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

[Heiner Kallweit]

On 08.08.2019 08:21, Yonglong Liu wrote:
> 
> 
> On 2019/8/8 14:11, Heiner Kallweit wrote:
>> On 08.08.2019 03:15, Yonglong Liu wrote:
>>>
>>>
>>> On 2019/8/8 0:47, Heiner Kallweit wrote:
>>>> On 07.08.2019 15:16, Yonglong Liu wrote:
>>>>> [   27.232781] hns3 0000:bd:00.3 eth7: net open
>>>>> [   27.237303] 8021q: adding VLAN 0 to HW filter on device eth7
>>>>> [   27.242972] IPv6: ADDRCONF(NETDEV_CHANGE): eth7: link becomes ready
>>>>> [   27.244449] hns3 0000:bd:00.3: invalid speed (-1)
>>>>> [   27.253904] hns3 0000:bd:00.3 eth7: failed to adjust link.
>>>>> [   27.259379] RTL8211F Gigabit Ethernet mii-0000:bd:00.3:07: PHY state change UP -> RUNNING
>>>>> [   27.924903] hns3 0000:bd:00.3 eth7: link up
>>>>> [   28.280479] RTL8211F Gigabit Ethernet mii-0000:bd:00.3:07: PHY state change RUNNING -> NOLINK
>>>>> [   29.208452] hns3 0000:bd:00.3 eth7: link down
>>>>> [   32.376745] RTL8211F Gigabit Ethernet mii-0000:bd:00.3:07: PHY state change NOLINK -> RUNNING
>>>>> [   33.208448] hns3 0000:bd:00.3 eth7: link up
>>>>> [   35.253821] hns3 0000:bd:00.3 eth7: net stop
>>>>> [   35.258270] hns3 0000:bd:00.3 eth7: link down
>>>>>
>>>>> When using rtl8211f in polling mode, may get a invalid speed,
>>>>> because of reading a fake link up and autoneg complete status
>>>>> immediately after starting autoneg:
>>>>>
>>>>>         ifconfig-1176  [007] ....    27.232763: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x00 val:0x1040
>>>>>   kworker/u257:1-670   [015] ....    27.232805: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x04 val:0x01e1
>>>>>   kworker/u257:1-670   [015] ....    27.232815: mdio_access: mii-0000:bd:00.3 write phy:0x07 reg:0x04 val:0x05e1
>>>>>   kworker/u257:1-670   [015] ....    27.232869: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x01 val:0x79ad
>>>>>   kworker/u257:1-670   [015] ....    27.232904: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x09 val:0x0200
>>>>>   kworker/u257:1-670   [015] ....    27.232940: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x00 val:0x1040
>>>>>   kworker/u257:1-670   [015] ....    27.232949: mdio_access: mii-0000:bd:00.3 write phy:0x07 reg:0x00 val:0x1240
>>>>>   kworker/u257:1-670   [015] ....    27.233003: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x01 val:0x79ad
>>>>>   kworker/u257:1-670   [015] ....    27.233039: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x0a val:0x3002
>>>>>   kworker/u257:1-670   [015] ....    27.233074: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x09 val:0x0200
>>>>>   kworker/u257:1-670   [015] ....    27.233110: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x05 val:0x0000
>>>>>   kworker/u257:1-670   [000] ....    28.280475: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x01 val:0x7989
>>>>>   kworker/u257:1-670   [000] ....    29.304471: mdio_access: mii-0000:bd:00.3 read  phy:0x07 reg:0x01 val:0x7989
>>>>>
>>>>> According to the datasheet of rtl8211f, to get the real time
>>>>> link status, need to read MII_BMSR twice.
>>>>>
>>>>> This patch add a read_status hook for rtl8211f, and do a fake
>>>>> phy_read before genphy_read_status(), so that can get real link
>>>>> status in genphy_read_status().
>>>>>
>>>>> Signed-off-by: Yonglong Liu <liuyonglong@xxxxxxxxxx>
>>>>> ---
>>>>>  drivers/net/phy/realtek.c | 13 +++++++++++++
>>>>>  1 file changed, 13 insertions(+)
>>>>>
>>>> Is this an accidental resubmit? Because we discussed this in
>>>> https://marc.info/?t=156413509900003&r=1&w=2 and a fix has
>>>> been applied already.
>>>>
>>>> Heiner
>>>>
>>>> .
>>>>
>>>
>>> In https://marc.info/?t=156413509900003&r=1&w=2 , the invalid speed
>>> recurrence rate is almost 100%, and I had test the solution about
>>> 5 times and it works. But yesterday it happen again suddenly, and than
>>> I fount that the recurrence rate reduce to 10%. This time we get 0x79ad
>>> after autoneg started which is not 0x798d from last discussion.
>>>
>>>
>>>
>> OK, I'll have a look.
>> However the approach is wrong. The double read is related to the latching
>> of link-down events. This is done by all PHY's and not specific to RT8211F.
>> Also it's not related to the problem. I assume any sufficient delay would
>> do instead of the read.
>>
>> .
>>
> 
> So you will send a new patch to fix this problem? I am waiting for it,
> and can do a full test this time.
> 
> 
Can you try the following? This delay should give thy PHY enough time
to clear both bits before the following read is done.

diff --git a/drivers/net/phy/phy.c b/drivers/net/phy/phy.c
index ef7aa738e..32f327a44 100644
--- a/drivers/net/phy/phy.c
+++ b/drivers/net/phy/phy.c
@@ -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);
+
 	if (phy_is_started(phydev))
 		err = phy_check_link_status(phydev);
 out_unlock:
-- 
2.22.0