On 11/11/2023 3:21 AM, lihuisong (C) wrote:
> 在 2023/11/11 9:59, Jie Hai 写道:
>> From: Dengdui Huang <huangdeng...@huawei.com>
>>
>> Currently, hns3 VF driver uses the following points to match
>> the response and request message for the mailbox synchronous
>> message between VF and PF.
>> 1. req_msg_data which is consist of message code and subcode,
>> is used to match request and response.
>> 2. head means the number of send success for VF.
>> 3. tail means the number of receive success for VF.
>> 4. lost means the number of send timeout for VF.
>> And 'head', 'tail' and 'lost' are dynamically updated during
>> the communication.
>>
>> Now there is a issue that all sync mailbox message will
>> send failure forever at the flollowing case:
>> 1. VF sends the message A
>> then head=UINT32_MAX-1, tail=UINT32_MAX-3, lost=2.
>> 2. VF sends the message B
>> then head=UINT32_MAX, tail=UINT32_MAX-2, lost=2.
>> 3. VF sends the message C, the message will be timeout because
>> it can't get the response within 500ms.
>> then head=0, tail=0, lost=2
>> note: tail is assigned to head if tail > head according to
>> current code logic. From now on, all subsequent sync milbox
>> messages fail to be sent.
>>
>> It's very complicated to use the fields 'lost','tail','head'.
>> The code and subcode of the request sync mailbox are used as the
>> matching code of the message, which is used to match the response
>> message for receiving the synchronization response.
>>
>> This patch drops these fields and uses the following solution
>> to solve this issue:
>> In the handling response message process, using the req_msg_data
>> of the request and response message to judge whether the sync
>> mailbox message has been received.
>>
>> Fixes: 463e748964f5 ("net/hns3: support mailbox")
>> Cc: sta...@dpdk.org
>>
>> Signed-off-by: Dengdui Huang <huangdeng...@huawei.com>
>> Signed-off-by: Jie Hai <haij...@huawei.com>
>>
>
> Acked-by: Huisong Li <lihuis...@huawei.com>
>
Applied to dpdk-next-net/main, thanks.
