Hi,
On 28/08/2020 13:52, Carasec Elena wrote:
> Hello,
>
> We admit that the mistake was from our part.
> New tests' results have revealed that from DPDK v20.05 this issue
> does not exist anymore. So, this patch is applicable for DPDK
> v20.02 and older.
>
The "Fixed:" commit is first in 18.05, so looks like this is valid for
18.11 and 19.11 LTS branches. Please confirm, and it will need i40e
maintainers ack too.
thanks,
Kevin.
> Kind regards,
> Elena Carasec
>
>
>
> -------- Original Message --------
> Subject: RE: [dpdk-dev] [PATCH] i40e: fix segfault when using custom RSS
> key
> Date: 2020-08-28 08:38
> From: "Xing, Beilei" <beilei.x...@intel.com>
> To: Elena Carasec <xcara...@stud.fit.vutbr.cz>, "dev@dpdk.org"
> <dev@dpdk.org>
>
>> -----Original Message-----
>> From: dev <dev-boun...@dpdk.org> On Behalf Of Elena Carasec
>> Sent: Wednesday, August 26, 2020 7:28 PM
>> To: dev@dpdk.org
>> Cc: Elena Carasec <xcara...@stud.fit.vutbr.cz>; sta...@dpdk.org; Xing,
>> Beilei
>> <beilei.x...@intel.com>; Zhang, Qi Z <qi.z.zh...@intel.com>; Adrien
>> Mazarguil <adrien.mazarg...@6wind.com>; Andrew Rybchenko
>> <arybche...@solarflare.com>; Luca Boccassi <bl...@debian.org>; Jan
>> Viktorin <vikto...@cesnet.cz>
>> Subject: [dpdk-dev] [PATCH] i40e: fix segfault when using custom RSS
>> key
>>
>> &out->conf and in can point to the same memory area. Reinitialization
>> of
>> out->conf leads to setting in->key to NULL, but leaves key_len 40. This
>> leads to segfault on destruction of the RSS flow action. The segfault
>> happens
>> inside i40e_action_rss_same(), when comparing comp->key and
>> with->key, because both comp->key_len and with->key_len are 40 (should
>> be 0).
>>
>> Reproduction steps (testpmd):
>>
>> port stop 0
>> flow create 0 ingress pattern end actions rss func default level 0\
>> key
>> 6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5a6d5
>> a6d5a6d5a6d5a6d5a\
>> key_len 40 queues 0 end / end
>> port start 0
>> set link-up port 0
>> start
>> stop
>> set link-down port 0
>> port stop 0
>> flow destroy 0 rule 0
>> (Segmentation fault)
>
> Hi,
>
> Thanks for the fix.
> But according to Chenxu and Shougang's test, this issue doesn't exist
> after DPDK 20.05,
> where RSS configuration has been refactored.
> Could you please retry with the latest DPDK?
>
> BR,
> Beilei
>
>>
>> Fixes: ac8d22de2394 ("ethdev: flatten RSS configuration in flow API")
>>
>> Signed-off-by: Elena Carasec <xcara...@stud.fit.vutbr.cz>
>> Signed-off-by: Jan Viktorin <vikto...@cesnet.cz>
>> ---
>> drivers/net/i40e/i40e_ethdev.c | 2 ++
>> 1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/net/i40e/i40e_ethdev.c
>> b/drivers/net/i40e/i40e_ethdev.c
>> index 11c02b1..a5fe130 100644
>> --- a/drivers/net/i40e/i40e_ethdev.c
>> +++ b/drivers/net/i40e/i40e_ethdev.c
>> @@ -13211,6 +13211,8 @@ struct i40e_customized_pctype*
>> return -EINVAL;
>> if (!in->key && in->key_len)
>> return -EINVAL;
>> + if (&out->conf == in)
>> + return 0;
>> out->conf = (struct rte_flow_action_rss){
>> .func = in->func,
>> .level = in->level,
>> --
>> 1.8.3.1
>
