Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230325152417.5403-1-kerneljasonxing@gmail.com> <20230326013845.2110-1-hdanton@sina.com>
In-Reply-To: <20230326013845.2110-1-hdanton@sina.com>
From:   Jason Xing <kerneljasonxing@gmail.com>
Date:   Sun, 26 Mar 2023 11:30:45 +0800
Message-ID: <CAL+tcoCO47mgZeKM=h8OZWcebB-hKzyC9FiTgs2cPR25bE18UQ@mail.gmail.com>
Subject: Re: [PATCH net] net: fix raising a softirq on the current cpu with
 rps enabled
To:     Hillf Danton <hdanton@sina.com>
Cc:     pabeni@redhat.com, Eric Dumazet <edumazet@google.com>,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        Jason Xing <kernelxing@tencent.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Sun, Mar 26, 2023 at 9:39=E2=80=AFAM Hillf Danton <hdanton@sina.com> wro=
te:
>
> On Sat, Mar 25, 2023 at 8:26=E2=80=AFAM Jason Xing <kerneljasonxing@gmail=
.com>
> >
> > @@ -4594,8 +4594,6 @@ static int napi_schedule_rps(struct softnet_data =
*sd)
> >         if (sd !=3D3D mysd) {
> >                 sd->rps_ipi_next =3D3D mysd->rps_ipi_list;
> >                 mysd->rps_ipi_list =3D3D sd;
> > -
> > -               __raise_softirq_irqoff(NET_RX_SOFTIRQ);
> >                 return 1;
> >         }
>
> Nope, ipi should be sent. But no ipi can go without irq enabled.
>

Sorry, I didn't get it. IPI is sent in net_rx_action() and apparently
I didn't touch this part in this patch. Here is only about whether we
should raise an IRQ even if the skb will be enqueued into another cpu.

> So feel free to work out what sense made by disabling irq at the call sit=
e
> then directly send ipi instead.