On 11/30/2021 7:31 AM, Yong Wang wrote:
-----Original Message-----
From:"sahithi.sin...@oracle.com" <sahithi.sin...@oracle.com>
Date: Monday, November 8, 2021 at 12:23 AM
To: Yong Wang<yongw...@vmware.com>
Cc:"dev@dpdk.org" <dev@dpdk.org>, Sahithi Singam<sahithi.sin...@oracle.com>
Subject: [PATCH] net/vmxnet3: add spinlocks to register command access
From: Sahithi Singam<sahithi.sin...@oracle.com>
At present, there are no spinlocks around register command access.
This resulted in a race condition when two threads running on
two different cores invoked link_update function at the same time
to get link status. Due to this race condition, one of the threads
reported false link status value.
Signed-off-by: Sahithi Singam<sahithi.sin...@oracle.com>
---
Thanks Sahithi for the patch. As we discussed offline, in DPDK, the
expectation is that control level synchronization should be handled by the
application. In my knowledge, currently no PMD guarantee such synchronization
at driver callback level. It makes more sense to have the application manages
the synchronization as most likely it needs to work with multiple PMDs and it's
better to keep this behavior consistent across all PMDs (i.e, it does not make
a lot of sense to support this behavior only in one particular PMD).
ack,
updating patch status as rejected.