Re: net/tcp: warning in tcp_try_coalesce/skb_try_coalesce

[Dmitry Vyukov]

On Thu, Feb 2, 2017 at 3:03 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> On Thu, Feb 2, 2017 at 2:56 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>> On Thu, Feb 2, 2017 at 2:40 PM, Eric Dumazet <eric.dumazet@xxxxxxxxx> wrote:
>>> On Thu, 2017-02-02 at 13:59 +0100, Dmitry Vyukov wrote:
>>>
>>>>
>>>> Right, I can confirm that this is not happening on net-next tip
>>>> (62e13097c46c69dbd7544ab2cd585ccf48f360a4).
>>>> I reproduced it on a random month old commit on net-next, though
>>>> (23a8ed4a624324dc696c328f09bd502c4a3816f0). Let's consider that this
>>>> is accidentally fixed by something in that range.
>>>
>>> This also might depend on the .config
>>>
>>> I tried 23a8ed4a624324dc696c328f09bd502c4a3816f0 and could not trigger
>>> the warning .
>>
>>
>> I've attached my config.
>> But I've tried with the same config on both revisions.
>>
>> Well, there is also a chance that the bug is not fixed, but requires
>> slightly different reproducer after some changes.
>
> Hummm....
>
> My config includes KASAN+KCOV and I use gcc version 7.0.0 20161208
> (experimental) (GCC).
> I've now tried to turn off KASAN+KCOV and compiled with my stock
> 4.8.4, and this did _not_ trigger the warning on
> 23a8ed4a624324dc696c328f09bd502c4a3816f0.

Wait, I managed to reproduce it on
23a8ed4a624324dc696c328f09bd502c4a3816f0 without KASAN/KCOV with
4.8.4. It just required more time (like several minutes).

So I would expect that you should be able to reproduce it as well.
I've attached my new config.

Attachment: .config
Description: Binary data