On Wed, Dec 11, 2024 at 9:26 PM Jason Gunthorpe <jgg@xxxxxxxx> wrote: > > On Wed, Dec 11, 2024 at 11:33:38AM +0530, Selvin Xavier wrote: > > > ib_unregister will trigger the destroy_qp (at least QP1) and > > destroy_cqs. Those verbs will be trying to issue the FW command and > > we are trying to prevent sending it to FW here. We need a check > > here in the common path to avoid sending any commands and i dont see > > a way to handle this otherwise. Current check has a bug where the > > return code check was not correct and we ended up sending some of > > the commands that eventually timeout. > > > > Just to add, We have similar implementation in our L2 driver also, > > which we were trying to replicate using > > bnxt_re data structures. > > Doesn't that suggest this is layered wrong? Shouldn't these tests be > inside the shared command submission code? There are two separate communication channel for this adapter. One owned by L2 driver and another owned by RoCE driver. The RoCE driver checks are for the second communication channel. L2 doesn't use this channel. > > Jason
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature
