Re: R8169: Network lockups in 4.18.{8,9,10} (and 4.19 dev)

From: Maciej S. Szmigiero
Date: Thu Oct 11 2018 - 08:24:04 EST


On 11.10.2018 10:24, Chris Clayton wrote:
> On 11/10/2018 01:12, Maciej S. Szmigiero wrote:
>> On 11.10.2018 00:49, Chris Clayton wrote:
>>>> Now, knowing the "right" value you can experiment with what rtl_init_rxcfg()
>>>> writes (under the "default:" label for your NIC model).
>>>>
>>>
>>> This might be more interesting. Through a combination of viewing the output from pr_notice() and the output from
>>> "ethtool -d", I can see RxConfig with the following values
>>>
>>> During boot: 0x00028700
>>> Before suspend: 0x0002870e
>>> During resume: 0x00024000
>>> Post resume: 0x0002870e
>>>
>>> As I did with 4.18.10 early on in the process, I removed the call to rtl_init_rxcfg() from rtl_hw_start() and rebuilt,
>>> installed and rebooted. Now I see the following values:
>>>
>>> During boot: 0x00028700
>>> Before suspend: 0x0002870e
>>> During resume: 0x00024000
>>> Post resume: 0x0002400e
>>>
>>
>> Now we can finally see some difference...
>> Besides missing RX128_INT_EN (bit 15 or 0x8000) and RX_DMA_BURST
>> (bits 8-10 or 0x700) - that rtl_init_rxcfg() would normally set so this
>> is kind of expected - one can see that the working configuration
>> post-resume has bit 14 (or 0x4000) set, too.
>>
>> This bit is described in the driver as RX_MULTI_EN ("8111c only") and is
>> set by rtl_init_rxcfg() for example for RTL_GIGA_MAC_VER_35.
>>
>> RTL_GIGA_MAC_VER_35 is described in the driver as being in the same
>> family as your RTL_GIGA_MAC_VER_38, so can you please try the following
>> change:
>> --- r8169.c
>> +++ r8169.c
>> @@ -4271,6 +4271,7 @@ static void rtl_init_rxcfg(struct rtl816
>> case RTL_GIGA_MAC_VER_18 ... RTL_GIGA_MAC_VER_24:
>> case RTL_GIGA_MAC_VER_34:
>> case RTL_GIGA_MAC_VER_35:
>> + case RTL_GIGA_MAC_VER_38:
>> RTL_W32(tp, RxConfig, RX128_INT_EN | RX_MULTI_EN | RX_DMA_BURST);
>> break;
>> case RTL_GIGA_MAC_VER_40 ... RTL_GIGA_MAC_VER_51:
>>
>> This will add RX_MULTI_EN also for your chip model (you need to add back
>> the call to rtl_init_rxcfg() to rtl_hw_start(), naturally).
>>
>
> That's done the trick. With the above change applied, my network runs running fine after a suspend/resume cycle and the
> ping times are back in the 14-15ms range.

Nice!

I will submit a patch, it would be great if you could test it and then
add a "Tested-by:" tag.

> Chris

Maciej