linux-kernel - Re: [PATCH net] net: phy: rtl8211f: do a double read to get real time link status

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <7f18113e-268b-6a4a-af83-236cfa337fcd@gmail.com>
Date:   Fri, 9 Aug 2019 22:05:38 +0200
From:   Heiner Kallweit <hkallweit1@...il.com>
To:     Yonglong Liu <liuyonglong@...wei.com>, Andrew Lunn <andrew@...n.ch>
Cc:     davem@...emloft.net, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org, linuxarm@...wei.com,
        salil.mehta@...wei.com, yisen.zhuang@...wei.com,
        shiju.jose@...wei.com
Subject: Re: [PATCH net] net: phy: rtl8211f: do a double read to get real time
 link status

On 09.08.2019 06:57, Yonglong Liu wrote:
> 
> 
> On 2019/8/9 4:34, Andrew Lunn wrote:
>> On Thu, Aug 08, 2019 at 10:01:39PM +0200, Heiner Kallweit wrote:
>>> On 08.08.2019 21:40, Andrew Lunn wrote:
>>>>> @@ -568,6 +568,11 @@ int phy_start_aneg(struct phy_device *phydev)
>>>>>  	if (err < 0)
>>>>>  		goto out_unlock;
>>>>>  
>>>>> +	/* The PHY may not yet have cleared aneg-completed and link-up bit
>>>>> +	 * w/o this delay when the following read is done.
>>>>> +	 */
>>>>> +	usleep_range(1000, 2000);
>>>>> +
>>>>
>>>> Hi Heiner
>>>>
>>>> Does 802.3 C22 say anything about this?
>>>>
>>> C22 says:
>>> "The Auto-Negotiation process shall be restarted by setting bit 0.9 to a logic one. This bit is self-
>>> clearing, and a PHY shall return a value of one in bit 0.9 until the Auto-Negotiation process has been
>>> initiated."
>>>
>>> Maybe we should read bit 0.9 in genphy_update_link() after having read BMSR and report
>>> aneg-complete and link-up as false (no matter of their current value) if 0.9 is set.
>>
>> Yes. That sounds sensible.
>>
>>      Andrew
>>
>> .
>>
> 
> Hi Heiner:
> 	I have test more than 50 times, it works. Previously less
> than 20 times must be recurrence. so I think this patch solved the
> problem.
> 	And I checked about 40 times of the time gap between read
> and autoneg started, all of them is more than 2ms, as below:
> 
>   kworker/u257:1-670   [015] ....    27.182632: mdio_access: mii-0000:bd:00.3 write phy:0x07 reg:0x00 val:0x1240
>   kworker/u257:1-670   [015] ....    27.184670: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x01 val:0x7989
> 
> 

Instead of using this fixed delay, the following experimental patch
considers that fact that between triggering aneg start and actual
start of aneg (incl. clearing aneg-complete bit) Clause 22 requires
a PHY to keep bit 0.9 (aneg restart) set.
Could you please test this instead of the fixed-delay patch?

Thanks, Heiner

diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c
index b039632de..163295dbc 100644
--- a/drivers/net/phy/phy_device.c
+++ b/drivers/net/phy/phy_device.c
@@ -1741,7 +1741,17 @@ EXPORT_SYMBOL(genphy_aneg_done);
  */
 int genphy_update_link(struct phy_device *phydev)
 {
-	int status;
+	int status = 0, bmcr;
+
+	bmcr = phy_read(phydev, MII_BMCR);
+	if (bmcr < 0)
+		return bmcr;
+
+	/* Autoneg is being started, therefore disregard BMSR value and
+	 * report link as down.
+	 */
+	if (bmcr & BMCR_ANRESTART)
+		goto done;
 
 	/* The link state is latched low so that momentary link
 	 * drops can be detected. Do not double-read the status
-- 
2.22.0