[dpdk-dev] [PATCH v4 2/5] ixgbe: Bug fix: Properly configure Rx CRC stripping for x540 devices

Vlad Zolotarov vladz at cloudius-systems.com
Mon Mar 9 17:10:24 CET 2015



On 03/09/15 09:58, Thomas Monjalon wrote:
> 2015-03-09 09:08, Vlad Zolotarov:
>> On 03/08/15 23:12, Thomas Monjalon wrote:
>>> Hi Vlad,
>>>
>>> 2015-03-08 16:04, Vlad Zolotarov:
>>>> According to x540 spec chapter 8.2.4.8.9 CRCSTRIP field of RDRXCTL should
>>>> be configured to the same value as HLREG0.RXCRCSTRP.
>>>>
>>>> Clearing the RDRXCTL.RSCFRSTSIZE field for x540 is not required by the spec
>>>> but seems harmless.
>>>>
>>>> Signed-off-by: Vlad Zolotarov <vladz at cloudius-systems.com>
>>> You are mixing a fix (this patch) and enhancements (LRO) in the same series.
>>> Could you separate them please, as LRO is not going into 2.0 but this fix
>>> is a good candidate.
>> Pls, note that all patches in this series except for PATCH3 and PATCH5
>> are fixing real bugs. I can send them as a separate series if u'd like.
>> Pls., confirm.
> Yes you're right, patch 1 is also a fix and patch 4 seems to solve other
> issues. However, patch 4 makes also some refactoring and seems a bit risky.
> We need an ixgbe maintainer to decide wether we can merge it before the
> release. Or is it possible to have fixes of the patch 4 without the
> refactoring?
>
> Thanks Vlad.
> Sorry to request such split but this PMD is sensible and I don't want to
> have a risk of making it worst in release 2.0.

Well, IMHO PATCH4 here has the most important and critical fix among 
other patches - the whole port is going to be configured according to 
the last queue configuration. Consider the case when queues are 
configured differently and the last one meets vector Rx requirements: 
bulk allocation requirements, descriptors number is a power of 2 (there 
are per-port requirements too). But the first queue may be configured to 
have a different number of descriptors, which is NOT a power of two. 
Currently the port's rx_pkt_burst callback will be configured to use a 
vercot Rx callback and we would hit the same issue reported by Stephen 
and "fixed" in 352078e8e.




More information about the dev mailing list