[Openib-windows] post send request in invalid QP state

[Yossi Leybovich]

Fab

The same behavior is with invalid opcode , the HW will create CQE with
the required syndrome to the user.

Can we remove the IB_INVALID_WR_TYPE from the return values of
ib_post_send/recv.

10x
Yossi  

> -----Original Message-----
> From: ftillier.sst at gmail.com [mailto:ftillier.sst at gmail.com] 
> On Behalf Of Fabian Tillier
> Sent: Thursday, March 09, 2006 8:12 PM
> To: Yossi Leybovich
> Cc: openib-windows at openib.org; Ami Perlmutter
> Subject: Re: [Openib-windows] post send request in invalid QP state
> 
> Hi Yossi,
> 
> On 3/9/06, Yossi Leybovich <sleybo at mellanox.co.il> wrote:
> >
> > Fab
> >
> > In the documentation of IBAL its states , that if one post WQ to QP 
> > that is not in valid state (RTR)  it should get IB_INVALID_QP_STATE.
> > If you want to check qp state you need to issue query_qp 
> and still you  
> > have race with the wire ( like RTR that move to RTS or RTS 
> that move 
> > to ERR) The only state that may generate this error is 
> RESET which is 
> > not creating CQE, all other state return CQE with error so 
> let the user check the CQE.
> >
> > Our new implementation of mthca skip the QP state checkin ( 
> like gen2) 
> > and relay on the CQE.
> > Can we change the documentation of the verb and remove this 
> return value ?
> > or state that its for RESET state?
> 
> Can we track the reset state in SW, without having to do the 
> query_qp call?  If so, we should keep the error for the reset 
> case only. 
> Having the error reported via a CQE is fine for all other 
> cases, and we certainly want to avoid the overhead of the query.
> 
> If we remove the value, what is the behavior of a post 
> operation when the QP is in the reset state?  If it fails 
> silently but the post call succeeded, we have a problem.  A 
> simple flag would solve the problem. 
> Heck, even an assertion in debug builds during post would be 
> fine, so that we don't have the overhead on a release build.
> 
> - Fab
>