[PATCH net] net: mvneta: clear XDP pfmemalloc flag between frames

[PATCH net] net: mvneta: clear XDP pfmemalloc flag between frames

[Lorenzo Bianconi]

In mvneta_swbm_add_rx_fragment() the XDP pfmemalloc flag is set on the
xdp_buff if the page under pressure is a pfmemalloc one. The xdp_buff is
reused for the next packet and the flag is never cleared, so a pfmemalloc
fragment in one packet wrongly marks the skb of a subsequent packet as
pfmemalloc. Clear the flag in mvneta_swbm_rx_frame(), which is invoked for
each new frame.

Fixes: ed7a58cb40bd ("net: marvell: rely on xdp_update_skb_shared_info utility routine")
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>
---
 drivers/net/ethernet/marvell/mvneta.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
index 543e566425c1..32416999f3d6 100644
--- a/drivers/net/ethernet/marvell/mvneta.c
+++ b/drivers/net/ethernet/marvell/mvneta.c
@@ -2341,6 +2341,7 @@ mvneta_swbm_rx_frame(struct mvneta_port *pp,
 	/* Prefetch header */
 	prefetch(data);
 	xdp_buff_clear_frags_flag(xdp);
+	xdp_buff_clear_frag_pfmemalloc(xdp);
 	xdp_prepare_buff(xdp, data, pp->rx_offset_correction + MVNETA_MH_SIZE,
 			 data_len, true);
 }

---
base-commit: e3bfd25626b44b6fa61a13c17178922171d519ce
change-id: 20260919-mvneta-xdp-clear-frag-fix-7ac691e1659b

Best regards,
-- 
Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>

Re: [PATCH net] net: mvneta: clear XDP pfmemalloc flag between frames

[Simon Horman]

On Sun, Sep 20, 2026 at 09:55:09AM +0200, Lorenzo Bianconi wrote:
> In mvneta_swbm_add_rx_fragment() the XDP pfmemalloc flag is set on the
> xdp_buff if the page under pressure is a pfmemalloc one. The xdp_buff is
> reused for the next packet and the flag is never cleared, so a pfmemalloc
> fragment in one packet wrongly marks the skb of a subsequent packet as
> pfmemalloc. Clear the flag in mvneta_swbm_rx_frame(), which is invoked for
> each new frame.
> 
> Fixes: ed7a58cb40bd ("net: marvell: rely on xdp_update_skb_shared_info utility routine")
> Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>

Reviewed-by: Simon Horman <horms@kernel.org>

Re: [PATCH net] net: mvneta: clear XDP pfmemalloc flag between frames

[Jakub Kicinski]

On Sun, 20 Sep 2026 09:55:09 +0200 Lorenzo Bianconi wrote:
> In mvneta_swbm_add_rx_fragment() the XDP pfmemalloc flag is set on the
> xdp_buff if the page under pressure is a pfmemalloc one. The xdp_buff is
> reused for the next packet and the flag is never cleared, so a pfmemalloc
> fragment in one packet wrongly marks the skb of a subsequent packet as
> pfmemalloc. Clear the flag in mvneta_swbm_rx_frame(), which is invoked for
> each new frame.
> 
> Fixes: ed7a58cb40bd ("net: marvell: rely on xdp_update_skb_shared_info utility routine")
> Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>
> ---
>  drivers/net/ethernet/marvell/mvneta.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> index 543e566425c1..32416999f3d6 100644
> --- a/drivers/net/ethernet/marvell/mvneta.c
> +++ b/drivers/net/ethernet/marvell/mvneta.c
> @@ -2341,6 +2341,7 @@ mvneta_swbm_rx_frame(struct mvneta_port *pp,
>  	/* Prefetch header */
>  	prefetch(data);
>  	xdp_buff_clear_frags_flag(xdp);
> +	xdp_buff_clear_frag_pfmemalloc(xdp);

bit of a trap for your future self to clear the flags one by one.
Should we not write the flags to 0 ? Or maybe add:

/* Clear xdp_buf state for reuse on the same ring */
static __always_inline void xdp_reinit_buff(struct xdp_buff *xdp)
{
	xdp->flags = 0;
}

and call that?

>  	xdp_prepare_buff(xdp, data, pp->rx_offset_correction + MVNETA_MH_SIZE,
>  			 data_len, true);
>  }

Re: [PATCH net] net: mvneta: clear XDP pfmemalloc flag between frames

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 1683 bytes --]

> On Sun, 20 Sep 2026 09:55:09 +0200 Lorenzo Bianconi wrote:
> > In mvneta_swbm_add_rx_fragment() the XDP pfmemalloc flag is set on the
> > xdp_buff if the page under pressure is a pfmemalloc one. The xdp_buff is
> > reused for the next packet and the flag is never cleared, so a pfmemalloc
> > fragment in one packet wrongly marks the skb of a subsequent packet as
> > pfmemalloc. Clear the flag in mvneta_swbm_rx_frame(), which is invoked for
> > each new frame.
> > 
> > Fixes: ed7a58cb40bd ("net: marvell: rely on xdp_update_skb_shared_info utility routine")
> > Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>
> > ---
> >  drivers/net/ethernet/marvell/mvneta.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> > index 543e566425c1..32416999f3d6 100644
> > --- a/drivers/net/ethernet/marvell/mvneta.c
> > +++ b/drivers/net/ethernet/marvell/mvneta.c
> > @@ -2341,6 +2341,7 @@ mvneta_swbm_rx_frame(struct mvneta_port *pp,
> >  	/* Prefetch header */
> >  	prefetch(data);
> >  	xdp_buff_clear_frags_flag(xdp);
> > +	xdp_buff_clear_frag_pfmemalloc(xdp);
> 
> bit of a trap for your future self to clear the flags one by one.
> Should we not write the flags to 0 ? Or maybe add:
> 
> /* Clear xdp_buf state for reuse on the same ring */
> static __always_inline void xdp_reinit_buff(struct xdp_buff *xdp)
> {
> 	xdp->flags = 0;
> }
> 
> and call that?

ack, I agree. I will fix it in v2.

Regards,
Lorenzo

> 
> >  	xdp_prepare_buff(xdp, data, pp->rx_offset_correction + MVNETA_MH_SIZE,
> >  			 data_len, true);
> >  }
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]