Re: [EXTERNAL] Re: [v6 1/5] vhost: skip crypto op fetch before vring init

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi Gowri,

On 2/28/25 9:48 AM, David Marchand wrote:
> On Thu, Feb 27, 2025 at 7:07 PM Gowrishankar Muthukrishnan
> <gmuthukrishn@marvell.com> wrote:
>>>> Ha, and also you should be able to remove:
>>>> __rte_no_thread_safety_analysis /* FIXME: requires iotlb_lock? */ in
>>>> vhost_crypto_process_one_req() once implemented.
>>>>
>>>
>> Removing it would break compilation for thread safety flag.
>> http://mails.dpdk.org/archives/test-report/2025-February/857515.html
>>
>> It is due to local vc_req that is passed to func that requires iotlb lock
>> In vc_req->vq. Even though vc_req->vq is locked vq, GCC does not allow it, as I understand.
> 
> *cough* clang.
> 
>>
>>          vc_req = &data_req;
>>          vc_req->desc_idx = desc_idx;
>>          vc_req->dev = vcrypto->dev;
>>          vc_req->vq = vq;
> 
> The annotations won't handle this wrapping in the vc_req object.
> Just pass a vq object rather than the vc_req (which I don't see little
> point in having in the first place..) and adjust annotations.
> 
> 

Before your series arrived, we were wondering if we should not deprecate
Vhost crypto as it was not really maintained and we had no identified
user.

Since it seems you are going to use it, which is great, would you commit
to make the necessary changes to make it reliable? If yes, I would agree
to take your v8 as is for v25.03 if proper rework is done for v25.07,
would that work for you?

Maxime