Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
Message-ID: <97732cc0a75c0be3f2354075085e7fa6d78e82bb.camel@redhat.com>
Subject: Re: [PATCH net 6/7] net: hns3: fix VF reset fail issue
From:   Paolo Abeni <pabeni@redhat.com>
To:     Jijie Shao <shaojijie@huawei.com>, yisen.zhuang@huawei.com,
        salil.mehta@huawei.com, davem@davemloft.net, edumazet@google.com,
        kuba@kernel.org
Cc:     shenjian15@huawei.com, wangjie125@huawei.com,
        liuyonglong@huawei.com, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Date:   Thu, 02 Nov 2023 17:24:14 +0100
In-Reply-To: <c87cfcbc-8cd6-4a01-bac0-74113f7ca904@huawei.com>
References: <20231028025917.314305-1-shaojijie@huawei.com>
         <20231028025917.314305-7-shaojijie@huawei.com>
         <9bc9514044063bc57155fb786f970ca1d69758b4.camel@redhat.com>
         <c87cfcbc-8cd6-4a01-bac0-74113f7ca904@huawei.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Thu, 2023-11-02 at 20:16 +0800, Jijie Shao wrote:
> on 2023/11/2 18:45, Paolo Abeni wrote:
> > On Sat, 2023-10-28 at 10:59 +0800, Jijie Shao wrote:
> > >  =20
> > > -static void hclgevf_clear_event_cause(struct hclgevf_dev *hdev, u32 =
regclr)
> > > +static void hclgevf_clear_event_cause(struct hclgevf_dev *hdev, u32 =
regclr,
> > > +				      bool need_dalay)
> > >   {
> > > +#define HCLGEVF_RESET_DELAY		5
> > > +
> > > +	if (need_dalay)
> > > +		mdelay(HCLGEVF_RESET_DELAY);
> > 5ms delay in an interrupt handler is quite a lot. What about scheduling
> > a timer from the IH to clear the register when such delay is needed?
> >=20
> > Thanks!
> >=20
> > Paolo
>=20
> Using timer in this case will complicate the code and make maintenance di=
fficult.

Why?=20

Would something alike the following be ok? (plus reset_timer
initialization at vf creation and cleanup at vf removal time):

---
diff --git a/drivers/net/ethernet/hisilicon/hns3/hns3vf/hclgevf_main.c b/dr=
ivers/net/ethernet/hisilicon/hns3/hns3vf/hclgevf_main.c
index a4d68fb216fb..626bc67065fc 100644
--- a/drivers/net/ethernet/hisilicon/hns3/hns3vf/hclgevf_main.c
+++ b/drivers/net/ethernet/hisilicon/hns3/hns3vf/hclgevf_main.c
@@ -1974,6 +1974,14 @@ static enum hclgevf_evt_cause hclgevf_check_evt_caus=
e(struct hclgevf_dev *hdev,
 	return HCLGEVF_VECTOR0_EVENT_OTHER;
 }
=20
+static void hclgevf_reset_timer(struct timer_list *t)
+{
+	struct hclgevf_dev *hdev =3D from_timer(hclgevf_dev, t, reset_timer);
+
+	hclgevf_clear_event_cause(hdev, HCLGEVF_VECTOR0_EVENT_RST);
+	hclgevf_reset_task_schedule(hdev);
+}
+
 static irqreturn_t hclgevf_misc_irq_handle(int irq, void *data)
 {
 	enum hclgevf_evt_cause event_cause;
@@ -1982,13 +1990,13 @@ static irqreturn_t hclgevf_misc_irq_handle(int irq,=
 void *data)
=20
 	hclgevf_enable_vector(&hdev->misc_vector, false);
 	event_cause =3D hclgevf_check_evt_cause(hdev, &clearval);
+	if (event_cause =3D=3D HCLGEVF_VECTOR0_EVENT_RST)
+		mod_timer(hdev->reset_timer, jiffies + msecs_to_jiffies(5));
+
 	if (event_cause !=3D HCLGEVF_VECTOR0_EVENT_OTHER)
 		hclgevf_clear_event_cause(hdev, clearval);
=20
 	switch (event_cause) {
-	case HCLGEVF_VECTOR0_EVENT_RST:
-		hclgevf_reset_task_schedule(hdev);
-		break;
 	case HCLGEVF_VECTOR0_EVENT_MBX:
 		hclgevf_mbx_handler(hdev);
 		break;
---

> We consider reducing the delay time by polling. For example,
> the code cycles every 50 us to check whether the write register takes eff=
ect.
> If yes, the function returns immediately. or the code cycles until 5 ms.
>=20
> Is this method appropriate?

IMHO such solution will not remove the problem. How frequent is
expected to be the irq generating such delay?

Thanks

Paolo