linking

[Mike Bly]

[-- Attachment #1: Type: text/plain, Size: 885 bytes --]

Sriram,

I am glad the patch worked for you. Unfortunately, I am no longer working
in this space so you will need to take the lead on getting this pulled in.

-MikeB

On Mon, Mar 16, 2026, 1:50 AM Sriram Yagnaraman <
sriram.yagnaraman@ericsson.com> wrote:

> Hi Mike,
>
> Do you know the status of this patch? It looks like it hasn't received any
> review since it was posted in January 2025.
>
> We've independently hit the same bug (dst_off reset per-descriptor
> instead of per-packet in the RX slow path) and your fix solves our
> issue.
>
> If this patch is stale and unlikely to be picked up, I can send a
> new patchset against the latest tree that also includes a
> boundary check for MEMIF_DESC_FLAG_NEXT when n_slots is
> exhausted.
>
> Let me know if you'd prefer to respin your series or if I should send
> a new one.
>
> Thanks,
> Sriram
>

[-- Attachment #2: Type: text/html, Size: 1308 bytes --]