I am on vacation until July 12. I will be happy to review these when I get back.
Bob
-----Original Message-----
From: yangx.jy@xxxxxxxxxxx <yangx.jy@xxxxxxxxxxx>
Sent: Friday, July 8, 2022 1:34 AM
To: Jason Gunthorpe <jgg@xxxxxxxx>; Bob Pearson <rpearsonhpe@xxxxxxxxx>
Cc: linux-rdma@xxxxxxxxxxxxxxx; leon@xxxxxxxxxx; zyjzyj2000@xxxxxxxxx
Subject: Re: [PATCH] RDMA/rxe: Process received packets in time
On 2022/7/4 22:23, yangx.jy@xxxxxxxxxxx wrote:
> On 2022/7/4 20:55, Jason Gunthorpe wrote:
>> On Sun, Jul 03, 2022 at 11:56:25PM +0800, Xiao Yang wrote:
>>> If received packets (i.e. skb) stored in qp->resp_pkts cannot be
>>> processed in time, they may be ovewritten/reused and then lead to
>>> abnormal behavior.
>>
>> I just merged a bunch of fixed from Bob on atomics, do they solve
>> this problem too by chance?
> Hi Jason,
>
> Do you mean that I should use RDMA for-next branch to confirm the issue?
>
> I don't think Bob's patches can solve this problem by chance, but I
> will use RDMA for-next branch to test the issue on my slow vm.
Hi Jason,
I confirmed that the following patch avoided this issue on my slow vm.
commit dc1848388137d20e5786b976caa49a26889f36f3
Author: Bob Pearson <rpearsonhpe@xxxxxxxxx>
Date: Mon Jun 6 09:38:37 2022 -0500
RDMA/rxe: Merge normal and retry atomic flows
I think this issue is related to res->atomic.skb which has been removed by the patch. However I am still confused about why res->atomic.skb will be overwritten/reused as my commit message described.
Best Regards,
Xiao Yang
>
> Best Regards,
> Xiao Yang
>>
>> Jason
