[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CACKFLin+1whPs0qeM5xBb1yXx8FkFS_vGrW6PaGy41_XVH=SGg@mail.gmail.com>
Date: Fri, 8 Sep 2023 10:57:13 -0700
From: Michael Chan <michael.chan@...adcom.com>
To: Pavan Chebbi <pavan.chebbi@...adcom.com>
Cc: Sebastian Andrzej Siewior <bigeasy@...utronix.de>, netdev@...r.kernel.org, bpf@...r.kernel.org,
"David S. Miller" <davem@...emloft.net>, Alexei Starovoitov <ast@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>, Eric Dumazet <edumazet@...gle.com>,
Jakub Kicinski <kuba@...nel.org>, Jesper Dangaard Brouer <hawk@...nel.org>,
John Fastabend <john.fastabend@...il.com>, Paolo Abeni <pabeni@...hat.com>,
Thomas Gleixner <tglx@...utronix.de>, Andrew Gospodarek <gospo@...adcom.com>
Subject: Re: [PATCH net 2/4] bnxt_en: Flush XDP for bnxt_poll_nitroa0()'s NAPI
On Fri, Sep 8, 2023 at 9:30 AM Pavan Chebbi <pavan.chebbi@...adcom.com> wrote:
>
> On Fri, Sep 8, 2023 at 7:29 PM Sebastian Andrzej Siewior
> <bigeasy@...utronix.de> wrote:
> >
> > bnxt_poll_nitroa0() invokes bnxt_rx_pkt() which can run a XDP program
> > which in turn can return XDP_REDIRECT. bnxt_rx_pkt() is also used by
> > __bnxt_poll_work() which flushes (xdp_do_flush()) the packets after each
> > round. bnxt_poll_nitroa0() lacks this feature.
> > xdp_do_flush() should be invoked before leaving the NAPI callback.
> >
> > Invoke xdp_do_flush() after a redirect in bnxt_poll_nitroa0() NAPI.
> >
> > Cc: Michael Chan <michael.chan@...adcom.com>
> > Fixes: f18c2b77b2e4e ("bnxt_en: optimized XDP_REDIRECT support")
> > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
> > ---
> > drivers/net/ethernet/broadcom/bnxt/bnxt.c | 5 +++++
> > 1 file changed, 5 insertions(+)
> >
> > diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> > index 5cc0dbe121327..7551aa8068f8f 100644
> > --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> > +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> > @@ -2614,6 +2614,7 @@ static int bnxt_poll_nitroa0(struct napi_struct *napi, int budget)
> > struct rx_cmp_ext *rxcmp1;
> > u32 cp_cons, tmp_raw_cons;
> > u32 raw_cons = cpr->cp_raw_cons;
> > + bool flush_xdp = false;
>
> Michael can confirm but I don't think we need this additional variable.
> Since the event is always ORed, we could directly check if (event &
> BNXT_REDIRECT_EVENT) just like is done in __bnxt_poll_work().
If we have a mix of XDP_TX and XDP_REDIRECT during NAPI, event can be
cleared by XDP_TX. So this patch looks correct to me because of that.
Or we can make it consistent with __bnxt_poll_work() and assume that
XDP_TX won't mix with XDP_REDIRECT.
Handling a mix of XDP actions needs to be looked at separately. The
driver currently won't work well when that happens. I am working on
an internal patch to address that and will post it when it's ready.
Thanks.
>
> > u32 rx_pkts = 0;
> > u8 event = 0;
> >
> > @@ -2648,6 +2649,8 @@ static int bnxt_poll_nitroa0(struct napi_struct *napi, int budget)
> > rx_pkts++;
> > else if (rc == -EBUSY) /* partial completion */
> > break;
> > + if (event & BNXT_REDIRECT_EVENT)
> > + flush_xdp = true;
> > } else if (unlikely(TX_CMP_TYPE(txcmp) ==
> > CMPL_BASE_TYPE_HWRM_DONE)) {
> > bnxt_hwrm_handler(bp, txcmp);
> > @@ -2667,6 +2670,8 @@ static int bnxt_poll_nitroa0(struct napi_struct *napi, int budget)
> >
> > if (event & BNXT_AGG_EVENT)
> > bnxt_db_write(bp, &rxr->rx_agg_db, rxr->rx_agg_prod);
> > + if (flush_xdp)
> > + xdp_do_flush();
> >
> > if (!bnxt_has_work(bp, cpr) && rx_pkts < budget) {
> > napi_complete_done(napi, rx_pkts);
> > --
> > 2.40.1
> >
> >
Download attachment "smime.p7s" of type "application/pkcs7-signature" (4209 bytes)
Powered by blists - more mailing lists