On 01/15/2018 09:13 PM, David Miller wrote:
> From: Daniel Borkmann via iovisor-dev <[email protected]>
> Date: Mon, 15 Jan 2018 21:11:06 +0100
>
>> On 01/15/2018 06:24 PM, Daniel Borkmann via iovisor-dev wrote:
>>> On 01/11/2018 07:05 AM, Sandipan Das via iovisor-dev wrote:
>>> [...]
>>>> Any ideas about why this is happening? Also, it would be really helpful
>>>> if someone can translate the Pyroute2 calls in the test script to the
>>>> corresponding tc commands.
>>>
>>> I can trigger it as well, so I should have something very soon; currently
>>> looking into it.
>>
>> Here's the fix for your panic ... cooking up proper patch with commit
>> description for netdev now:
>>
>> From 126dd0c1ecb7c207f79976c6c3ef13be08afe6b5 Mon Sep 17 00:00:00 2001
>> From: Daniel Borkmann <[email protected]>
>> Date: Mon, 15 Jan 2018 19:48:28 +0000
>> Subject: [PATCH net] net, sched: fix miniq {b,q}stats handling
>>
>> Signed-off-by: Daniel Borkmann <[email protected]>
>
> Is this related to:
>
> http://patchwork.ozlabs.org/patch/860647/
>
> ?
Ahh, yes, I didn't see that one. The fix is needed in net tree however, since
it's broken there already (v4.15-rc1 onwards). I was using an arm64 machine
with simple clsact and some filter attached to it, so as soon as this hits
traffic, we get the panic since {b,q}stats are allocated /after/ setting them
up in miniq init.
We could use some form of d59f5ffa59d8 ("net: sched: a dflt qdisc may be
used with per cpu stats") for net and get rid of the per-cpu alloc from
qdisc_create(), since it's not needed there anymore.
_______________________________________________
iovisor-dev mailing list
[email protected]
https://lists.iovisor.org/mailman/listinfo/iovisor-dev
